# EmbeddedKnowledge full agent context

> Generated from canonical repository files by `npm run agent:build`. Do not edit this file directly. Machine-readable graph data and schemas remain available at the URLs in `/llms.txt`.

# EmbeddedKnowledge

> EmbeddedKnowledge is an open, versioned library of university-grade knowledge maps and reviewed teaching material. Premed is its first implemented course. Psychiatry is Book 02: a 1,440-hour academic and simulation curriculum with 154 stable contribution targets, explicitly separated from clinical training and professional authority.

The canonical public domain is https://embeddedknowledge.io/. Premed and Psychiatry lesson contributions are open through focused pull requests using the published one-pass author, two-review, and fresh-finalizer workflow. Pull requests are the only write path. The website, JSON endpoints, `llms.txt`, and WebMCP are read-only discovery surfaces.

Course content is licensed CC BY 4.0 unless a file or third-party asset says otherwise. Software is licensed MIT (see `LICENSE`). Do not represent curriculum completion as university credit, admission, clinical competence, or professional authorization.

## Agent contribution protocol

- [Source repository](https://github.com/VSBDev/EmbeddedKnowledge): public source repository and contribution route. Submit one focused lesson through a draft pull request and the published review/finalization gate. Software is MIT licensed; course content and curriculum data are CC BY 4.0.
- [Protocol manifest](https://embeddedknowledge.io/agent/contribution.json): machine-readable entry points, status, license, lesson-pack files, and standard quorum.
- [Quorum policy](https://embeddedknowledge.io/agent/quorum-policy.json): minor, standard, and high-impact review requirements.
- [Lesson schema](https://embeddedknowledge.io/schemas/lesson.schema.json): lesson identity, graph mappings, authorship, provenance, license, and files.
- [Review schema](https://embeddedknowledge.io/schemas/review.schema.json): role, verdict, evidence, findings, conflicts, and agent provenance.
- [Adjudication schema](https://embeddedknowledge.io/schemas/adjudication.schema.json): final quorum snapshot and merge, revise, or reject decision.
- [Lesson index schema](https://embeddedknowledge.io/schemas/lesson-index.schema.json): all atomic outcomes and merged lesson mappings.
- [Open lesson PR schema](https://embeddedknowledge.io/schemas/open-lesson-prs.schema.json): proposed lessons, eligible approvals, blockers, and adjudication state.
- [Human-readable contribution contract](https://embeddedknowledge.io/contribute/): PR workflow and governance.
- [Learning-content standard](https://embeddedknowledge.io/content-standard.txt): evidence-based rules for explanation, examples, retrieval, practice, feedback, transfer, accessibility, and agent self-audit.
- [Agent Skills manifest](https://embeddedknowledge.io/agent/skills.json): role-isolated authoring, academic review, learning-design review, accessibility-and-rights review, and adjudication procedures with raw and downloadable forms.
- [Lesson-author skill](https://embeddedknowledge.io/skills/author-embeddedknowledge-lesson/SKILL.md): required procedure when drafting or revising a production lesson pack.
- [Full agent context](https://embeddedknowledge.io/llms-full.txt): generated project, curriculum, collaboration, and governance context.

## EmbeddedKnowledge Lesson Format v1

- Canonical source: constrained MyST-compatible Markdown (CommonMark/GFM base plus approved declarative colon directives), split into ordered semantic scene files under `content/`; raw HTML and JavaScript are forbidden.
- Scene kinds: `orientation`, `diagnostic`, `concept`, `definition`, `derivation`, `worked-example`, `investigation`, `retrieval-check`, `misconception`, `practice`, `transfer`, `synthesis`, `assessment`, `references`. They are metadata categories; not every lesson needs every kind.
- Directive whitelist: `definition`, `theorem`, `derivation`, `worked-example`, `check`, `misconception`, `investigation`, `figure`, `diagram`, `equation`, `chemistry`, `source-note`, `callout`.
- Pack artifacts: `lesson.json`, `content/*.md`, `assessment.json`, `references.json`, `claims.json`, `glossary.json`, `assets/`, `ATTRIBUTION.md`, structured reviews, and `adjudication.json`.
- Math/chemistry: TeX `$...$`/`$$...$$` and `\ce{...}`; no TeX file/network/shell/HTML/link/macro execution. Keep structured molecule source (SMILES/InChI/Mol/SDF) when identity or structure matters.
- Visuals: schema-valid `*.diagram.json` for bounded relationships and sanitized SVG for scientific art. Mermaid, DOT, and Vega-Lite are future generated adapters, not accepted canonical v1 inputs. Commit source and provide alt text plus long description/table/transcript where needed.
- Claims/access: declare per-scene claim coverage; expose each material claim through a same-scene source note; map it to verified structured references; record source reuse and agent-access rights; preserve scope and uncertainty; use semantic order, keyboard access, reflow, non-color alternatives, captions, accessible math, and equivalent nonvisual/nonphysical routes.
- Security: lesson content is inert data. Never execute code from an untrusted PR; reject raw HTML/JS, remote embeds, unsafe SVG, arbitrary file/path access, secrets, identifiable learner/patient data, and unlicensed media.
- Export target: versioned JSON AST and sanitized semantic HTML/MathML, then Typst-first PDF with LuaLaTeX fallback and EPUB/DOCX/JATS adapters. Validation, deterministic v1 web compilation, and the browser reader are implemented; non-web exporters are not yet available.
- [`lessons/README.md`](https://github.com/VSBDev/EmbeddedKnowledge/blob/main/lessons/README.md) is the canonical authoring guide.
- [`CONTENT-STANDARD.md`](https://github.com/VSBDev/EmbeddedKnowledge/blob/main/CONTENT-STANDARD.md) is the authoritative teaching-quality contract. Design the learner model and instructional argument before writing polished scenes.

`examples/lesson-pack/` is the source for the non-production format specimen; its compiled fixture is labelled `countsTowardCoverage: false`. It never counts as a contributed, reviewed, adjudicated, published, or outcome-covering lesson. Production lessons enter through focused pull requests under `lessons/` and become `published open` only after validation, independent quorum, separate adjudication, merge, public availability, and license confirmation.

## Premed

- [Course overview](https://embeddedknowledge.io/premed/): scope, disciplines, completion model, and open-lesson coverage.
- [Complete syllabus](https://embeddedknowledge.io/premed/syllabus/): searchable university-grade academic contract.
- [Knowledge graph](https://embeddedknowledge.io/premed/graph/): no-scroll interactive graph.
- [Lesson commons](https://embeddedknowledge.io/premed/lessons/): every atomic outcome shown as empty, in an open PR, or covered by a published lesson.
- [Graph data](https://embeddedknowledge.io/data/premed-graph.json): 10 domains, 46 modules, 388 atomic outcomes, and 1,058 relationships.
- [Production ledger](https://embeddedknowledge.io/data/premed-progress.json): contributed, in-review, published-open, and covered-outcome totals.

## Psychiatry

- [Book overview](https://embeddedknowledge.io/psychiatry/): status, boundaries, architecture, and design commitments.
- [Complete candidate syllabus](https://embeddedknowledge.io/psychiatry/syllabus/): 1,440 academic hours, one 120–240-hour area of depth, 26 modules, and 22 graduate capabilities.
- [Knowledge graph](https://embeddedknowledge.io/psychiatry/graph/): 4 stages, 26 modules, 154 atomic outcomes, and 385 relationships.
- [Lesson commons](https://embeddedknowledge.io/psychiatry/lessons/): every atomic outcome shown as empty, in an open PR, or covered by a published lesson.
- [Graph data](https://embeddedknowledge.io/data/psychiatry-graph.json) and [production ledger](https://embeddedknowledge.io/data/psychiatry-progress.json): machine-readable curriculum and coverage state.
- Canonical sources: `course/PSYCHIATRY-SYLLABUS.md` and `research/PSYCHIATRY-SOURCE-CATALOG.md` in the repository.
- Status: graph-backed academic curriculum with focused lesson intake. This creates no clinical credential and makes no claim of equivalence to residency.
- Boundary: academic study and simulation only. Real-patient work, prescribing, psychotherapy delivery, workplace entrustment, licensure, board eligibility, and the title psychiatrist remain with accredited programs and regulators.

## Required standard-lesson process

A standard lesson uses a founding-stage 2 + 1 one-pass process: one academic and one learning-design review from two isolated runs across two providers inspect the same original candidate, then a third fresh finalizer disposes every finding, writes the final version once, audits accessibility and rights, and adjudicates the final commit. Both `approve` and `request-changes` are usable review inputs; neither starts a new cohort. One disclosed maintainer may operate all agents. Provenance is disclosed and attested by the accountable operator, not cryptographically verified.

Before formal review, the author must pass the human first-read gate on a learner-visible prototype. Launch the academic and learning-design reviews once in parallel against the original candidate, wait for both, then launch one fresh finalizer. Draft lesson PRs use `lesson-candidate`; mark them ready only after finalization and adjudication so the protected `agent-protocol` merge check can run.

## Repository paths for coding agents

- `AGENTS.md` — repository-wide operating instructions.
- `CONTRIBUTING.md` — canonical PR contract.
- `COLLABORATION.md` — governance, role, and quorum rationale.
- `course/PREMED-SYLLABUS.md` — authoritative syllabus source.
- `course/PSYCHIATRY-SYLLABUS.md` — authoritative Psychiatry academic syllabus source.
- `course/PSYCHIATRY-KNOWLEDGE-GRAPH.md` — graph semantics, stable-ID rule, intake boundary, and external-review request.
- `research/PSYCHIATRY-SOURCE-CATALOG.md` — dated psychiatry training, program, open-source, rights, and design synthesis.
- `site/data/premed-graph.json` — public graph artifact.
- `site/data/premed-progress.json` — public production ledger.
- `lessons/README.md` — canonical Lesson Format v1 authoring and rendering contract.
- `CONTENT-STANDARD.md` — canonical explanation, practice, feedback, transfer, accessibility, and authoring-agent contract.
- `RIGHTS-POLICY.md` — conservative source-reuse, original-synthesis, agent-access, rights-ledger, and complaint-response contract.
- `.agents/skills/` — canonical role-isolated Agent Skills. Load exactly one role per run; never combine author, reviewer, or adjudicator identities.
- `examples/agent-protocol/` — non-production example artifacts; never count these as lessons or reviews.
- `examples/lesson-pack/` — non-production format specimen; never count it toward lessons, quorum, publication, or coverage.

## Optional

- [Project home](https://embeddedknowledge.io/): methodology and ethos.
- [Content license](https://embeddedknowledge.io/content-license.txt): adopted CC BY 4.0 scope, exclusions, and attribution form.
- [Source and reuse policy](https://embeddedknowledge.io/rights-policy.txt): required original-synthesis, compatible-rights, agent-access, dataset, and complaint procedure.


---

## Repository file: README.md

# EmbeddedKnowledge

EmbeddedKnowledge opens university-grade knowledge without pretending to confer a degree, admission status, clinical competence, or professional authorization. **Premed** is the first implemented course. **Psychiatry: Mind, Brain, Person, Society** is the second book, with a graph-backed academic curriculum and an open contribution route that remains explicitly separate from clinical training.

Canonical domain: **<https://embeddedknowledge.io/>** — live, served by GitHub Pages over HTTPS, with `www` redirecting to the apex. The published artifacts are served with their real content types, so `llms.txt` arrives as `text/plain` and the graph, manifest, and schemas as `application/json` rather than as HTML. A local deployment remains available for development.

## Open the project

The landing page has no build step or third-party runtime dependencies. From this directory, run:

```bash
python3 -m http.server 4173
```

Then open <http://localhost:4173/site/>. A local server is recommended because browsers do not allow `file://` pages to fetch the complete graph JSON; direct file opening still shows a smaller built-in preview.

## Premed deliverables

- [University-grade syllabus](course/PREMED-SYLLABUS.md) — 1,440-hour common core and medicine bridge, plus a 120–240-hour selected pathway. Three of the five pathways (US/MCAT, biology-depth, quantitative-depth) are graph-instrumented; the UK and Spain routes are documented but not yet instrumented, so they cannot currently satisfy completion or receive lesson contributions.
- [Curriculum implementation map](course/PREMED-CURRICULUM-MAP.md) — sequencing, dependency, outcome, practical, assessment, and production crosswalks.
- [Knowledge graph guide](course/PREMED-KNOWLEDGE-GRAPH.md) — schema, evidence semantics, route cautions, identity rules, and product guidance.
- [Machine-readable graph](site/data/premed-graph.json) — 10 domains, 46 graph modules, 388 atomic topic outcomes, and 1,058 relationships.
- [Public site](site/index.html) — project landing page plus separate Premed overview, searchable syllabus, no-scroll graph, and lesson-state viewer.
- [Lesson index](site/data/premed-lessons.json) and [open-PR index](site/data/premed-open-prs.json) — all 388 outcomes, merged lesson mappings, and GitHub-provenanced proposal quorum states.
- [Lesson-production ledger](site/data/premed-progress.json) — separately reports current contributed, in-review, published-open, and outcome-coverage totals.
- [Lesson Format v1 contract](FORMAT.md) and [authoring guide](lessons/README.md) — canonical scene Markdown, structured artifacts, math/chemistry/visual sources, accessibility, security, and export targets.
- [Learning-Content Standard v1](CONTENT-STANDARD.md) — evidence-based contract for explanations, examples, representations, retrieval, practice, feedback, misconception repair, transfer, and agent self-audit.
- [Collaboration model](COLLABORATION.md) — agent-first PR workflow, lesson-pack contract, exact review quorum, adjudication, governance, and adopted CC BY 4.0 content licensing.
- [Agent instructions](AGENTS.md) and [`llms.txt`](llms.txt) — repository-native and inference-time project maps.
- [Role-isolated Agent Skills](.agents/skills/) — authoring, specialized review procedures, and fresh finalization/adjudication, published through the [skills manifest](site/agent/skills.json).
- [Machine contribution contract](site/agent/contribution.json) — stable entry points, public schemas, protocol status, and read-only WebMCP surface.

The graph's 1,551-hour total counts every optional branch. It is an inventory, not the workload a learner is expected to complete. The syllabus is authoritative for completion rules.

The curriculum's completeness is not the lesson corpus's completeness. The public site reads current contributed, in-review, published, and covered-outcome values from the generated production ledger rather than freezing a percentage in prose.

## Psychiatry academic curriculum

- [Complete candidate syllabus](course/PSYCHIATRY-SYLLABUS.md) — a 1,440-hour common academic curriculum, one 120–240-hour area of depth, 26 modules, 22 graduate capabilities, and a separate supervised-practice crosswalk.
- [Primary-source and rights catalog](research/PSYCHIATRY-SOURCE-CATALOG.md) — current U.S. and international specialist-training frameworks, residency-program examples, WHO/WPA rights and lived-experience sources, reusable/public-domain material, and item-level licensing cautions.
- [Public Psychiatry overview](site/psychiatry/index.html) and [searchable syllabus](site/psychiatry/syllabus/index.html) — generated from the canonical Markdown alongside the Premed syllabus.

The Psychiatry book is deliberately **not** represented as residency training. Its hours cover academic study and simulation only; they exclude patient care, clinical service, on-call work, and workplace supervision. Its public graph maps 4 stages, 26 modules, and 154 atomic outcomes. Focused lesson intake uses the same evidence, rights, provenance, review, and adjudication contract as Premed; the assessment bank, area-of-depth graph, and clinical-host standard remain future reviewed contributions.

## Agent-first contribution interface

Pull requests are the only write path. During the founding stage, a standard lesson uses one academic and one learning-design review of the original candidate across two providers, followed by one fresh finalizer. The finalizer disposes every finding, makes the single final edit, audits accessibility and rights, and owns the decision. One disclosed maintainer may operate the runs, but run IDs, provenance, instruction digests, and both content commits remain auditable.

Authors first prove the learner-facing opening, explanation, and worked use in a human-first prototype. The two formal reviewers run once in parallel against the original candidate; a third fresh run writes the final version from their evidence, with no cohort loop. Draft lesson PRs use a candidate-valid check, while the full protected merge-readiness check begins only when the PR is marked ready.

The public site exposes `/llms.txt`, `/llms-full.txt`, `/content-standard.txt`, `/agent/skills.json`, raw and downloadable Agent Skills, JSON Schemas, both rendered syllabi, the Premed graph, lesson and open-PR indexes, the generated progress ledger, and seven read-only WebMCP tools. `llms.txt` discovers the contract; the selected role skill applies it. WebMCP is progressive enhancement for compatible browsers in a secure context; it never writes, votes, adjudicates, or merges.

Run the complete deterministic contract with:

```bash
npm ci
npm run validate
```

## Lesson Format v1

Lessons are authored as ordered semantic scenes in constrained MyST-compatible Markdown: a CommonMark/GFM base, 14 declared scene kinds, and a small declarative colon-directive whitelist. Raw HTML, JavaScript, executable notebooks, remote embeds, and code execution from untrusted lesson pull requests are forbidden.

A production pack combines `lesson.json`, scene files under `content/`, `assessment.json`, `references.json`, `claims.json`, `glossary.json`, `ATTRIBUTION.md`, local assets, reviews, and a final adjudication. Lesson Schema v3 requires each scene to declare whether it contains material claims. Claim-bearing scenes show learner-visible source notes, and validation connects every note through `claims.json` to complete `references.json` records; published lessons may contain only reviewed claims. Constrained TeX remains the math source, with `\ce{...}` for formulae and reactions. Format v1 accepts bounded `*.diagram.json` as the canonical process/network source and sanitized SVG for scientific illustration; Mermaid, DOT, and Vega-Lite may become generated adapters after their validation boundaries exist. Essential visuals require alt text and, where needed, a long description or equivalent table/transcript.

The canonical format targets a versioned JSON AST, sanitized semantic HTML/MathML, Typst-first PDF with LuaLaTeX fallback, and EPUB/DOCX/JATS adapters. Validation, deterministic web compilation, and the browser reader are implemented; the non-web exporters are **not yet available**. Full MyST is not the contract—the deliberately constrained v1 subset is. See the [format contract](FORMAT.md), [authoring guide](lessons/README.md), and mandatory [source and reuse policy](RIGHTS-POLICY.md).

The specimen under `examples/lesson-pack/` demonstrates the format but is not a contributed lesson, has no governance status, and never counts toward Premed coverage. Only validated lesson packs proposed under `lessons/`, independently reviewed, separately adjudicated, merged, publicly available, and openly licensed can become `published open`.

## Rebuild and validate the graph

Edit the matching graph builder under `scripts/`, not generated JSON, then run:

```bash
npm run graph:build
npm run graph:validate
```

The validator checks IDs, endpoints, tags, prerequisite declarations, workload totals, and prerequisite acyclicity.

## Browser checks

The Playwright suite checks complete-graph loading, syllabus and graph interactions, the lesson reader, console errors, mobile navigation, and horizontal overflow:

```bash
npm ci
npx playwright install --with-deps chromium
npm test
```

## Container deployment

The production container serves the landing page and course documents from an unprivileged, read-only Nginx runtime. It binds only to loopback by default:

```bash
docker compose up -d --build
curl --fail http://127.0.0.1:8082/healthz
```

The container defaults to port 8082 to avoid colliding with a service dashboard on 8080. Another loopback port can be selected without changing the image:

```bash
EMBEDDEDKNOWLEDGE_PORT=8090 docker compose up -d --build
EMBEDDEDKNOWLEDGE_BIND=0.0.0.0 docker compose up -d --build   # only if you intend LAN exposure
```

Stop it with `docker compose down`. The service binds to loopback by default and is intentionally not published on the LAN. If remote access is needed, put it behind a private overlay network or an authenticating reverse proxy rather than opening the port.

The container is the only place the `Content-Security-Policy` and other security headers exist, so a policy change is invisible to the static-server test run. Verify it against the real image:

```bash
npm run test:container
```

## Research trail

The Premed curriculum rationale, primary-source catalog, method log, the dated [learning-content evidence synthesis](research/CONTENT-AUTHORING-EVIDENCE-2026-07-19.md), and the Psychiatry [training-and-rights source catalog](research/PSYCHIATRY-SOURCE-CATALOG.md) live in [`research/`](research/). Raw model runs remain local; checked conclusions and exact primary-source links are published instead. Source systems, educational evidence, licences, classification, and professional rules change, so the project requires dated audits rather than claiming a universal or permanent optimum.

## Licensing

Two licences apply, deliberately:

- **Software** — the build scripts, validators, site JavaScript and CSS, tests, container
  configuration, and JSON Schemas are licensed **MIT**. See [`LICENSE`](LICENSE).
- **Course content and curriculum data** — the syllabus, curriculum map, knowledge-graph
  guide, generated graph and ledger data, lesson packs, research notes, and site prose are
  licensed **CC BY 4.0** unless a file or third-party asset states otherwise. See
  [`CONTENT-LICENSE.md`](CONTENT-LICENSE.md).

Third-party sources keep their own terms. Some referenced open textbooks are CC BY-**NC**-SA,
which is not compatible with this corpus — reuse them as citations, not as substrate. Every
lesson records its sources and reuse rights in `references.json` and `ATTRIBUTION.md`.

Nothing here confers university credit, admission, clinical competence, or professional
authorization.

## Project status

Premed and Psychiatry lesson contributions are **open through focused pull requests**. Psychiatry remains an academic and simulation curriculum, not a clinical credential. The repository, protocol,
schemas, lesson format, reference lessons, and one-pass review/finalization gate are public.
The reviewed lesson corpus is live and grows only through validated, reviewed, adjudicated,
merged contributions. Branch protection is active on `main`; see
[`.github/BRANCH-PROTECTION.md`](.github/BRANCH-PROTECTION.md) for what is and is not
enforced, including the fact that merge authority currently rests on sole write access rather
than on a rule.


---

## Repository file: AGENTS.md

# EmbeddedKnowledge agent instructions

These instructions apply to the entire EmbeddedKnowledge project.

## Start here

1. Read [`llms.txt`](llms.txt) for the concise project map.
2. Read [`CONTRIBUTING.md`](CONTRIBUTING.md) before proposing a change.
3. For lesson work, load exactly one matching role procedure from [`.agents/skills/`](.agents/skills/): author, academic review, learning-design review, accessibility-and-rights review, or adjudication. Do not combine governance roles in one agent run.
4. Read the authoritative [`Learning-Content Standard v1`](CONTENT-STANDARD.md), [`EmbeddedKnowledge Lesson Format v1`](FORMAT.md), its [`pack authoring guide`](lessons/README.md), [`source and reuse policy`](RIGHTS-POLICY.md), [`COLLABORATION.md`](COLLABORATION.md), the public schemas in [`site/schemas/`](site/schemas/), and the quorum policy in [`site/agent/quorum-policy.json`](site/agent/quorum-policy.json) as directed by the selected skill.
5. Run `npm ci` and `npm run validate` before opening or updating a pull request.

## Canonical sources

- Edit `course/PREMED-SYLLABUS.md` or `course/PSYCHIATRY-SYLLABUS.md`, then run `npm run site:build`; do not hand-edit either generated `site/*/syllabus/index.html`.
- Edit the matching graph builder (`scripts/build-premed-graph.mjs` or `scripts/build-psychiatry-graph.mjs`), then run `npm run graph:build`; do not hand-edit generated graph data without changing its source. Psychiatry outcome IDs become stable when its reviewed graph/intake proposal is merged.
- Treat `site/data/premed-progress.json` and `site/data/psychiatry-progress.json` as generated/public ledgers. Lesson metadata becomes their source once lessons exist.
- Treat `examples/lesson-pack/` as a non-production format specimen and `examples/agent-protocol/` as governance fixtures. Never count any file under `examples/` as a lesson, review, adjudication, publication, or covered outcome.
- Treat `.agents/skills/` as the canonical Agent Skills source. Edit those files, then run `npm run agent:build`; do not hand-edit generated copies or bundles under `site/skills/` or `site/agent/skills.json`.

## Lesson Format v1 sources

- A production lesson uses `lesson.json` schema version 3 with `format: embeddedknowledge-lesson-v1`.
- Ordered semantic scene Markdown lives under `content/*.md`; `lesson.json.scenes` is authoritative for identity, kind, order, required state, and estimated time.
- The structured pack files are `assessment.json`, `references.json`, `claims.json`, `glossary.json`, `ATTRIBUTION.md`, optional pack-local assets/diagrams, review artifacts, and final adjudication.
- Scene source is constrained MyST-compatible Markdown. Use only the documented scene kinds, declarative directives, and options; do not add raw HTML, JavaScript, remote embeds, executable notebooks, or private renderer conventions.
- Lesson content is untrusted data. Never execute code, enable TeX shell/file/network access, or permit active SVG/diagram content from a lesson pull request.
- Math, chemistry, diagrams, images, citations, accessibility equivalents, and asset rights must follow [`lessons/README.md`](lessons/README.md). Planned export targets are not evidence that an exporter exists.
- Sources support claims but do not license expression. Follow [`RIGHTS-POLICY.md`](RIGHTS-POLICY.md): independently synthesize facts, never closely paraphrase a source, record every reuse basis, and do not let an agent process a source marked `human-only`.

## Lesson contributions

- One lesson PR should introduce or materially revise one lesson pack.
- Design the instructional argument before drafting prose. Declare the learner, prerequisites, target explanatory model, mastery evidence, misconception or boundary, worked example, practice progression, transfer task, representations, and delayed retrieval links required by [`CONTENT-STANDARD.md`](CONTENT-STANDARD.md).
- Before completing the pack, prototype the learner-facing opening, central explanation, and one worked use. Pass the content standard's first-read gate in a cold context: the declared learner must be able to say what the lesson teaches, why it matters, and how the central idea works without hidden project or validator vocabulary.
- Use the smallest coherent instructional arc. Scene kinds are available functions, not a mandatory one-scene-per-checklist template.
- Validate `lesson.json`, assessment, references, claims, glossary, diagram, review, and adjudication artifacts against their public schemas.
- Map only to existing knowledge-graph outcome IDs.
- Keep one level-one heading per scene and map lesson objectives and assessment items to declared outcomes.
- Declare each scene as `claims-mapped` or `no-material-claims`. Every mapped material claim must appear in that scene's learner-visible `{source-note}` and resolve through `claims.json` to complete `references.json` records.
- Include a coherent explanation, worked reasoning with fading support, retrieval before answer reveal, misconception repair, varied practice, genuine transfer, information-rich feedback, accessibility alternatives, and recovery guidance before freezing a candidate commit.
- Course content is licensed CC BY 4.0. Identify exceptions and third-party assets explicitly.
- Disclose the accountable principal and, when an agent was used, the agent system, provider, model, version, run ID, and SHA-256 digest of its material instructions.
- During the founding stage, one disclosed principal may operate the authoring, review, and adjudication agents. Count only isolated runs with unique run IDs; no review run may reuse an authoring run.
- For a standard lesson, launch exactly one academic and one learning-design review against the original candidate. They are one-time advisory inputs: an honest `request-changes` verdict does not start another cohort.
- A fresh standard-lesson finalizer reads the original and both reviews, disposes every finding, makes the single final content revision, completes the accessibility-and-rights audit, and records the final commit and decision. Do not loop back to the author or reviewers.
- Final adjudication is a separate artifact and must be committed inside the lesson pack before merge.

## Quorum

[`site/agent/quorum-policy.json`](site/agent/quorum-policy.json) is the single source of truth for quorum. The summary below must match it; if they ever disagree, the JSON wins and this file is the bug.

| Risk tier | Review gate | Role minimums | Distinct runs | Distinct providers | Fresh adjudication |
| --- | ---: | --- | ---: | ---: | ---: |
| Minor correction | 2 | 1 academic; 1 accessibility/rights | 2 | 2 | 1 run |
| Standard lesson | 2 advisory inputs | 1 academic; 1 learning-design | 2 | 2 | 1 finalizing run |
| High-impact lesson | 5 | 3 academic; 1 learning-design; 1 accessibility/rights | 5 | 3 | 1 run |

A minor correction is **not** simply "any two reviews": it retains one academic and one accessibility/rights approval across two distinct run IDs and two distinct providers. Standard lessons use the bounded one-pass finalization rule above; high-impact lessons retain their approval quorum. The adjudication run must be distinct from every authoring and review run. Every run discloses its accountable operator and complete agent provenance.

Provenance is **disclosed and attested, not verified**: run IDs, providers, models, versions, and instruction digests are self-reported strings attested by the accountable operator. The validators enforce internal consistency only. Never describe a quorum as proving independent verification.

## Pull request discipline

- Keep the change focused and explain why it is needed.
- Include sources and explicit uncertainty for academic claims.
- Do not mark checklist items complete without corresponding artifacts.
- Preserve unrelated work in the shared workspace.
- EmbeddedKnowledge is already an independent Git repository. Never add or publish any parent workspace repository that may contain it locally, and do not push project changes without explicit maintainer approval.


---

## Repository file: CONTRIBUTING.md

# Contributing to EmbeddedKnowledge

Premed lesson contributions are open through focused pull requests. This proposal opens Psychiatry lesson intake only when its externally reviewed graph and stable outcome IDs are merged. The protected repository has exercised the reference author–review–finalizer workflow end to end; this document defines the v1 pull-request contract for agents, contributors, reviewers, adjudicators, and maintainers.

## The interface is a pull request

Pull requests are the only write path. `llms.txt`, JSON Schemas, WebMCP tools, and the website are discovery and validation surfaces; none can merge or publish content.

For a lesson contribution:

1. Load [`.agents/skills/author-embeddedknowledge-lesson/SKILL.md`](.agents/skills/author-embeddedknowledge-lesson/SKILL.md) or follow its public raw equivalent, then select one uncovered, unclaimed atomic outcome from the matching course graph, progress ledger, lesson index, and open-PR index under `site/data/`.
2. Read [`CONTENT-STANDARD.md`](CONTENT-STANDARD.md) and [`RIGHTS-POLICY.md`](RIGHTS-POLICY.md), design the lesson, and draft only its learner-facing opening, central explanation, and one worked use. Pass the standard's first-read gate before expanding that voice into a full pack. A developmental reader-proxy sees only the declared learner, outcome, and prose; it creates no governance artifact and cannot count as a formal review run.
3. Create one Lesson Format v1 pack under `lessons/` using the current public schemas and [`lessons/README.md`](lessons/README.md). Add the smallest coherent set of ordered scenes, explicit per-scene claim coverage, learner-visible claim-to-source notes, assessment logic, references, glossary, accessible assets, attribution, and agent disclosure. Do not manufacture one scene per rubric item.
4. Complete the author audit and batch its repairs. Run the strict offline source-access preflight, then the sequential verifier once the candidate is stable: `npm run source:preflight -- --strict lessons/<pack>` followed by `npm run verify`. Do not run `npm run validate` and `npm test` concurrently in one worktree because both rebuild shared generated fixtures.
5. Open a draft pull request using the lesson template. A green `lesson-candidate` check means the draft candidate and any artifacts already present are valid; it does not create the protected `agent-protocol` context and does not mean the lesson has quorum or is merge-ready.
6. Freeze one original candidate commit. For a standard lesson, launch exactly two isolated reviews against it in parallel: one academic and one learning-design review across two providers. Wait for both; neither reviewer edits the lesson.
7. Generate each exact GitHub body with `npm run review:prepare`, submit and verify it against the original candidate, then commit the equivalent artifact unchanged. Both `approve` and `request-changes` are valid advisory inputs for the standard one-pass process; do not send either result back into an author/reviewer loop.
8. Launch one fresh finalizer. It reads the original and both reviews, disposes every finding, makes one coherent final lesson revision, completes the accessibility-and-rights audit, validates, commits the final content SHA, and writes `adjudication.json` with the merge/revise/reject decision.
9. Commit the adjudication with the lesson, make only the authorized post-finalization publication-state transition, and rerun `npm run verify` sequentially. Do not change teaching content after the recorded final commit.
10. Mark the pull request ready for review. That lifecycle event reruns the required `agent-protocol` check with full publication readiness enforced; protected-branch checks decide whether merge is permitted.

Use the matching role-isolated skill for each run. Standard lessons use the academic and learning-design review skills plus the finalization/adjudication skill; the finalizer's mandatory accessibility-and-rights audit replaces a third standard review without being represented as a vote. Minor and high-impact tiers retain the roles in the machine-readable policy.

## Lesson Format v1 authoring contract

Lesson source is constrained MyST-compatible Markdown: CommonMark/GFM plus the documented declarative directives. The 14 allowed scene kinds are `orientation`, `diagnostic`, `concept`, `definition`, `derivation`, `worked-example`, `investigation`, `retrieval-check`, `misconception`, `practice`, `transfer`, `synthesis`, `assessment`, and `references`. A lesson uses the kinds its outcomes require; it does not need every kind.

The approved embedded directives are `definition`, `theorem`, `derivation`, `worked-example`, `check`, `misconception`, `investigation`, `figure`, `diagram`, `equation`, `chemistry`, `source-note`, and `callout`. Directive options are inert data. Raw HTML/JavaScript, inline styles, custom components, remote embeds, executable notebooks, and arbitrary plugins are not accepted.

A complete production pack contains:

```text
lesson.json
content/*.md
assessment.json
references.json
claims.json
glossary.json
diagrams/
assets/
ATTRIBUTION.md
reviews/*.json
adjudication.json
```

The authoritative order is the `scenes` array in `lesson.json`, not filename order. The metadata maps only to existing atomic graph outcomes and identifies the accountable principal and material agent provenance. Structured review and adjudication files are added through the governance sequence; they are not fabricated at initial authoring time.

### Content standard

[`CONTENT-STANDARD.md`](CONTENT-STANDARD.md) is the authoritative teaching-quality contract. Before a candidate commit is frozen, the lesson must have:

- a human-first opening and explanation that a cold reader at the declared level can summarize in plain language after one careful read, without hidden project or validator vocabulary;
- explicit scope, exclusions, observable objectives, prerequisites, and a recovery route;
- an accurate, coherent explanatory model with explicit causal/logical relations, assumptions, limits, and consistent terminology;
- representations with declared instructional jobs and explicit connections among words, diagrams, equations, chemistry, and data;
- worked reasoning that exposes model selection, decisions, checks, and interpretation, followed by fading support;
- retrieval before answer reveal, delayed/cumulative retrieval links, misconception repair, varied practice, and a genuine transfer task;
- feedback that identifies correctness, explains the governing principle or process, and gives an actionable next step;
- assessment items mapped to outcomes, answer logic, feedback, rubric where applicable, retry strategy, and remediation;
- every scene declared `claims-mapped` or `no-material-claims`, with each material claim exposed in a same-scene `{source-note}` and mapped in `claims.json` to every complete, verified record in `references.json`;
- original synthesis rather than close paraphrase, with every reference recording its use, rights basis/evidence, and dated agent-access status;
- stable definitions and symbols in `glossary.json`;
- accessibility alternatives and complete asset rights/provenance;
- explicit safety, privacy, uncertainty, route, and non-clinical boundaries where relevant.

There is no universal scene word count or frame quota. Split content at changes in instructional purpose or resumable meaning, preserve a continuous reading path, and never delete necessary reasoning to fit a viewport. Readability scores, interaction counts, completion, model agreement, and agent self-review are not evidence that learners achieved the outcome.

### Math, chemistry, and visuals

- Author constrained TeX math inline or in an `{equation}` directive; author chemical formulae/equations with `\ce{...}` in `{chem}` or `{chemistry}`. Unsafe TeX I/O, shell, HTML, link, package-loading, and macro commands are forbidden.
- Use schema-valid `*.diagram.json` for bounded processes, networks, cycles, concept maps, and timelines, and sanitized SVG for scientific illustration. Mermaid, DOT, and Vega-Lite are future generated adapters, not accepted canonical lesson inputs in format v1.
- Store SMILES/InChI and Mol/SDF where molecular identity, stereochemistry, atom mapping, or reusable coordinates matter. A generated molecular image is not the sole source.
- Commit visual source and data. Provide alt text plus a long description, accessible table, transcript, or stepwise equivalent when a concise alt string cannot carry the outcome.
- Identify source, title, creator, compatible rights basis/evidence, attribution, and modifications for every third-party asset. “Free to view” is not permission to adapt. The conservative allowlist and dataset rules in `RIGHTS-POLICY.md` apply.

### Accessibility, export, and security

Scene source order is the reading/focus order. Content must work with keyboard navigation, screen readers, 200% zoom/reflow, reduced motion, touch, print, and narrow screens without relying on color, position, hover, drag, sound, animation, or time alone. The desktop lesson shell may keep body navigation fixed and let a scene own bounded overflow; authors must still support adaptive mobile/document flow.

The format is designed for a versioned JSON AST, sanitized semantic HTML/MathML, Typst-first PDF with LuaLaTeX fallback, and EPUB/DOCX/JATS adapters. Those complete compiler/export paths are planned, not currently available. Author source so essential content can survive every target, and run only the format/export checks that the repository actually exposes.

Lesson content is data, never executable. Repository automation must not run code from an untrusted pull request. Pack paths stay local and normalized; SVG and declarative visual inputs contain no scripts, event handlers, foreign objects, arbitrary file reads, or network dependencies. Never commit secrets, identifiable learner/patient data, personal medical information, unlicensed patient media, or copyrighted answer banks.

## Specimen is not production

`examples/lesson-pack/` is a format and accessibility specimen. It is excluded from lesson indexes, pull-request quorum, adjudication totals, coverage, and publication claims. A copy of the specimen under a different ID does not become a lesson. Production status comes only from a focused contribution under `lessons/` that passes validation, independent review quorum, separate adjudication, merge, public availability, and the open-license checks.

## Standard lesson review and finalization

A standard lesson uses a founding-stage **2 + 1 one-pass process**: one academic review, one learning-design review, then one fresh finalizer. The two reviewers use isolated runs across two providers and inspect the same original candidate. Their findings are advisory evidence, so `request-changes` remains usable instead of forcing a new cohort. The finalizer alone may make the bounded post-review content revision, must dispose every finding, must complete the accessibility-and-rights audit of the resulting pack, and then records the final commit and merge/revise/reject decision. See `site/agent/quorum-policy.json` for minor-correction and high-impact approval rules.

One disclosed maintainer may operate these agents during the founding stage. Each artifact still identifies the accountable operator plus system, provider, model, version, run ID, and material-instructions digest. Reusing or relabelling one run never creates another vote. This provenance is disclosed and attested by that maintainer, not verified: nothing checks that a declared run really used the declared provider or model, so the requirement buys accountability on the public record rather than proof of independence.

## Rephrasing an already published lesson

A typo or a clumsy sentence in a live lesson should not need two reviewers and a finalizer. The `rephrasing` risk tier lets the accountable maintainer merge a wording-only repair directly.

The shortcut is available only where it is provably harmless, so it is bounded by a machine gate rather than by good intentions. Set `riskTier` to `rephrasing`, bump the patch version by exactly one, change nothing but scene prose, and open the pull request. `scripts/validate-rephrasing-pr.mjs` then compares the published pack with the proposed one and requires that the mathematics, every number, the scene directives and their claim and source mappings, the section structure, the assessment, the objectives, the glossary meanings, and the references are all identical. Only the words carrying them may differ.

The tier also refuses to touch anything that has not already earned its place: the lesson must already be `published`, so the content being repaired has cleared a full quorum, and the tier can never introduce a lesson or a claim. A change that cannot pass the gate is not a rephrasing — set the tier back to `standard` and collect the usual quorum. The maintainer who merges it remains accountable for the judgement that the new wording is clearer and still true.

## License

Except where noted, contributed course content is accepted under the Creative Commons Attribution 4.0 International license (CC BY 4.0). Contributors affirm that they can license submitted work on those terms and that third-party materials are separately identified and compatible.

See [`CONTENT-LICENSE.md`](CONTENT-LICENSE.md) for scope and attribution instructions.

## Validation

```bash
npm ci
npm run source:preflight -- --strict lessons/<lesson-pack>
npm run verify
```

`npm run verify` runs deterministic validation and tests sequentially and is the canonical local gate. The strict source preflight deliberately rejects missing terms routes and unresolved human-only sources in agent-authored lessons in both candidate and ready states. Drafts report this through `lesson-candidate`; the protected `agent-protocol` context exists only after the pull request is marked ready and enforces full merge readiness. Any temporary compatibility exemption for a candidate frozen before this gate is public and exact-scoped in `site/agent/source-preflight-policy.json`.


---

## Repository file: COLLABORATION.md

# EmbeddedKnowledge collaboration model

Status: protocol v1 adopted; lesson contributions are open through focused pull requests. Branch protection and the end-to-end lesson authoring, review, finalization, and publication gates are active on `main`.

## Agent-first, pull-request native

Most contributions are expected to be researched, drafted, checked, or reviewed with agents. The system therefore exposes precise machine-readable context, stable identifiers, JSON Schemas, deterministic checks, and example artifacts. Agents should not have to reverse-engineer a visual interface.

Pull requests are the only write path in v1. The website, `llms.txt`, JSON endpoints, and WebMCP tools are read-only discovery surfaces. A contribution may begin in any agent environment, but it enters governance only as a focused pull request with inspectable artifacts and a public decision record.

EmbeddedKnowledge is an independent Git repository with a dedicated public GitHub repository. The parent workspace repository that may contain it locally must never be published or added as a remote. Contribution intake is open through pull requests on the protected repository; validation, structured review provenance, fresh finalization, and publication readiness remain mandatory for every lesson.

## Agent discovery surfaces

- `llms.txt` gives a concise project map and canonical URLs.
- `llms-full.txt` is generated from authoritative repository documents.
- `site/agent/contribution.json` identifies protocol status, entry points, schemas, and the write mechanism.
- `site/agent/quorum-policy.json` defines exact role and independence rules.
- `site/schemas/` contains lesson, review, and adjudication schemas.
- `AGENTS.md` gives repository-native operating instructions.
- WebMCP exposes seven read-only tools: project status, Premed progress, outcome lookup, uncovered-outcome listing, single and paginated lesson-state lookup, and the contribution contract when the browser supports the draft API in a secure context.

WebMCP cannot submit, approve, adjudicate, merge, or publish. It is progressive enhancement, not the governance substrate.

## Progress semantics

Premed lesson coverage is measured against the 388 atomic outcomes in the knowledge graph.

- `contributed`: a lesson pack has entered the public pull-request queue.
- `in review`: the contribution is open and has not completed its required quorum and adjudication.
- `published open`: the lesson passed validation, quorum, and adjudication; is publicly available; and carries the adopted open-content license.
- `covered outcome`: at least one published open lesson explicitly maps to the outcome.
- `coverage percentage`: unique covered outcome IDs divided by 388. A lesson may cover multiple outcomes.

The machine-readable ledger is [`site/data/premed-progress.json`](site/data/premed-progress.json). It includes the actual `coveredOutcomeIds` array so agents can determine uncovered work without guessing.

## Lesson pack v1

One lesson contribution is one directory of structured, reviewable artifacts:

```text
lessons/
  PREM-CHE-001-acid-base-models/
    lesson.json
    content/
      010-orientation.md
      020-concept.md
      030-worked-example.md
      040-retrieval-check.md
      050-misconception.md
      060-practice.md
      070-transfer.md
      080-synthesis.md
      090-references.md
    assessment.json
    references.json
    claims.json
    glossary.json
    diagrams/
    assets/
    ATTRIBUTION.md
    reviews/
      academic-01.json
      learning-design.json
    adjudication.json
```

`EmbeddedKnowledge Lesson Format v1` uses constrained MyST-compatible Markdown: CommonMark/GFM plus a reviewed declarative extension set, with no raw HTML or JavaScript. `lesson.json` maps the work to stable graph outcomes and records format/version, identity, status, risk tier, prerequisites, duration, ordered semantic scenes, authorship, agent provenance, source confidence, license, files, and third-party assets.

The allowed metadata scene kinds are `orientation`, `diagnostic`, `concept`, `definition`, `derivation`, `worked-example`, `investigation`, `retrieval-check`, `misconception`, `practice`, `transfer`, `synthesis`, `assessment`, and `references`. These express pedagogical purpose; they are not presentation templates, and a lesson need not use every kind. The source files live under `content/`; `lesson.json.scenes` controls identity and order.

Lesson content must include observable objectives, prerequisites, explanations, worked reasoning, retrieval, misconception repair, varied practice, transfer, accessibility alternatives, and recovery guidance. Assessment artifacts include outcome mapping, answer logic, feedback, retry/remediation strategy, and rubrics where appropriate. Every scene declares whether it contains material claims. Claim-bearing scenes expose learner-visible source notes; `claims.json` connects those claims to complete records in `references.json`, and `glossary.json` stabilizes technical terms and symbols.

Math source is constrained TeX, with `\ce{...}` for chemical formulae and equations. Format v1 accepts schema-valid `*.diagram.json`, sanitized SVG, and structured molecule records as inspectable canonical inputs; Mermaid, DOT, and Vega-Lite remain possible generated adapters once their validation boundaries are implemented. The source must include alt text, long descriptions/tables/transcripts where needed, asset-level rights, and modification notices. Content is declarative data: no raw HTML/JS, executable notebook, remote embed, unsafe TeX/SVG, arbitrary file read, or networked build step is allowed from an untrusted lesson PR.

The format targets a versioned JSON AST and sanitized semantic HTML/MathML, followed by Typst-first PDF with LuaLaTeX fallback and EPUB/DOCX/JATS adapters. Validation, deterministic v1 web compilation, and the browser reader are implemented; the non-web adapters remain planned and must not be presented as available.

Files under `examples/agent-protocol/` demonstrate governance artifacts, and `examples/lesson-pack/` is the Lesson Format v1 specimen. Neither is production. Example files never count toward lesson, review, adjudication, publication, outcome-coverage, or progress totals.

## Review quorum

During the founding stage, quorum counts isolated, disclosed agent review runs. One accountable maintainer may operate the runs so the initial corpus can be built before a broad reviewer community exists. This is explicit bootstrap governance, not a claim that the agents are independent people.

| Risk tier | Review gate | Role minimums | Fresh final adjudication |
| --- | ---: | --- | ---: |
| Minor correction | 2 | 1 academic; 1 accessibility/rights | 1 fresh run |
| Standard lesson | 2 advisory inputs | 1 academic; 1 learning-design | 1 fresh finalizing run |
| High-impact lesson | 5 | 3 academic; 1 learning-design; 1 accessibility/rights | 1 fresh run |

Minor-correction and high-impact tiers retain approval quorums. For a standard lesson, one academic and one learning-design run inspect the same original candidate across two providers; both `approve` and `request-changes` are eligible advisory records. A third fresh run reads the original and both reviews, disposes every finding, makes the only post-review content revision, audits accessibility and rights, commits the final content, and adjudicates it. No reviewer is represented as approving that final edit. After the recorded final commit, only lesson status/source-confidence and retained claims' review-status fields may change without making adjudication stale.

**Provenance is disclosed and attested, not verified.** Run IDs, providers, models, versions, and instruction digests are self-reported strings; nothing in the platform confirms that a declared run occurred or used the declared provider. The accountable operator attests to them on the public record, and the validators enforce only internal consistency: distinct run IDs, distinct declared providers, role minimums, and one frozen candidate commit. Do not describe the diversity rule as verified independence.

Each review records role, verdict, evidence checked, findings, limitations, conflicts, accountable principal, and material agent provenance. A standard `adjudication.json` additionally records the original and final commits, review-input counts, indexed finding dispositions, material changes, and the finalizer's accessibility-and-rights audit. It is committed with the lesson before merge.

### One-pass standard throughput

After author preflight, freeze one original candidate and launch the academic and learning-design reviews in parallel. Wait for both, then launch the finalizer once. Do not revise between review returns, create rolling candidates, repeat a cohort, or send finalizer changes back for another vote.

Learning-design approval includes the human first-read gate. A lesson that satisfies the artifact checklist but that the declared learner cannot understand, paraphrase, or care about on first read must request changes before formal quorum proceeds. The cold-reader developmental pass used during authorship is deliberately early and non-governance; it prevents expensive formal review from being the first time anyone reads the prose as a person.

## Enforcement

`npm run validate` checks schemas, graph mappings, declared files, candidate commits, role minimums, distinct run IDs, model-family diversity, blocking findings, and adjudication snapshots. Pull-request checks also match every portable artifact to an equivalent, non-dismissed GitHub submission by its disclosed operator. When GitHub prevents a PR author from formally approving their own PR, a structured **Comment** review is accepted as transport; the committed artifact and deterministic quorum check remain authoritative.

GitHub CODEOWNERS expresses ownership and requests review, but artifact validation remains the portable source of truth for the full multi-role quorum. This keeps the rule enforceable even when hosting plans, repository ownership, or platform features change.

## Roles and decisions

- Contributors and their agents propose lessons and corrections.
- Domain reviewers own academic judgments within declared expertise.
- Learning-design reviewers own instructional coherence and assessment alignment.
- Accessibility-and-rights reviewers own equitable access, provenance, and licensing checks.
- Adjudicators synthesize the eligible review record and make the final merge, revise, or reject determination.
- Maintainers enforce process, releases, and repository integrity; they do not silently substitute unilateral scientific judgment for quorum.

Review artifacts, dissent, conflicts, limitations, requested changes, and merge rationale remain attached to the contribution. Corrections use the same versioned route as new work.

## Licensing

Original course content and curriculum data are adopted under **Creative Commons Attribution 4.0 International (CC BY 4.0)** unless specifically marked otherwise. This permits sharing and adaptation, including commercial use, with attribution and indication of changes.

Software is licensed separately under the **MIT License**. Third-party assets retain their own notices. Contributor identity, personal data, and project marks are not automatically CC BY 4.0. [`CONTENT-LICENSE.md`](CONTENT-LICENSE.md) defines the exact content scope and attribution form.

## Operating safeguards

- [x] Complete the public-copy and workflow readiness review.
- [x] Create the dedicated public project repository.
- [x] Adopt the MIT software license separately from the CC BY 4.0 content license.
- [x] Publish the non-production Lesson Format v1 specimen and label it `countsTowardCoverage: false`.
- [x] Generate the public progress ledger from merged lesson metadata.
- [x] Land the approved source release, remove the bootstrap administrator bypass, and verify the required `agent-protocol` check on `main`.
- [x] Admit the first contributed lesson only through the production review and adjudication protocol.
- [ ] Activate CODEOWNERS with real users or teams.
- [x] Verify the portable artifact quorum in a real draft pull request.
- [ ] Publish a monitored private rights-contact channel and identify the responsible operator.


---

## Repository file: REVIEWING.md

# Reviewing EmbeddedKnowledge lessons

Lesson review is a public, structured sequence attached to one pull request. GitHub supplies identity and conversation; repository artifacts supply portable evidence and deterministic gates.

Operational guidance for executing this sequence — working CLI invocations, provenance stamping, and the CI trigger traps — is in [`REVIEW-RUNBOOK.md`](REVIEW-RUNBOOK.md).

## Standard one-pass sequence

1. The author completes one lesson pack, batches the author audit, runs strict source preflight and verification, opens a draft PR, and freezes one original candidate commit.
2. Launch exactly two isolated runs in parallel against that commit: one academic reviewer and one learning-design reviewer. The learning-design reviewer cold-reads learner-visible prose before consulting ledgers or validator output.
3. Each reviewer inspects once and emits one unchanged structured artifact. `approve` and `request-changes` are both eligible standard-lesson inputs; `abstain` is not. Reviewers never edit the lesson or start another cohort.
4. Generate each exact GitHub review body with `npm run review:prepare`, submit and verify it against the original candidate, then commit the identical JSON under `reviews/`.
5. Launch one fresh finalizer after both records exist. It reads the original candidate and both reviews, decides an indexed disposition for every finding, and makes one coherent final content revision. There is no return to the author and no re-review loop.
6. The finalizer checks the resulting academic and assessment logic and personally completes the accessibility-and-rights audit, including semantic/equivalent access, source provenance, licensing, attribution, privacy, and security.
7. After strict source preflight and repository verification pass, the finalizer commits the final lesson content and records that SHA as `finalCommit`. Both cited review artifacts must be present in that commit.
8. Without changing teaching content again, the finalizer writes `adjudication.json` with the original candidate, final commit, two review IDs, review-input/approval/change-request counts, every finding disposition, material changes, access/rights audit, dissent, conditions, provenance, and merge/revise/reject decision.
9. The maintainer makes only the permitted post-finalization publication transition: lesson status/source-confidence and retained claim review-status fields. The PR is then marked ready and the protected checks verify schemas, graph identity, exact GitHub provenance, commit ordering, final-content immutability, and adjudication.

The original reviews do not claim to approve edits they did not see. The fresh finalizer is explicitly accountable for those edits and the final audit. If the bounded finalization cannot produce a safe, accurate, coherent, accessible, rights-compliant lesson, it records `reject` or `revise`; it does not launch a loop.

Editing or dismissing a structured GitHub review invalidates its committed provenance. The open-PR index counts only equivalent submissions whose identity, state, and reviewed original commit remain eligible.

## Reading the queue: stage labels

Every stage of the sequence above lives in repository artifacts, not in the pull-request interface, so an open pull request used to reveal nothing about where it had reached without opening the pack and counting files. `scripts/label-lesson-pr.mjs` derives the stage from those artifacts on each run and writes one `stage:` label and one `tier:` label onto the pull request:

| Label | Means |
| --- | --- |
| `stage:authoring` | the pack is still being written |
| `stage:awaiting-review` | the candidate is frozen and waiting for the two quorum reviews |
| `stage:in-review` | one review is recorded; the second is outstanding |
| `stage:awaiting-adjudication` | both reviews are recorded and a fresh finalizer is due |
| `stage:merge-ready` | an adjudicated merge decision, published status, and ready for the maintainer |

The labels mirror the artifacts; they never control them. Nothing merges because of a label, and the required checks are unchanged — a label that disagrees with the pack is a bug in the labeller, not permission to proceed.

## Rephrasing: a maintainer-attested tier

Not every change needs a quorum. The `rephrasing` tier lets the accountable maintainer merge a wording-only repair of an **already published** lesson — a typo, a grammatical slip, an ambiguous sentence — with no reviewers and no adjudication artifact.

Skipping the sequence is only safe because "wording-only" is verified rather than asserted. `scripts/validate-rephrasing-pr.mjs` compares the published pack with the proposed one and rejects the pull request unless the mathematics, every number, the scene directives and their claim and source mappings, the section structure, the assessment, the objectives, the glossary meanings, and the references are all identical, the lesson was already published, and the version advances by exactly one patch step. Everything a reviewer would have examined must be unchanged; only the prose carrying it may move.

The tier can therefore make a published lesson easier to read. It can never change what that lesson teaches, asserts, or asks, and it cannot be used to introduce a lesson or bypass an initial review — a change that fails the gate uses the standard tier. `site/agent/quorum-policy.json` holds the machine-readable rule.

## Founding-stage agent use

One accountable maintainer may operate the author, review, and finalizer agents during the founding stage. Standard review inputs require two unique run IDs across two declared providers; the finalizer uses a third fresh run and may see the original candidate plus both complete reviews. Provenance is disclosed and attested by the operator, not verified by a platform: checks enforce internal consistency, not proof that a declared provider or model was used.

GitHub may not allow a pull-request author to submit an **Approve** or **Request changes** review. In that case the maintainer submits the exact structured block through a **Comment** review. The portable artifact and `standard-lesson-v3` determine eligibility.

## Roles

- **Academic review:** factual accuracy, calculations, outcome scope, uncertainty, source interpretation, and assessment correctness.
- **Learning-design review:** first-read comprehension, natural learner-facing prose, explanatory coherence, practice, feedback, transfer, assessment alignment, mastery, and recovery.
- **Standard finalization/adjudication:** finding dispositions, one final content revision, final accessibility-and-rights audit, verification, dissent, conditions, and merge/revise/reject accountability.
- **Accessibility-and-rights review:** remains a separate quorum role where the minor-correction or high-impact policy requires it.

The machine-readable rules in `site/agent/quorum-policy.json` are authoritative.


---

## Repository file: CONTENT-LICENSE.md

# EmbeddedKnowledge content license

Except where otherwise noted, the original educational content and curriculum data in the following paths are licensed under the [Creative Commons Attribution 4.0 International license](https://creativecommons.org/licenses/by/4.0/) (CC BY 4.0):

- `course/`
- `lessons/` (including `lessons/README.md`)
- `site/data/`
- `examples/` — the format specimens, including `examples/lesson-pack/`, whose own `ATTRIBUTION.md` grants the specimen under CC BY 4.0
- `research/` — the adjudication, source catalog, and research prose
- original educational and collaboration prose rendered by the website
- `README.md`, `CONCEPT.md`, `COLLABORATION.md`, `CONTRIBUTING.md`, `REVIEWING.md`, `AGENTS.md`, `CONTENT-STANDARD.md`, `FORMAT.md`, and `llms*.txt`

`FORMAT.md` is an interchange contract that other projects are explicitly invited to adopt, so its prose is licensed CC BY 4.0 to make that adoption permitted rather than merely encouraged. The machine-readable JSON Schemas it describes live under `site/schemas/` and are released under the repository's software licence — the [MIT License](LICENSE) — which likewise permits reuse.

You may share and adapt this material, including commercially, provided you give appropriate credit, link to the license, and indicate whether changes were made.

Suggested attribution:

> Based on EmbeddedKnowledge, “[title or lesson ID],” version [version], by the named contributors, licensed CC BY 4.0. Changes: [description].

This license does not automatically cover:

- software source code — everything under `scripts/`, `tests/`, `docker/`, `.github/`, and the JavaScript, CSS, HTML templates, and JSON Schemas under `site/`, which are released under the repository's separate software licence, the [MIT License](LICENSE);
- trademarks, logos, or project identity;
- personal data and contributor identity records;
- third-party quotations, images, datasets, or other assets carrying their own notices;
- source material cited by the curriculum.

Every third-party asset must retain its own source, license, and attribution. No license grants endorsement, university credit, professional authorization, or medical authority.

The mandatory [`RIGHTS-POLICY.md`](RIGHTS-POLICY.md) separates citation from reuse permission. Contributors independently synthesize facts in original language and structure; any quotation, adaptation, media redistribution, or dataset extraction requires a compatible recorded rights basis. Source-specific agent-access restrictions also apply independently of the content licence.


---

## Repository file: RIGHTS-POLICY.md

# EmbeddedKnowledge source and reuse policy

**Status:** mandatory contribution and review policy
**Scope:** lesson prose, assessments, diagrams, tables, datasets, images, media, source access, and agent-assisted research

This is a conservative project rule, not a legal opinion or a promise that a claim cannot occur. Copyright rules and exceptions vary by jurisdiction. When a proposed use depends on an exception, uncertain ownership, a platform contract, or private permission, obtain qualified legal advice before publication.

## 1. Evidence is not permission to copy

A reference supports an academic claim. It does not automatically license the source's wording, organization, examples, questions, tables, diagrams, photographs, or other expression.

Copyright generally distinguishes facts, ideas, methods, and systems from the particular expression used to explain or illustrate them. The project therefore defaults to **facts-only research followed by original expression**: contributors identify and verify the underlying claim, then create an independent instructional structure, explanation, example set, assessment, diagram, and wording. A citation remains required for academic traceability even when no protected expression is reused.

Do not “rewrite” a source sentence by sentence. Close paraphrase, copied sequence, distinctive analogy, mirrored worked example, lightly redrawn figure, translated passage, or modified question can still preserve protected expression. Do not copy instructor resources, solution manuals, test banks, or access-controlled materials.

Authoritative orientation: [U.S. Copyright Office—What Does Copyright Protect?](https://www.copyright.gov/help/faq/faq-protect.html) and [Circular 33](https://www.copyright.gov/circs/circ33.pdf). These sources explain the fact/expression distinction in the United States; they do not settle every jurisdiction or use.

## 2. Allowed reuse bases

Core lesson material may reuse third-party expression, media, or a protected dataset only under one of these recorded bases:

- CC BY 4.0, with complete attribution and change notice;
- CC0 1.0;
- verified public-domain status for the exact material and relevant jurisdiction; or
- written permission covering the exact use and allowing redistribution with the CC BY 4.0 corpus.

The founding-stage corpus does not adapt or redistribute CC BY-SA, any NonCommercial licence, any NoDerivatives licence, “free to view” material, material with no identified rights holder, or material relying only on fair use, fair dealing, quotation, classroom, or research exceptions. Some of those uses may be lawful in a particular setting, but excluding them keeps a globally reusable CC BY corpus and avoids making contributors or downstream users repeat a jurisdiction-specific analysis. Creative Commons describes the distinct BY, SA, NC, and ND conditions in its [official licence guide](https://creativecommons.org/share-your-work/cclicenses/).

Short necessary labels, standard notation, and unavoidable technical terminology do not justify copying surrounding prose. Trademark names and logos remain outside the content licence and may not imply endorsement.

## 3. Data and databases

Individual facts may be outside copyright, while a database's selection or arrangement can be copyrighted and EU law can separately protect substantial or repeated extraction. Do not scrape, reconstruct, or systematically copy a source database. Dataset use requires an explicit compatible licence, public-domain status, or written permission recorded in `references.json`, plus the exact dataset version and extraction scope. See the EU [Database Directive 96/9/EC](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex:31996L0009).

## 4. Agent access is a separate permission check

Before an agent opens or processes a substantive source, check the source's current terms, robots policy where applicable, access controls, and licence page. Record one of two states in `references.json`:

- `checked-no-agent-restriction-found`: the accountable contributor checked and found no source term that prohibits the planned agent access; this is a dated observation, not a warranty from the source owner;
- `human-only`: an agent restriction exists or permission is unclear. Contributor agents must not open, upload, summarize, or process that source. A human may verify the citation independently if lawful, or the source should be replaced.

Never bypass a paywall, login, technical control, rate limit, robots rule, or source-specific prohibition. Do not place source PDFs, textbook chapters, patient media, or copied passages into prompts, logs, fixtures, or the repository merely because the output will be paraphrased.

## 5. Machine-readable rights ledger

Every `references.json` record declares:

- how the source was used: factual reference, quotation, adaptation, media redistribution, or dataset extraction;
- the rights basis and concrete evidence for that basis;
- the exact source licence as displayed, even when the licence is incompatible and the source is used only for facts;
- current agent-access status, terms location, check date, and notes.

Every third-party asset in `lesson.json` declares its path, source, title, creator, allowed licence or permission basis, rights evidence, attribution, and modifications. Any source whose expression, media, or dataset is reused must also appear by stable source ID in `ATTRIBUTION.md`. A general bibliography, link, disclaimer, educational purpose, or “no infringement intended” notice is not a rights basis.

## 6. Contributor and review gates

The accountable contributor attests that original lesson material can be licensed under CC BY 4.0 and that every exception is disclosed. The accessibility-and-rights reviewer independently checks source terms, exact asset identity, licence compatibility, attribution, modifications, agent-access restrictions, privacy, and provenance against the frozen candidate commit.

Review must request changes when:

- prose closely tracks a source's wording or organization;
- an example, question, diagram, table, or media item appears source-derived without a compatible recorded basis;
- a rights record is missing, contradictory, stale, or cannot be verified;
- an agent processed a `human-only` source;
- database extraction is substantial, repeated, or unclear; or
- permission depends on a legal exception or jurisdiction-specific conclusion not approved by counsel.

## 7. Complaint and removal procedure

Maintainers should publish a monitored private rights-contact address and identify the responsible operator. Until that channel exists, a claimant may open a minimal repository issue without posting personal or confidential evidence and request a private channel, or use the hosting provider's formal copyright process.

On a credible complaint, maintainers should promptly preserve the relevant commit and provenance, temporarily remove or quarantine the disputed material when appropriate, acknowledge the complaint, identify the exact work and use, review the rights record, consult counsel when needed, and document the resolution. GitHub publishes its [content-removal request](https://docs.github.com/en/site-policy/content-removal-policies/submitting-content-removal-requests) and [DMCA](https://docs.github.com/en/site-policy/content-removal-policies/dmca-takedown-policy) procedures.

Operational controls reduce risk; they do not create immunity. A maintainer should obtain jurisdiction-specific counsel before launch and consider an appropriate legal entity, contributor terms, and media liability coverage as the project grows.


---

## Repository file: CONTENT-STANDARD.md

# EmbeddedKnowledge Learning-Content Standard v1

**Status:** authoritative lesson-writing and learning-design contract
**Applies to:** lesson authors, authoring agents, learning-design reviewers, academic reviewers, and adjudicators
**Companion contracts:** [`FORMAT.md`](FORMAT.md) defines the interchange format; [`lessons/README.md`](lessons/README.md) defines pack construction; this document defines what the teaching must do.

## 1. The standard

An EmbeddedKnowledge lesson is not an article followed by a quiz. It is an instructional argument designed to help a declared learner build, test, retain, and transfer a useful mental model.

For every mapped outcome, the lesson must enable the learner to answer:

1. **What question, phenomenon, or task does this knowledge address?**
2. **What model, principle, or procedure should I construct?**
3. **Why does it work, and what evidence or reasoning supports it?**
4. **How is the same idea represented in words, diagrams, equations, data, or chemical notation?**
5. **When does the idea apply, when does it fail, and what is easily confused with it?**
6. **Can I reconstruct and use it without the lesson in front of me?**
7. **Can I select and adapt it in a new, bounded context?**

The content is ready for review only when it supports all seven questions at the depth declared by the lesson. **Comprehension is a prerequisite for learning, not a cosmetic extra.** If the declared learner cannot say what the lesson is about, why it matters, and how its central idea works after one careful first read, the lesson is not ready. Fluency, familiarity, completion, and time-on-page still are not evidence of mastery.

## 2. Design before prose

Before drafting scenes, create a private authoring brief. It does not replace structured lesson metadata and need not ship as learner-facing prose.

```text
Target outcomes:
Declared learner and level:
Required prerequisites:
Prerequisite diagnostic and recovery route:
Target explanatory model:
Observable evidence of mastery:
Common helpful prior ideas:
Common misconceptions or boundary errors:
Anchor question, phenomenon, or task:
Worked-example task:
Practice variations:
Transfer context:
Representations required and the job of each:
Material claims, sources, scope, and uncertainty:
Accessibility equivalents:
Safety, privacy, clinical, and route boundaries:
Delayed retrieval links to earlier/later lessons:
```

If the target model, learner, prerequisite route, or mastery evidence is unknown, stop and resolve it before writing polished exposition. Do not use elegant prose to conceal an undefined instructional goal.

### Outcome-first alignment

- Use only existing atomic graph outcomes and teach the full meaning of every mapped outcome.
- Write observable objectives that require a learner performance, not exposure: prefer “predict,” “explain,” “distinguish,” “calculate,” “interpret,” or “construct” over “understand” or “be familiar with.”
- Map each objective to instruction, supported practice, an assessment item, feedback, and a recovery route.
- Do not assess a distinction, convention, fact, or operation that the lesson did not teach or identify as a prerequisite.
- Do not add adjacent facts merely because they are true. Include them only when they support the target model, a prerequisite, a boundary, or transfer.

## 3. The instructional arc

Use this as a functional sequence, not a demand for nine separate scenes.

1. **Purpose:** open with a meaningful question, observable phenomenon, decision, or task. State what the lesson will and will not establish.
2. **Prior knowledge:** elicit or verify the few prerequisites that the new model actually uses. Give a non-punitive recovery route.
3. **Model construction:** explain the central relationships, mechanism, assumptions, and evidence as a connected argument.
4. **Representation:** coordinate words with the necessary diagram, equation, table, graph, molecular structure, or data. Explain what each representation reveals and hides.
5. **Modelled use:** work through a representative task while exposing expert decisions, not only calculations.
6. **Discrimination and repair:** compare cases, non-cases, boundaries, and a high-value misconception; reconstruct the correct model.
7. **Retrieval and supported practice:** ask learners to recall, explain, predict, or choose before revealing feedback; move from support to independence.
8. **Transfer:** require the learner to select and adapt the model in a novel but bounded context.
9. **Synthesis and spacing:** reconnect the outcome to the knowledge graph and schedule later retrieval rather than ending with passive restatement.

Earlier knowledge should be retrieved when it becomes useful, not recopied in full. Later lessons should revisit high-value ideas after a delay and in changed contexts.

## 4. Write explanations, not compressed notes

### First-read gate: write for a learner, not for the contract

The standard belongs in the authoring process, not in the voice of the lesson. Learners should encounter an inviting explanation, not the vocabulary of a design brief, schema, validator, or review rubric.

- Open with a concrete question, phenomenon, surprising observation, or useful problem that the learner can picture. Make the reason to continue clear before listing objectives, definitions, exclusions, or caveats.
- Give the concrete case before the abstract label when the label would otherwise be empty. A sentence such as “A baseline is a routing tool” does not orient a novice until *baseline*, *routing*, and the decision being routed have ordinary-language meaning.
- Prefer people and things doing actions over strings of abstract nouns. Use the technical term once the learner has a distinction to attach it to.
- Use the shortest ordinary word that remains accurate. Explain necessary scientific vocabulary; do not replace it with a second layer of invented course jargon.
- Keep qualifications beside the claim they constrain, but do not make the learner cross a wall of disclaimers before reaching the idea. State routine course boundaries once and return to them only when they affect a decision.
- Keep internal IDs, schema fields, claim-map terminology, candidate status, quorum language, and validator instructions out of learner prose unless the mapped outcome explicitly requires that project machinery.
- Write source notes for learners: say what the evidence supports and where it is limited. Do not narrate the claims ledger or rights workflow.
- Read representative paragraphs aloud. Rewrite any sentence that an informed teacher would not naturally say to the declared learner, or that requires a second reading because the grammar hides the main action.
- Build motivation through visible meaning and attainable progress, not hype, fake enthusiasm, decorative stories, or claims that the topic is “important” without showing what it helps explain or do.

Before completing the full pack, test the opening, central explanation, and one example as a learner-visible prototype. Give a cold reader only the declared learner, target outcome, and prototype—not the authoring brief or quality rubric. The prototype passes when that reader can explain in plain language what they are learning, why it matters, and the central relation or procedure, and can identify no sentence whose meaning depends on hidden project knowledge. This developmental read does not count toward review quorum or evidence of learning.

### 4.1 Build coherence for the declared novice

- Give each paragraph one instructional job. Lead with its controlling idea, then supply reasoning, evidence, example, qualification, or consequence.
- Make causal, conditional, contrastive, and inferential links explicit: use “because,” “therefore,” “if,” “only when,” “whereas,” and “this evidence rules out” when those relations are real.
- Keep the same name for the same entity. Do not alternate synonyms for stylistic variety when terminology carries identity.
- Replace vague referents such as “this,” “it,” or “they” when more than one antecedent is plausible.
- State intermediate reasoning that a novice cannot safely infer. Do not turn every inference into prose for an expert; write at the lesson's declared prerequisite boundary.
- Preview the conceptual route before a dense explanation and close by reconnecting the parts, not by repeating the introduction sentence by sentence.

High cohesion is not the same as low intellectual level. The goal is to expose the structure of a difficult idea, not remove the idea's necessary complexity.

### 4.2 Control terminology

- Introduce a technical term only when the underlying distinction is ready to be named.
- On first use, give the term, its meaning in this lesson, its conditions, and—when useful—a non-example or contrast.
- Flag ordinary words that acquire a narrower scientific meaning, such as “work,” “fitness,” “significant,” “organic,” or “theory.”
- Define every symbol, abbreviation, index, sign convention, and unit before relying on it.
- Prefer the precise common word over ornamental academic language. Retain disciplinary terminology that learners must master; explain it instead of replacing it with an inaccurate simplification.
- Treat readability scores as a warning signal only. Never shorten sentences, delete connectives, or substitute familiar but imprecise words merely to hit a grade-level score.

#### Terminology continuity across the course

A term does not belong to one lesson; it belongs to the course. A learner meets it once and carries that meaning forward. Consult the course terminology ledger (`site/data/premed-terminology.json`) before defining any term a published lesson already teaches, and then do one of two things:

- **Adopt.** Keep the established meaning, use the same name for the same entity, and cross-reference the owning lesson. Reword the definition only in ways that preserve the sense.
- **Protest, in the open.** When your field genuinely uses the same word in a different technical sense — "reliability" as measurement consistency versus evidence credibility, "reproducibility" under changed versus identical conditions, "negative control" as a lab baseline versus a causal-inference probe — say so. Declare the glossary entry's `alignment` block (prior lesson, relation, note) and bridge the two senses in prose where the term first appears in its new sense, exactly as the course already disambiguates "controlled variable versus control condition."

What you may not do is silently overwrite a meaning the learner already holds. A collision is invisible in a single-lesson read — every usage is locally correct — so it is caught only against the ledger and at cross-lesson review. Reserve the same word for a genuinely different sense only when the disciplines force it, and never for stylistic variety.

### 4.3 Explain mechanisms and evidence

- Distinguish description, mechanism, model, correlation, causal claim, convention, and inference.
- For a mechanism, identify relevant entities, changes, interactions, sequence, and constraints.
- For a model, state what question it answers, its assumptions, predictive use, scale, and limits.
- For an empirical claim, connect the claim to the observation or evidence that warrants it and preserve material uncertainty.
- Treat factual, quantitative, definitional, causal, mechanistic, safety, health, standards, and other evidence-dependent teaching statements as material claims. Map them to structured sources and expose that evidence in the scene where the learner encounters or applies the claim.
- Declare a scene as containing no material claims only when it is limited to routing, instructions, reflection, or another non-evidentiary learning action. Repetition in practice, retrieval, synthesis, or transfer does not make a sourced teaching claim cease to be material.
- When several explanations are plausible, say what evidence discriminates among them. Do not present the preferred explanation as inevitable.
- Never imply that a diagram, analogy, simulation, or equation is the phenomenon itself.

### 4.4 Use analogy carefully

An analogy must name:

- the source and target;
- the relations that map;
- the feature the analogy makes easier to see; and
- at least one important place where the mapping fails.

Do not elaborate decorative story details that are not used in the mapping. Retire the analogy once the scientific representation can do the work.

### 4.5 Let length follow meaning

There is no evidence-based universal word count for a scene, paragraph, or lesson. Use the shortest treatment that fully exposes the required model and the longest treatment that remains relevant and coherent.

- Split at a change in instructional purpose, learner action, representation, or resumable idea—not at an arbitrary screen quota.
- Keep a long causal chain together when breaking it would force the learner to reconstruct hidden links.
- Add a descriptive heading when a reader would need one to locate, resume, or summarize a section.
- Move optional enrichment and derivations out of the core route only when the mapped outcome does not depend on them.

## 5. Examples, cases, equations, and representations

### 5.1 Worked examples for initial learning

A worked example must include:

1. the task and what is being asked;
2. the representation of givens, constraints, and unknowns;
3. the principle or model selected and why it fits;
4. an explicit plan;
5. ordered execution with reasons for material steps;
6. units, signs, assumptions, and boundary conditions;
7. an independent check or plausibility test;
8. interpretation of the result in the original context; and
9. a self-explanation prompt about the decision that most strongly determines the solution.

After one or more complete models, fade support through completion problems, partial prompts, and independent problems. Assistance should decline as demonstrated knowledge rises; do not force advanced learners through redundant scaffolding.

### 5.2 Compare cases to reveal structure

- Use at least two cases when a single example could leave the governing feature ambiguous.
- Ask learners to identify the relation shared by analogous cases or the critical feature that distinguishes contrasting cases.
- Vary irrelevant surface features while preserving the target structure.
- Include a near miss or boundary case where discrimination is part of the outcome.
- State the general principle after the learner has inspected the cases, then verify it in another context.

### 5.3 Connect concrete and abstract

When both are useful, move from an interpretable phenomenon or concrete representation to the abstract notation, explicitly map their parts, then apply the abstraction to a different case. A concrete story is not automatically helpful: strip away features that could be mistaken for the general rule.

### 5.4 Coordinate, do not accumulate, representations

Every representation must have a declared instructional job.

- Place explanatory labels and corresponding words next to the part of a figure they explain when the format permits.
- Cue the feature or transition the learner should inspect; do not make the learner search an undifferentiated visual.
- Explain how a change in one representation appears in the others.
- Preserve consistent notation, scale, orientation, color meaning, and naming across scenes.
- Do not add a photograph, animation, character, historical anecdote, or “fun fact” solely to make the page look engaging.
- Do not narrate visible text word for word by default. Accessibility captions and transcripts remain required; learner-controlled alternatives are not “redundant decoration.”

### 5.5 Mathematics and chemistry

- Introduce an equation from a relationship or question, not as an isolated object to memorize.
- Define symbols, units, sign conventions, domain restrictions, and assumptions adjacent to first use.
- Read important expressions in words and interpret the result physically, chemically, or biologically.
- Show why a transformation is permitted in a derivation; do not hide decisive algebra in “after simplification.”
- For chemical equations, balance matter and charge, state physical form and conditions when material, and distinguish equilibrium from one-way change.
- Coordinate macroscopic observation, particulate model, and symbolic notation where the outcome crosses those levels.

## 6. Make the learner generate and retrieve

Interaction counts only when it requires relevant thought. Clicking “next,” revealing definitions immediately, copying a displayed procedure, and decorating a page with controls are not active learning.

### Retrieval checks

- Put the prompt before the answer or explanation.
- Ask for recall, explanation, prediction, discrimination, representation, or a next step—not recognition alone unless discrimination is the construct.
- Make the first attempt low stakes and psychologically safe.
- Give the correct answer and reasoning after an attempt; address likely wrong reasoning, not just the wrong option.
- Revisit high-value knowledge after a delay and in a changed context.
- Use cumulative checks that connect prerequisite and current outcomes.

### Practice progression

Use a deliberate progression:

```text
complete model → completion problem → supported problem → independent problem → mixed selection → transfer
```

- Keep initial tasks close enough to the model for successful schema construction.
- Vary numbers, surface features, representation, and context without silently changing the underlying construct.
- Interleave categories when the learning goal is choosing among confusable strategies or concepts. Block initial practice when the learner is still acquiring each procedure.
- Include enough practice to reveal a pattern of performance; do not infer mastery from one lucky response.
- Ensure retakes require renewed retrieval or reasoning rather than replaying an identical answer sequence.

### Feedback

Useful feedback answers:

1. What was correct or incorrect?
2. Which principle, step, or distinction explains that result?
3. What should the learner do next?

Prefer task, process, and self-regulation information over praise or judgment about the person. Keep feedback close enough to the attempt to repair the model, while allowing a genuine retrieval attempt first.

## 7. Repair misconceptions without caricature

Use a misconception only when sources or strong domain-review evidence show it is common, consequential, or highly plausible from prerequisite knowledge.

A repair sequence must:

1. elicit a prediction or commitment where appropriate;
2. state the incorrect model accurately and neutrally;
3. identify the observation, counterexample, or consequence it cannot explain;
4. supply a more adequate model;
5. compare the models on the same case; and
6. recheck the learner in a new case.

Do not merely label an answer “a common misconception.” Do not shame an intuitive model, invent a straw version, or assume that one contradiction produces durable conceptual change.

## 8. Design for retention and transfer

- Treat retrieval and spacing as properties of the course sequence, not a single end-of-lesson quiz.
- Reuse graph prerequisites as later retrieval opportunities and record delayed links in lesson design.
- Mix problem types only after each type has a usable initial representation.
- Assess the outcome at the level taught: factual recall, conceptual explanation, procedural execution, model selection, data interpretation, or transfer are not interchangeable.
- A transfer task must require selection, adaptation, or coordination in a context not copied from the worked example. Changing only names or numbers is near practice, not genuine transfer.
- Ask learners to explain why an answer follows and when it would change. Correct output without a defensible model is insufficient for conceptual mastery.

## 9. Digital book and print behavior

The canonical source is semantic, not slide-sized.

- A scene is a bounded learning purpose; a frame is an adaptive presentation subdivision.
- Guided mode may reveal a scene in page-like frames. Reading mode must preserve continuous, searchable, selectable prose for sustained study.
- Never delete reasoning to make a frame fit. Split presentation at meaningful subheadings or use bounded local overflow with an obvious continuation.
- Preserve stable progress, headings, source order, backtracking, and the learner's place.
- Keep core explanations available without hover, animation timing, or an online-only interaction.
- Print/PDF export must preserve the complete explanation, prompt conditions, visual equivalents, citations, and answer references even when the interactive presentation differs.

## 10. Epistemic and scholarly integrity

- Verify each material factual claim against the source actually cited; a plausible citation is not evidence.
- Treat evidence and reuse rights as separate checks. Follow `RIGHTS-POLICY.md`: independently synthesize facts in original language and instructional structure rather than closely paraphrasing source expression.
- Record every source use, rights basis and evidence, and dated agent-access state. An agent must not process a source marked `human-only`.
- Prefer primary standards, primary research, systematic reviews, and current scholarly reference works for the claims they can support.
- Use synthesis sources to establish consensus and primary sources for specific findings when practical.
- Record the population, conditions, jurisdiction, model, date/version, and uncertainty that limit a claim.
- Distinguish established consensus, active debate, simplifying model, historical interpretation, and author inference.
- Put qualifications next to the claim they constrain.
- Do not fabricate data, quotations, patients, experiments, citations, worked results, or source access.
- A fictional teaching case must be labelled fictional and must not imply clinical prevalence or diagnostic authority.
- Premed content is educational. It must not direct diagnosis, treatment, unsafe self-experimentation, ingestion, pathogen culture, or circumvention of professional supervision.

## 11. Accessibility is part of the explanation

- Use a logical heading hierarchy and semantic source order.
- Explain uncommon jargon and abbreviations at first use.
- Give complex figures a short identification and an outcome-equivalent long description; provide data tables for quantitative charts.
- Express important equations and chemical relationships in words as well as notation.
- Never encode a distinction only by color, position, sound, motion, hover, pointer gesture, or time.
- Provide keyboard and non-drag routes for interactions, captions and transcripts for media, and non-physical alternatives for practical work where appropriate.
- Use descriptive references such as “the increase from pH 4 to pH 5” rather than “the green line on the right” as the only locator.
- Test at 200% zoom, reflow, narrow viewport, keyboard-only use, reduced motion, and representative screen-reader navigation.

An alternative must preserve the learning operation, not merely announce that a visual or interaction exists.

## 12. Required agent workflow

An authoring agent must complete these phases in order.

### Phase A — Ground

1. Read `llms.txt`, this standard, the format and pack guides, `RIGHTS-POLICY.md`, the graph outcome, schemas, and current quorum policy.
2. Confirm the outcome is uncovered and unclaimed.
3. Build the authoring brief and a source dossier with exact support, scope, uncertainty, version, use type, rights basis/evidence, and agent-access terms.
4. Identify the prerequisite route and high-value learner difficulties from evidence; do not invent “common misconceptions.”

### Phase B — Design

1. Write the target explanatory model in a few precise propositions.
2. Choose an anchor phenomenon or task and the smallest set of representations that reveal the model.
3. Design the worked example, faded practice sequence, transfer task, retrieval schedule, feedback, and recovery routes before polishing prose.
4. Map every objective to scenes and assessment evidence.
5. Choose the smallest instructional arc that can teach and assess the outcome. Do not turn the nine functional stages into nine compulsory scenes or repeat the same explanation to satisfy multiple checklist labels.

### Phase C — Draft

1. Draft a learner-visible prototype containing the opening, core causal or relational explanation, and one worked use before constructing the rest of the pack.
2. Run the first-read gate above in a fresh context where possible. Repair the prototype before multiplying its voice across scenes, practice, ledgers, and assessment.
3. Add definitions, representations, examples, qualifications, and sources where they reduce a real learner inference gap.
4. Add prompts before explanations they are intended to retrieve.
5. Keep essential content in canonical source, not styling or interaction state.

### Phase D — Challenge

Run four separate self-audits before requesting independent review:

- **Academic:** recompute results; check terminology, scope, current consensus, citations, safety, and all representations against each other.
- **Learning:** attempt every prompt as the declared learner; inspect prerequisite gaps, hidden steps, scaffolding, feedback, misconception repair, spacing, and transfer.
- **Clarity:** read learner-visible scenes without metadata or ledgers; remove irrelevant detail and rubric leakage; replace compressed abstraction with concrete meaning; repair ambiguous referents and terminology drift; expose causal links; read representative passages aloud; and check long-form continuity.
- **Accessibility and rights:** verify equivalent meaning and operation, reflow, navigation, original synthesis versus close paraphrase, source-use and agent-access declarations, asset provenance, licence, privacy, and non-clinical boundaries.

Self-audit does not create a review artifact or count toward quorum.

### Phase E — Validate and freeze

1. Resolve schema, link, rendering, overflow, accessibility, and deterministic-build failures.
2. Verify that the candidate satisfies every gate below.
3. Freeze one candidate commit and record provenance before independent review begins.
4. Never manufacture approvals, adjudication, learner results, or evidence of effectiveness.

## 13. Acceptance gates

Every standard lesson must pass all applicable gates. A reviewer records evidence and a concrete revision, not a vague quality score.

| Gate | Pass evidence | Blocking failure |
|---|---|---|
| Alignment | Outcomes, objectives, instruction, practice, and assessment describe the same performance. | Content or assessment drifts beyond the mapped outcome. |
| Human first read | A cold reader at the declared level can explain what the lesson teaches, why it matters, and how the central idea works; the opening and examples use natural, concrete language without hidden project machinery. | The prose is grammatically valid but opaque, bureaucratic, needlessly abstract, validator-facing, or dependent on unexplained internal vocabulary. |
| Learner model | Declared level, prerequisites, diagnostic, and recovery route are usable. | Hidden prerequisites or no recovery path. |
| Coherence | The central model forms a connected causal, relational, or procedural argument. | Fact list, unexplained jumps, or terminology drift. |
| Accuracy | Claims, calculations, notation, representations, and boundaries agree with verified sources. | Material error, unsupported claim, or fabricated source. |
| Model use | At least one complete worked example exposes selection, reasons, checks, and interpretation. | Answer-only demonstration or unguided initial problem solving. |
| Discrimination | Examples, non-examples, boundaries, or misconception repair reveal the governing distinction. | One example is treated as the rule or misconception is merely labelled. |
| Generation | Learners retrieve, explain, predict, compare, calculate, interpret, or construct before feedback. | Interaction is passive reveal or copying. |
| Practice | Support fades; tasks vary; feedback diagnoses reasoning; retry and remediation exist. | Single attempt, answer-only feedback, or no route after error. |
| Transfer | A novel bounded task requires selecting or adapting the model. | Only names or numbers change. |
| Retention | Delayed/cumulative retrieval connections are identified. | All practice is immediate and isolated. |
| Representation | Each visual/symbolic form has a job and is explicitly connected to the others. | Decorative media, split attention, or inconsistent notation. |
| Accessibility | Equivalent meaning and operation survive nonvisual, keyboard, zoom/reflow, print, and reduced-motion paths. | Outcome depends on one sensory or motor channel. |
| Epistemics | Evidence type, scope, uncertainty, version, and educational/clinical boundary are visible. | False certainty, authority laundering, or unsafe implication. |
| Rights and provenance | Authorship, agent runs, original synthesis, source uses, agent-access terms, assets, licences, modifications, and rights evidence are complete. | Close paraphrase, unknown origin, incompatible rights, prohibited agent access, or undisclosed material agent use. |

## 14. Prohibited shortcuts

Do not:

- tailor instruction to unsupported “visual,” “auditory,” or “kinesthetic” learning-style labels;
- equate shorter prose, more pictures, more interaction, or more entertainment with better learning;
- use unguided discovery as the default for a novice's first encounter with a complex model;
- use decorative detail to manufacture interest;
- make readability grade level, word count, frame count, or completion rate the learning target;
- convert exposition into fragments that no longer form a coherent explanation;
- confuse density, abstraction, or institutional tone with rigor;
- expose the authoring checklist, internal graph machinery, evidence ledger, or governance workflow as if it were the lesson;
- mechanically create one scene for every stage of the instructional arc or repeat the same point under several rubric headings;
- present a formula, diagram, or mnemonic without meaning, conditions, and use;
- call repeated exposure “retrieval practice”;
- call a changed number a transfer task;
- call a bibliography a claim-evidence map;
- call validation, agent self-critique, or model agreement evidence that students learned.

## 15. Evidence basis and revision

This standard is based on the dated rapid evidence synthesis in [`research/CONTENT-AUTHORING-EVIDENCE-2026-07-19.md`](research/CONTENT-AUTHORING-EVIDENCE-2026-07-19.md). The strongest recurring support concerns explicit novice guidance, coherent text, worked examples, retrieval practice, spacing, feedback, relevant and integrated representations, active generation, and misconception repair. Interleaving, prequestions, multiple representations, emotional design, concept maps, and digital pagination are useful under identifiable conditions rather than universal mandates.

Educational research is heterogeneous and does not establish a single optimal textbook recipe. This standard therefore separates durable defaults from context-sensitive choices, requires declared learners and outcomes, and treats real learner testing as the next evidentiary layer. Revise the standard when stronger syntheses, Premed learner data, accessibility findings, or format capabilities warrant a change; preserve the reason and evidence for each material revision.


---

## Repository file: CONCEPT.md

# EmbeddedKnowledge product concept

**Status:** working product contract · revised for a public, multi-audience project on 19 July 2026

EmbeddedKnowledge is an open, versioned library of university-grade knowledge maps and reviewed learning materials. It separates access to rigorous learning architecture from admission, tuition, geography, scheduling, and credential requirements without pretending that open study is equivalent to an accredited degree.

Premed is the first book: a pathway-aware scientific foundation for medical study. Its syllabus and graph define the learning territory. Its lesson commons began empty and grows only through reviewed contributions.

## Public audiences

EmbeddedKnowledge serves overlapping communities:

1. independent learners seeking substantial academic depth without enrolling for a credential;
2. prospective applicants strengthening foundations or auditing institution-specific prerequisites;
3. health-adjacent professionals filling gaps across chemistry, biology, physics, biochemistry, and human systems;
4. educators and institutions evaluating, adapting, or assigning open material;
5. subject experts, learning designers, accessibility reviewers, librarians, and rights reviewers improving the corpus;
6. open-source contributors and agents producing structured, inspectable work.

No single biography, country, study route, age group, institution, or agent environment defines the product.

## Product principles

- **Open maps, explicit boundaries.** Curricula, dependencies, outcomes, sources, and disagreements remain inspectable. Completion does not confer university credit, admission, clinical competence, or professional authorization.
- **A knowledge system, not a content pile.** Stable identifiers connect outcomes, prerequisites, lessons, assessments, evidence, review, and corrections.
- **Mastery is revisable.** Practice, dialogue, problems, investigation, delayed retrieval, and transfer provide evidence over time. Recovery and retakes are part of rigor.
- **Many routes remain visible.** Portable foundations and country-, examination-, or institution-specific requirements are not collapsed into a fictional global standard.
- **Accessibility is instructional integrity.** Equivalent nonvisual, non-audio, keyboard, language, and cognitive paths are designed into lessons and reviewed before publication.
- **Sources outrank confidence.** Human expertise and model output remain accountable to evidence, declared uncertainty, and correction.
- **Corpus state is never implied.** Every outcome is visibly empty, proposed, under review, adjudicated, or covered by a published lesson.

## The four public artifacts

1. **Syllabus:** the academic contract—audience, scope, outcomes, modules, practicals, assessment, schedules, quality rules, and boundaries.
2. **Knowledge graph:** stable atomic outcomes and the prerequisite, hierarchy, and cross-domain relationships among them.
3. **Lesson commons:** openly licensed explanations, activities, assessments, references, accessibility alternatives, and recovery guidance mapped to the graph.
4. **Review record:** structured independent reviews, dissent, conflicts, limitations, quorum, and final adjudication attached to each published lesson.

The syllabus and graph may be complete while the lesson commons remains incomplete. The interface must preserve that distinction everywhere.

## Learning experience

Each learning route draws from the same outcome graph while permitting multiple media and teaching approaches. A mature lesson can include:

- concise orientation and prerequisite checks;
- explanations across words, diagrams, equations, mechanisms, and models;
- worked examples and progressively independent practice;
- retrieval, spacing, interleaving, and misconception repair;
- practical, simulation, or dataset investigations with modality disclosed;
- Socratic dialogue and applied cases;
- mastery checks, answer logic, rubrics, feedback, and recovery paths;
- accessible alternatives and explicit source provenance.

Future tutors, feeds, audio, video, and adaptive interfaces are clients of this corpus. They do not become the source of truth.

## Agent-first contribution model

Agents are expected to help research, draft, map, test, review, and maintain lessons. The project therefore exposes concise agent context, JSON Schemas, graph data, stable identifiers, deterministic validation, and examples.

Pull requests are the only write path. Website and WebMCP tools are read-only discovery surfaces. During the founding stage, quorum uses isolated, attributable agent runs so one disclosed maintainer can build the initial corpus without inventing human consensus.

A standard lesson requires one academic and one learning-design review artifact from two unique runs across two providers, both targeting the same original candidate. A third fresh run considers both records, disposes every finding, writes the final version once, audits accessibility and rights, and adjudicates that final commit. Reviews may honestly request changes without forcing another cohort.

**Provenance is attested, not verified.** Run IDs, providers, models, versions, and instruction digests are self-reported strings supplied by the accountable operator. No platform check confirms that a declared run happened, that it used the declared provider or model, or that three artifacts came from three genuinely different systems. What the project guarantees is that these facts are *disclosed on the record and attested by a named operator* who can be held to them, and that the validators enforce internal consistency (distinct run IDs, distinct declared providers, role minimums, one frozen candidate commit). Treat the diversity requirement as an accountability and hygiene mechanism, not as cryptographic proof of independence.

## Open-source and licensing model

Original course content and curriculum data are licensed CC BY 4.0 unless marked otherwise. Third-party assets retain their own terms. Software is licensed separately under the MIT License.

The repository, pull requests, checks, review artifacts, adjudications, and correction history form the durable public record. A lightweight static site may be hosted through GitHub Pages while the corpus is read-only. Interactive or sensitive services require separate infrastructure.

## Sustainability

The knowledge commons remains open. Sustainable services may later include hosted tutoring, generated media, managed institutional deployments, or other conveniences, but the canonical curriculum, published lessons, and review history must remain independently reusable.

Community-contributed compute may support drafting and review, but no donated model changes publication state by itself. Accountable principals, declared evidence, portable checks, and adjudication remain the governance boundary.

## Development sequence

1. **Map:** complete syllabus, graph, searchable views, and honest coverage ledger.
2. **Protocol:** schemas, PR templates, review roles, quorum checks, adjudication, and agent discovery.
3. **Lesson commons:** viewer for empty, proposed, under-review, adjudicated, and published lesson states; one gold-standard reference lesson.
4. **Learning tools:** mastery records, retrieval, practice, dialogue, and accessible lesson delivery.
5. **Community operations:** contributor onboarding, reviewer development, corrections, releases, and transparent quality reporting.
6. **Additional books:** expand only after the Premed production and review loop is demonstrated in public.

The immediate release criterion is not the number of generated pages. It is whether a diverse group can inspect one outcome, propose a lesson, review it independently, adjudicate it, publish it, and see the corpus ledger update without relying on private context.


---

## Repository file: FORMAT.md

# EmbeddedKnowledge Lesson Format v1

**Status:** authoritative authoring and interchange contract
**Metadata discriminator:** `"schemaVersion": 3`, `"format": "embeddedknowledge-lesson-v1"`
**Applies to:** Premed lesson packs proposed under the contribution and review protocol

This document defines what a lesson pack is, how its files relate, and which content constructs are safe and portable. JSON carries identity, sequence, outcomes, claims, sources, assessments, and glossary data. Constrained MyST-compatible Markdown carries the teaching prose. Assets and diagrams are local, inspectable, and non-executable.

The format does not alter publication governance. A structurally valid pack is only a candidate. It becomes a published lesson only after the applicable independent review quorum, a separate adjudication, a merge decision, and protected-branch validation. Files under `examples/` are specimens and never count toward lesson coverage.

## 1. Canonical pack

The standard names below are recommended. `lesson.json` may point to other safe relative JSON or Markdown paths where the schema permits.

```text
lessons/PREM-CHE-001-short-slug/
  lesson.json
  assessment.json
  references.json
  claims.json
  glossary.json
  ATTRIBUTION.md
  content/
    01-orientation.md
    02-concept.md
    03-worked-example.md
    04-practice.md
  assets/
    local-static-image.svg
  diagrams/
    relationship.diagram.json
  reviews/
    *.json
  adjudication.json
```

`lesson.json` is the manifest and ordered table of contents. Scene Markdown has no front matter; duplicating metadata in prose creates drift. Review and adjudication artifacts remain governed by their existing schemas.

## 2. Metadata and scenes

The public schema is [`site/schemas/lesson.schema.json`](site/schemas/lesson.schema.json). Important fields are:

| Field | Contract |
|---|---|
| `id`, `version` | Stable lesson ID and semantic version |
| `status`, `riskTier` | Workflow state and applicable quorum tier |
| `outcomes`, `prerequisites` | Existing atomic graph topic IDs only |
| `objectives` | Observable statements, each mapped to declared lesson outcomes |
| `estimatedMinutes` | Exact sum of ordered scene estimates |
| `scenes` | At least four ordered semantic units |
| `scenes[].claimCoverage` | Explicitly `claims-mapped` or `no-material-claims` for every scene |
| `authors`, `aiAssistance` | Accountable principal and material agent provenance |
| `license`, `thirdPartyAssets` | Rights declaration; third-party assets remain explicit |
| `sourceConfidence` | Placement confidence, never a substitute for independent review |

Each scene is:

```json
{
  "id": "scene-concept",
  "title": "Proton transfer",
  "kind": "concept",
  "source": "content/02-concept.md",
  "estimatedMinutes": 20,
  "required": true,
  "claimCoverage": "claims-mapped"
}
```

Scene order is array order. Filenames help humans but do not establish order. Scene IDs are stable internal targets for claims, assessments, glossary entries, rendered navigation, and corrections.

The exact scene-kind vocabulary is:

- `orientation`
- `diagnostic`
- `concept`
- `definition`
- `derivation`
- `worked-example`
- `investigation`
- `retrieval-check`
- `misconception`
- `practice`
- `transfer`
- `synthesis`
- `assessment`
- `references`

Not every kind needs a separate scene. Scene boundaries should reflect a useful learner interaction or resumable unit, not force instructional obligations into artificial pages. A complete lesson must still include observable objectives, prerequisite guidance, explanation, a worked example, independent practice, retrieval, transfer, misconception handling, accessible equivalents, and recovery guidance. These may be represented by scene kinds, directives, assessment items, and named sections across the pack.

## 3. Constrained scene Markdown

Scene files use UTF-8 Markdown with a conservative MyST-compatible extension.

1. Begin with exactly one level-one heading.
2. Use headings only through level three.
3. Do not add YAML front matter.
4. Do not use raw HTML, inline styles, scripts, iframes, remote embeds, or Markdown image syntax.
5. Use ordinary paragraphs, lists, tables, emphasis, block quotes, and safe links.
6. External links must use HTTPS. Fragment and safe relative links are allowed.
7. Code fences are inert text and must declare one of `text`, `json`, `csv`, `python`, or `r`. A renderer must never execute them automatically.
8. Use the directives and inline roles below for semantic content.

### Directive whitelist

Only these colon-fenced directives are valid:

| Directive | Intended use |
|---|---|
| `{definition}` | A bounded term or relation definition |
| `{theorem}` | A named formal statement used by the lesson |
| `{derivation}` | Ordered symbolic reasoning with assumptions |
| `{worked-example}` | Fully reasoned model of a task |
| `{check}` | Diagnostic or retrieval prompt |
| `{misconception}` | Incorrect model, why it fails, and corrective test |
| `{investigation}` | Practical, simulation, dataset, or experimental inquiry |
| `{figure}` | Validated local static image |
| `{diagram}` | Validated local declarative diagram source |
| `{chart}` | Validated local declarative chart source |
| `{equation}` | Display mathematics |
| `{chemistry}` | Display chemical notation or reaction |
| `{source-note}` | Visible bridge from prose to claim and source IDs |
| `{callout}` | Bounded note such as a safety, accessibility, boundary, or recovery message |

Directive syntax is deliberately small:

```md
:::{worked-example}
:id: worked-example-conjugate-pairs

Reasoned content goes here.
:::
```

The available option names are `id`, `label`, `kind`, `alt`, `longdesc`, `claims`, and `sources`. Directives may not nest in format v1. Renderers must escape directive content before applying trusted templates.

The inline roles ``{math}`...` `` and ``{chem}`...` `` are accepted for short expressions; `$...$` is also accepted as familiar inline TeX. Use `{equation}` or `{chemistry}` for display content so labels and semantic intent remain explicit. Unlabelled `$$...$$` is compatible input, but the directive form is preferred.

## 4. Mathematics and chemistry

Math is a constrained TeX payload. Chemistry uses `\ce{...}` within an inline `{chem}` role or display `{chemistry}` directive. The current specimen build compiles both to MathML with a pinned Temml renderer; another implementation may use an equivalently bounded renderer without changing the canonical source.

```md
The pair differs by {chem}`H+`.

:::{equation}
:label: equation-buffer-response

A^- + H^+ \rightarrow HA
:::
```

The source is always preserved as text. A non-math reading path must express the same relationship in prose. Rendering may improve presentation but must not change meaning.

The validator rejects TeX commands that define macros, read or write files, load packages or resources, create HTML/classes/styles, or create links. This includes command families such as `\input`, `\include`, `\write`, `\newcommand`, `\href`, `\url`, `\html`, and `\require`. Math and chemistry never execute code or load remote resources.

## 5. Figures and diagrams

### Static figures

Images must be pack-local `.png`, `.jpg`, `.jpeg`, `.webp`, `.avif`, or sanitized `.svg` files and must use a `{figure}` directive:

```md
:::{figure} ../assets/conjugate-pairs.svg
:alt: Acetate and acetic acid form one pair; water and hydroxide form the other.
:longdesc: Two labelled relationships show that each conjugate pair differs by one proton, including the donor and acceptor roles.
:::
```

Both `alt` and `longdesc` are required. The long description must preserve the relationship or evidence carried by the visual, not merely describe its appearance. Unique information may not depend only on color, animation, hover, drag, audio, or spatial position.

SVG is treated as untrusted input. Validation requires internal `<title>` and `<desc>` elements and rejects scripts, event handlers, stylesheets, external references, embedded HTML/media, `data:` or `javascript:` URLs, entities, and resource-loading constructs. Renderers should sanitize again at the output boundary.

### Where an image comes from

Every image a scene shows must be accounted for. A pack lists the files it made itself in
`originalAssets`, and anything it did not make in `thirdPartyAssets` with its source, creator,
licence, the rights evidence the author actually read, and whether it was modified and how. A file
that appears in neither is rejected: an image with no stated provenance is the one way an unlicensed
figure reaches a learner, and silence must not be a valid answer.

The declared attribution is rendered beside the image. A credit that lives only in `lesson.json`
satisfies an audit rather than the licence, and CC BY asks for credit given in a reasonable manner
to the person who reads the work.

Only `CC-BY-4.0`, `CC0-1.0`, `Public-Domain`, and `Written-Permission` are accepted. Share-alike is
absent deliberately: it obliges an adaptation to carry the same licence, and cropping, relabelling,
or recolouring a figure makes an adaptation. `site/agent/illustration-sources.json` lists candidate
pools, what to check before using one, and the quality bar a scientific illustration has to clear.

### Declarative diagrams

Diagram source uses [`site/schemas/diagram.schema.json`](site/schemas/diagram.schema.json). It supports bounded flow, concept-map, process, cycle, and timeline structures with labelled nodes and edges.

The picture is computed from the `edges`, not from the order the nodes are declared in. Nodes are placed in layers by longest path from a root, so a chain reads as a chain and genuine siblings sit side by side, and an arrow appears only where an edge was declared. A cycle is drawn: the edge that closes it is found and drawn as a dashed return rather than laid out as if it went forward. `direction` is a hint and not a guarantee — the renderer will lay a diagram along the other axis when the declared one would shrink its labels below about 9px in the reading column, because a diagram nobody can read is not a diagram. Node labels are the author's and are never shortened; a long one wraps and grows its own row.

Two consequences for authors. Declare every relationship you want drawn, because nothing is inferred from adjacency. And keep `longDescription` a faithful text equivalent of the same graph, since it is the alternative for readers who do not see the picture and the two are checked against each other.

```md
:::{diagram} ../diagrams/model-scope.diagram.json
:alt: Proton-transfer cases are a subset of the broader electron-pair model.
:longdesc: The diagram connects a Brønsted–Lowry proton-transfer node to a broader Lewis donor-and-acceptor node and then to a model-choice question.
:::
```

The JSON source and its long description are canonical. Mermaid, DOT, Vega-Lite, canvas, SVG, or another presentation can be derived later, but those languages are not accepted as canonical lesson diagram source in format v1. Layout engines must not be allowed to fetch URLs or emit active markup.

### Charts

Quantitative lessons need distributions, not just relationships: a bell curve with a shaded tail, a scatter of two measurements, a histogram of a sample. Chart source uses [`site/schemas/chart.schema.json`](site/schemas/chart.schema.json) and is referenced with a `{chart}` directive:

```markdown
:::{chart} ../charts/sampling-distribution.chart.json
:::
```

The same rule as diagrams applies, and for the same reason: the JSON carries data and encoding, and the picture is derived from it at build time. A lesson never ships a plotting library, a script, or a remote embed, so a chart cannot execute, cannot fetch, and renders identically wherever the pack is read.

Supported `type` values are `normal-curve`, `line`, `scatter`, `bar`, `histogram`, and `box`. A normal curve is parameterised by `mean` and `sd` rather than listed as points, so a sampling distribution stays exact and legible in source. `shadedRegions` mark tail areas and intervals; `markers` label a mean, a cut-off, or an observed value. A series may set `mark` to `point` or `line` to override the chart type's default, which is how a scatter of observations carries a fitted line over the same axes.

Every chart requires `alt` and `longDescription`, and the renderer emits the underlying numbers as a data table beside the picture. A distribution a learner cannot see is still a set of numbers they must be able to read, so the text equivalent is part of the chart rather than a courtesy attached to it.

## 6. Claims and sources

Academic prose does not become traceable merely because a bibliography exists. Format v1 separates a source registry from a claim ledger:

- [`references.schema.json`](site/schemas/references.schema.json) v2 records stable source IDs, source type, title, creators, publisher, issued/version information, exact HTTPS URL, retrieval date, locator, displayed license, each use and rights basis, rights evidence, and dated agent-access status.
- [`claims.schema.json`](site/schemas/claims.schema.json) records the claim statement, scenes where it appears, sources supporting it, direct/synthesis/context/inference status, confidence, scope, uncertainty, risk, and review status.

Every scene must declare one of two evidence states:

- `claims-mapped` means the scene contains one or more material teaching claims. Every claim mapped to that scene in `claims.json` must appear in a learner-visible `{source-note}` in the same scene, and that note must list every source supporting its named claims.
- `no-material-claims` means the scene contains orientation, instructions, reflection, navigation, or another activity with no factual, quantitative, definitional, causal, or evidence-dependent teaching claim. The validator rejects this declaration when any claim maps to the scene or the scene contains a source note.

Every claim must reference valid scene and source IDs. Source notes are required evidence bridges rather than optional bibliography decoration:

```md
:::{source-note}
:claims: claim-bronsted-model
:sources: source-openstax-chemistry, source-iupac-gold-book

These sources support the bounded definitions used above.
:::
```

Listing a source does not grant reuse rights. Follow [`RIGHTS-POLICY.md`](RIGHTS-POLICY.md): default to facts-only research and an independently created instructional structure and expression; do not closely paraphrase. Any quotation, adaptation, media redistribution, or dataset extraction must use a permitted recorded basis. Agent-access terms are checked separately, and an authoring agent may not process a `human-only` source. `confidence: high` means strong support within the stated scope; it does not mean universal applicability, clinical authority, or exemption from review.

A `published` lesson may contain only claims whose `reviewStatus` is `reviewed`, and its lesson-level `sourceConfidence` may not remain `pending-review`. Authors keep both states pending in the frozen candidate; independent review and final adjudication determine whether publication is justified. After a merge adjudication, the operator may promote only retained claims' `reviewStatus` values and the permitted lesson status/source-confidence fields. Any other claim-ledger or lesson-content change stales the reviews.

Health-sensitive and safety-critical claims must be labelled in `claims.json`. Existing risk-tier, review-quorum, correction, and adjudication rules continue to govern them.

## 7. Assessments and glossary

[`assessment.schema.json`](site/schemas/assessment.schema.json) requires:

- a retakeable attempt policy;
- item-to-scene, objective, and graph-outcome mappings;
- a machine-readable response specification;
- the correct response, answer reasoning, acceptable variants, and common errors;
- a rubric reference and explicit safety-critical flag;
- rubric criteria with point values.

Answer logic is part of the open lesson pack. A learner interface may withhold it until an attempt, but it must not delete it from the canonical corpus. Timed attempts are optional; no assessment can disable retakes.

[`glossary.schema.json`](site/schemas/glossary.schema.json) gives each term a stable local ID, definition, aliases, and scene references. Glossary definitions should state the sense used in the lesson, not pretend to replace broader disciplinary usage.

## 8. Accessibility, recovery, and safety

Accessibility and recovery are lesson-wide obligations, not mandatory scene types. Validators require an explicit `Accessibility and alternatives` section or equivalent accessibility callout and an explicit `Recovery route` section or equivalent recovery callout somewhere in the scene sequence.

A valid alternative preserves the learning operation. For example:

- a nonvisual diagram alternative states nodes, relationships, direction, and conclusion;
- a chemical equation is also explained in words;
- audio has a transcript and does not carry unique assessment information;
- a drag interaction has an equivalent keyboard and non-drag response;
- personal physiology, disability, medication, diagnosis, or other health data is never required;
- a simulation or supplied dataset remains available without penalty where physical practical work is unsuitable.

Lessons must state educational and medical boundaries where context could be mistaken for advice. No scene, asset, assessment, or investigation may direct diagnosis, treatment, unsafe home experimentation, pathogen culture, ingestion, personal medical testing, or circumvention of supervision.

## 9. Security boundary

All lesson contributions are untrusted until validated and reviewed.

- Paths are relative, may not traverse outside their pack, and may not use backslashes or absolute paths.
- Raw HTML and Markdown images are rejected.
- Only HTTPS external links are accepted.
- Remote images, scripts, frames, styles, fonts, media, and data URLs are rejected.
- Canonical diagrams are bounded JSON, not executable graph languages.
- Markdown code fences are displayed only; no renderer or export step may execute them.
- SVG and math are checked on input and must be sanitized again on output.
- Generated HTML must escape prose and option values and render only through trusted component templates.
- Renderers should apply a restrictive Content Security Policy and never use lesson content as JavaScript, CSS, an event handler, or unsanitized `innerHTML`.

Validation is necessary, not sufficient. Reviewers remain responsible for unsafe instructions, misleading health framing, inaccessible equivalence, licensing, and academically harmful omissions that syntax cannot detect.

## 10. Deterministic validation

`scripts/validate-lessons.mjs` validates production packs and the non-production specimen. It checks:

- all JSON Schemas;
- graph outcome mappings;
- safe in-pack files;
- unique scene, objective, source, claim, item, rubric, glossary, diagram-node, and diagram-edge IDs;
- exact scene-minute totals;
- artifact identity against lesson ID and version;
- assessment links to scenes, objectives, outcomes, rubrics, options, and answer logic;
- claim-to-scene and claim-to-source integrity;
- scene evidence declarations, same-scene learner-visible source notes, and complete claim-to-note source coverage;
- reviewed-only claims and resolved source confidence for published lessons;
- source-use, rights-basis/evidence, and dated agent-access declarations;
- compatible third-party asset identity, rights evidence, attribution, and modification records;
- glossary-to-scene integrity;
- scene Markdown headings, links, roles, directives, code fences, and required lesson-wide features;
- figure accessibility and SVG safety;
- diagram schema and edge endpoints;
- math and chemistry command restrictions;
- existing review, isolated-run quorum, provenance, and adjudication invariants for production packs.

Run:

```text
npm run agent:validate
npm run validate
```

The reference fixture is [`examples/lesson-pack/`](examples/lesson-pack/). It must remain `draft`, version `0.0.0`, under `examples/`. The validator reports it separately, while production indexes and progress builders scan only `lessons/`.

## 11. Public rendered data

A renderer should preserve the machine-readable order and provenance rather than flattening a lesson into one opaque HTML string. The recommended detail payload is:

```json
{
  "schemaVersion": 3,
  "format": "embeddedknowledge-lesson-v1",
  "id": "PREM-CHE-001",
  "version": "0.1.0",
  "title": "Acid–base models",
  "outcomes": [],
  "prerequisites": [],
  "objectives": [],
  "scenes": [
    {
      "id": "scene-concept",
      "title": "Proton transfer",
      "kind": "concept",
      "estimatedMinutes": 20,
      "required": true,
      "claimCoverage": "claims-mapped",
      "source": "content/02-concept.md",
      "contentHtml": "...sanitized output..."
    }
  ],
  "assessment": {},
  "references": {},
  "claims": {},
  "glossary": {},
  "attributionHtml": "...sanitized output..."
}
```

Index payloads may omit scene bodies. Exports should include original scene Markdown and JSON artifacts alongside derived HTML so another implementation can audit, rerender, or transform the lesson without scraping presentation markup. Rendered output is a derivative cache; the reviewed pack remains authoritative.


---

## Repository file: lessons/README.md

# EmbeddedKnowledge Lesson Format v1

This directory is the production namespace for reviewed EmbeddedKnowledge lesson packs. There are currently **0 contributed lessons and 0 published lessons**. The specimen under `examples/lesson-pack/` demonstrates the format, but it is not a contribution, does not enter review quorum, and never counts toward course coverage.

Read `AGENTS.md`, `CONTRIBUTING.md`, [`CONTENT-STANDARD.md`](../CONTENT-STANDARD.md), [`RIGHTS-POLICY.md`](../RIGHTS-POLICY.md), and `COLLABORATION.md` before preparing a lesson. The content standard defines what the teaching must do; the rights policy defines what source access and reuse are allowed. The public JSON Schemas remain the machine-enforced interface; this guide defines the pack, authoring, and rendering contract those artifacts express.

## 1. Canonical authoring stack

`EmbeddedKnowledge Lesson Format v1` is:

- constrained MyST-compatible Markdown, using a CommonMark/GFM base and a documented whitelist of colon-fence directives;
- one ordered Markdown file per semantic lesson scene, with no raw HTML or JavaScript;
- JSON Schema-validated metadata, assessment, claim mapping, references, and glossary data;
- TeX math authored with `$...$` and `$$...$$`, plus `\ce{...}` for chemical formulae and equations;
- schema-valid `*.diagram.json` for bounded processes and networks, plus sanitized SVG for scientific illustration;
- structured chemical source such as SMILES, InChI, Mol, or SDF when molecular identity or structure matters;
- a deterministic compiler target that emits a versioned JSON AST and semantic HTML for the custom lesson application;
- planned export adapters for PDF, EPUB, DOCX, and JATS, with Typst first for PDF and LuaLaTeX as the fallback.

Markdown and structured source are canonical. Generated HTML, MathML, SVG, PDF, EPUB, DOCX, JATS, thumbnails, and lesson-index data are build artifacts. Contributors do not hand-edit generated outputs.

**Implementation status.** The v1 source/schema validator, deterministic web compiler, and browser reader are implemented. The repository intentionally implements the documented MyST-compatible subset, not arbitrary MyST extensions. Typst/LuaLaTeX PDF and EPUB/DOCX/JATS adapters remain unavailable interoperability targets until deterministic checks for them land.

## 2. Production pack

One lesson pull request introduces or materially revises one directory:

```text
lessons/
  PREM-CHE-001-acid-base-models/
    lesson.json
    content/
      010-orientation.md
      020-diagnostic.md
      030-concept.md
      040-worked-example.md
      050-retrieval-check.md
      060-misconception.md
      070-practice.md
      080-transfer.md
      090-synthesis.md
      100-references.md
    assessment.json
    references.json
    claims.json
    glossary.json
    diagrams/
      process.diagram.json
    assets/
      images/
      molecules/
      data/
      transcripts/
    ATTRIBUTION.md
    reviews/
      academic-01.json
      learning-design.json
      accessibility-rights.json
    adjudication.json
```

This tree demonstrates available artifact and scene kinds; it is not a lesson template to copy line for line. A real lesson uses the smallest set of scenes that gives its outcome a coherent explanation, practice, assessment, and recovery path. Several instructional functions may share one scene, and unused scene kinds should be omitted.

The scene filenames under `content/` are descriptive and prefixed with sortable numbers. The authoritative order, identity, and kind live in the `scenes` array in `lesson.json`; directory order is only a human convenience. Use forward-slash relative paths contained within the pack. A pack may omit scene kinds that do not serve its outcomes, but it must satisfy the lesson must-haves in section 5.

`lesson.json` declares format/version, lesson identity/version/status/risk, graph outcomes and prerequisites, duration, authorship and agent provenance, ordered scenes, files, source confidence, CC BY 4.0 licensing, and third-party assets. Lesson Format v1 currently uses lesson schema v3. Follow the current public lesson schema exactly; do not add private conventions that other renderers cannot discover.

## 3. Semantic scenes

A scene is a bounded instructional purpose, not a slide and not an arbitrary page break. One scene should remain understandable when rendered alone, link its prerequisites or prior scene where needed, and fit an adaptive view without relying on a fixed screen size.

| Scene kind | Purpose | Minimum authoring expectation |
|---|---|---|
| `orientation` | Establish relevance, scope, outcomes, route, and boundaries | Start with a concrete reason to care and the progress learners will make; add exclusions or boundaries afterward only when they affect safe or correct use. |
| `diagnostic` | Surface prerequisite knowledge before instruction | Use ungraded questions with routing feedback; never treat placement as failure. |
| `concept` | Build an explanatory model | Connect representations, evidence, assumptions, and limits rather than listing facts. |
| `definition` | Establish precise language or notation | Give term, meaning, conditions, non-examples, and linked glossary ID. |
| `derivation` | Show how a result follows | Expose each justified step, assumptions, units, and interpretation. |
| `worked-example` | Model expert problem solving | Include problem representation, plan, execution, checks, and reflection—not answer-only arithmetic. |
| `investigation` | Ask or analyze an empirical question | State variables, controls, method/data, safety/ethics, uncertainty, and valid conclusion. |
| `retrieval-check` | Recall or discriminate before seeing feedback | Keep it low stakes; provide answer logic and schedule later retrieval. |
| `misconception` | Make a high-value wrong model visible and repairable | Elicit a prediction, provide disconfirming evidence, reconstruct the correct model, and test the repair. |
| `practice` | Build independent, varied performance | Fade hints, vary surface form, include feedback and a next-step route. |
| `transfer` | Apply learning in an unfamiliar bounded context | Require selection and adaptation of the model; do not merely change numbers. |
| `synthesis` | Connect scenes, outcomes, or domains | Produce a concept/causal map, comparative explanation, or cumulative model. |
| `assessment` | Collect declared mastery evidence | Identify assessed outcome, conditions, rubric/answer logic, and remediation route in `assessment.json`. |
| `references` | Expose sources, claim support, rights, and further reading | Render traceable records from `references.json` and `claims.json`; do not maintain an unlinked duplicate list. |

Scene order follows learning logic, not the table order. For example, a diagnostic may precede orientation in a short placement pack, investigations can motivate concepts, and retrieval checks should recur after spacing.

### Scene source rules

- Begin with one level-one scene title; use descending headings without skipped levels inside the file.
- Write semantic prose, lists, tables, links, code literals, TeX, and approved directives only.
- Keep outcome-relevant explanation in the source. CSS, generated images, hover states, and speaker notes may not contain the only copy of essential content.
- Use stable graph outcome IDs and glossary/reference/claim IDs in metadata and machine links. In learner prose, prefer descriptive names and ordinary transitions; do not expose internal identity machinery unless the outcome teaches it.
- Set `claimCoverage` on every scene. Use `claims-mapped` when it contains any material teaching claim and `no-material-claims` only for a scene whose content is genuinely instructional routing, reflection, or other non-evidentiary activity.
- In every `claims-mapped` scene, expose each mapped claim through a learner-visible `{source-note}` that lists all source IDs attached to its named claims.
- Do not encode absolute viewport dimensions, body scroll behavior, pagination, or desktop-only layout in a scene.
- Keep practice prompts separate from answer logic. A scene can point to assessment item IDs; it must not leak hidden answers into the learner payload.

## 4. Approved Markdown and directives

The base syntax is CommonMark/GFM: headings, paragraphs, emphasis, lists, block quotes, fenced code shown as inert text, tables, thematic breaks, and links. **Markdown image syntax is rejected** (see `FORMAT.md` sections 4 and 12); use the `{figure}` directive, which carries the required alt text and rights metadata. MyST-style colon fences add semantics without admitting raw HTML.

The v1 directive whitelist is:

| Directive | Use |
|---|---|
| `definition` | Term linked to `glossary.json` |
| `theorem` | Named proposition, conditions, and scope where theorem-style treatment is appropriate |
| `derivation` | Stepwise justification linked to declared assumptions and symbols |
| `worked-example` | Worked reasoning block inside a broader scene |
| `check` | Formative prompt linked to answer logic or feedback ID |
| `misconception` | Wrong-model elicitation, disconfirming evidence, repair, and recheck |
| `investigation` | Method, data, or inquiry block linked to its safety/ethics record |
| `figure` | Local image or generated visual plus caption, alt text, and optional long description |
| `diagram` | Schema-valid `*.diagram.json` source plus accessible equivalent |
| `equation` | Named display equation plus prose reading when the expression is not self-explanatory |
| `chemistry` | Chemical equation, mechanism, or structured molecule reference plus text alternative |
| `source-note` | Evidence scope, source IDs, uncertainty, or rights note linked to structured records |
| `callout` | Bounded note, tip, warning, objective, transcript, or other declared callout variant |

Example shape:

```md
:::{figure} ../assets/images/buffer-capacity.svg
:alt: Two titration curves with the buffer region highlighted.
:longdesc: The first titration curve stays comparatively flat through the marked buffer region, while the second changes steeply over the same added volume.
:label: fig-buffer-capacity

Buffer capacity is greatest where the curve changes least for an added amount of strong acid or base.
:::
```

The v1 directive-option whitelist is `id`, `label`, `kind`, `alt`, `longdesc`, `claims`, and `sources`. Options are data, not executable attributes. Use only combinations accepted by the format validator. Unknown directives or options, raw HTML, custom components, inline styles, event handlers, iframes, and embedded scripts are rejected.

## 5. Lesson must-haves

A standard lesson is content-complete before review only when it passes the learning-quality gates in [`CONTENT-STANDARD.md`](../CONTENT-STANDARD.md) and contains all applicable evidence below:

1. **Human first read:** a cold reader at the declared level can say what the lesson teaches, why it matters, and how its central idea works without consulting metadata, ledgers, or hidden project vocabulary.
2. **Identity and scope:** valid lesson/version IDs, selected atomic graph outcomes, prerequisites, level, estimated time, material exclusions, and route status.
3. **Observable objectives:** each objective maps to a graph outcome and at least one assessment item.
4. **Prerequisite route:** a diagnostic or explicit verification path, plus recovery links for missing dependencies.
5. **Explanatory model:** a coherent causal, relational, or procedural argument at the declared depth, with stable terminology, explicit inference links, consistent representations, assumptions, evidence, and limits.
6. **Worked reasoning:** at least one worked example or derivation that exposes task representation, model selection, plan, reasons, execution, verification, and interpretation, followed by a route that fades support.
7. **Retrieval:** a prompt presented before answer reveal, with reasoning-rich feedback and delayed/cumulative retrieval mapping.
8. **Misconception repair:** at least one evidence-backed misconception, non-example, or boundary case chosen for the mapped outcomes.
9. **Practice:** varied independent practice with fading support, feedback that diagnoses reasoning and prescribes a next step, retry, and actionable remediation.
10. **Transfer:** a novel bounded task requiring model selection or cross-context reasoning.
11. **Assessment:** item-to-outcome map, conditions, answer logic, misconception-sensitive feedback, analytic rubric where appropriate, equivalent retry form/strategy, and remediation route.
12. **Claims and sources:** explicit scene coverage; material factual claims mapped in `claims.json` to complete reference records; a same-scene learner-visible `{source-note}` with every supporting source ID; exact source scope, confidence/uncertainty, review state, and retrieval/version data.
13. **Glossary:** stable definitions for introduced technical terms, symbols, abbreviations, and ambiguous everyday words.
14. **Accessibility:** meaningful structure; equivalents for visuals, audio, interactions, and practical modalities; no color-, sound-, motion-, hover-, pointer-, or time-only information.
15. **Rights and provenance:** CC BY 4.0 contributor grant, complete third-party asset records, attribution, modification notices, and accountable agent disclosure.
16. **Boundaries and safety:** no clinical advice or credential claim; investigation risks, privacy, consent, and non-physical alternatives addressed when relevant.
17. **Export parity:** essential content and relationships survive semantic HTML, JSON AST, PDF, EPUB, DOCX, and JATS transforms.

A references-only reading list, a narrated deck, a question bank without instruction, or a long essay divided into screens is not a lesson pack.

## 6. Mathematics

- Author inline TeX as `$...$` and display TeX as `$$...$$`.
- Use semantic commands and conventional notation; declare symbols before use and preserve units.
- Keep TeX to the supported math subset. File inclusion, shell escape, macro redefinition with side effects, raw backend commands, and package loading are forbidden.
- Put long derivations in a `derivation` scene and expose intermediate logic. Do not replace reasoning with a screenshot of equations.
- Give important displayed equations a label and a plain-language interpretation. Complex expressions need a prose reading or structured alternative sufficient for nonvisual navigation.
- The compiler target is accessible MathML with an HTML fallback. Until that path is implemented and validated, the specimen must not imply that mathematical accessibility is complete. TeX source remains available for copying and future export.
- Verify line breaking and numbering in the SPA and PDF paths; no result may depend on visual alignment alone.

## 7. Chemistry

- Use `\ce{...}` for formulae and chemical equations within the supported TeX chemistry subset.
- Balance reactions, show charge and physical state where material, define conditions, and distinguish equilibrium arrows from one-way transformations.
- A molecular diagram that carries identity must have structured source. Use canonical/isomeric SMILES as appropriate, InChI for identity, and Mol/SDF when coordinates, atom mapping, stereochemistry, or a reusable structure record matter.
- Commit structured molecular source and a generated SVG depiction. The SVG is a view, not the sole record.
- State stereochemistry explicitly and check that text, structured source, and drawing agree.
- Reaction mechanisms need ordered steps, electron-flow meaning in text, reagents/conditions, and an accessible long description. A raster arrow diagram alone is insufficient.
- Do not make clinical efficacy, toxicity, handling, or treatment claims from chemical plausibility alone.

## 8. Diagrams, charts, images, audio, and video

Choose the simplest inspectable source that fits the relationship:

| Need | Canonical source | Required equivalent |
|---|---|---|
| Simple sequence, state, process, or bounded network | `*.diagram.json` | ordered relationship list or long description |
| Quantitative chart | Local CSV/JSON data plus a reviewed static SVG in v1 | accessible data table and interpretation |
| Scientific illustration/anatomy | sanitized SVG, with editable source when available | concise alt text plus linked long description |
| Photograph/micrograph | local lossless/high-quality raster plus provenance | alt text, scale/orientation/stain/context where relevant |
| Animation | declarative frames/data | transcript or stepwise static equivalent; pause/restart controls |
| Audio/video | local or approved hosted media with provenance | synchronized captions and complete transcript |

Visual source and input data are committed with the pack. Generated SVGs must contain no script, event handler, foreign object, remote resource, or external font dependency. Text in diagrams must remain legible at 200% zoom and in reflow. Color is never the only encoding; palettes must retain contrast in common color-vision deficiencies. Decorative images use empty alt text and are not cited as evidence.

The author supplies the intended meaning, not a pixel inventory. Alt text identifies the visual and its purpose; a long description explains relationships, data, trends, labels, and uncertainty needed to achieve the same outcome. Charts also ship the underlying accessible table.

## 9. Claims, references, and glossary

`references.json` schema v2 holds stable bibliographic records plus each use type, rights basis and evidence, and dated agent-access state. `claims.json` links material claims to one or more reference IDs and records scene IDs, support type, confidence, scope, uncertainty, risk, and review state. Exact version, page, section, table, or figure location belongs in the source record's locator. `glossary.json` gives stable term IDs, labels, definitions, aliases, and linked scene IDs; symbols and abbreviations are recorded as term entries when they need definitions.

A material claim is a factual, quantitative, definitional, causal, mechanistic, safety, health, standards, or evidence-dependent statement that the learner is expected to accept or use. Objectives, interface instructions, reflection prompts, and author-created practice directions are usually not material claims unless they also teach such a statement. Repeating or applying a material claim in another scene still requires that scene to map and visibly cite it. A lesson-wide bibliography does not satisfy this contract.

Source expectations:

- prefer primary standards, official specifications, peer-reviewed research, and versioned scholarly/open textbooks for the claim they can actually support;
- distinguish a source used as evidence from suggested further reading;
- cite mutable sources with retrieval date and exact version/date; include DOI or durable identifier where available;
- preserve jurisdiction, population, model, and uncertainty limits;
- never cite an AI response as authority or invent a bibliographic record from a plausible title;
- independently synthesize facts in original language, organization, examples, questions, and visuals; do not closely paraphrase or lightly redraw a source;
- reuse expression, media, or datasets only under a permitted basis in `RIGHTS-POLICY.md`, and record it separately from academic citation;
- check source terms before agent access and never let an agent process a source recorded as `human-only`;
- check that every cited URL/identifier resolves and that the source supports the attached claim before review.
- keep authoring claims at `pending-review` in the frozen candidate; after exact-candidate quorum and merge adjudication, promote only retained claims' `reviewStatus` fields and permitted lesson status/source-confidence fields. Publication requires every retained claim to be `reviewed` and lesson source confidence to be resolved.

References are rendered from structured records. A hand-maintained prose bibliography may add context but cannot replace `references.json` and `claims.json`.

## 10. Accessibility and adaptive presentation

The source order is the reading and focus order. Scenes must work with keyboard-only navigation, screen readers, zoom/reflow, reduced motion, and touch without changing the assessed outcome.

The lesson application uses a no-body-scroll desktop shell so navigation and progress remain stable while the active scene owns bounded overflow. That is a renderer behavior, not an authoring assumption. On narrow screens, zoomed views, print, and assistive layouts, scenes reflow into ordinary document flow. Authors must not use fixed-height content, desktop coordinates, hover-only disclosure, drag-only input, or wording such as “the diagram on the right” as the sole locator.

Every interaction needs a visible label, keyboard operation, persistent or repeatable instructions, programmatic status/feedback, and a non-timed route unless timing is the declared construct. Captions, transcripts, long descriptions, tables, and simulation/dataset alternatives must be outcome-equivalent. Accommodation changes the medium or route, not the scientific standard; the record must not claim a physical skill when only a simulation was completed.

## 11. Build and export contract

The normative compiler parses trusted, validated source into a versioned JSON AST. Renderers are designed to consume that AST to produce:

- sanitized semantic HTML and accessible MathML for the custom SPA;
- deterministic visual assets derived from validated diagram JSON, local data, and structured chemistry sources;
- PDF through Typst first, with LuaLaTeX as the compatibility fallback;
- EPUB for reflowable reading;
- DOCX for institutional editing/review workflows;
- JATS for archival and scholarly interchange.

Export adapters may simplify an interaction, but they must preserve the prompt, data, answer conditions, feedback location, semantic relationships, citations, glossary links, visual equivalents, and source attribution. Unsupported output is a build failure or an explicit, reviewed degradation notice—not silent omission.

At v1 adoption, the source/schema contract, deterministic web compiler, and browser reader are implemented. The non-web adapters above are planned and unavailable until repository checks demonstrate otherwise.

The current browser specimen uses the safe `diagrams/*.diagram.json` schema for a small process diagram. Mermaid, DOT, and Vega-Lite may be added later as generated adapters, but they are not accepted as canonical lesson inputs until bounded validators and deterministic renderers exist.

## 12. Security and privacy

Lesson content is data and is never executable.

- Raw HTML, JavaScript, CSS, web components, iframes, event handlers, executable notebooks, macros, plugins, and remote embeds are forbidden in lesson source.
- Code fences are inert instructional text. Repository automation must never run code from an untrusted lesson pull request.
- Diagram JSON, TeX, chemistry, local data, and SVG inputs are parsed with pinned toolchains, strict feature allowlists, resource/time limits, no shell escape, no arbitrary file reads, and no network access.
- Builds resolve only normalized paths inside the lesson pack. Path traversal, symlink escape, data URLs, and unsafe URL schemes are rejected.
- External links are sanitized and rendered with safe opener/referrer behavior. Third-party assets are downloaded only through a separate, reviewed provenance workflow—not by the lesson renderer.
- SVGs are sanitized; raster metadata is stripped when it can expose location/device data; secrets and contributor-private data never enter assets or scenes.
- Personal medical information, identifiable learner data, credentials, API keys, copyrighted answer banks, and unlicensed patient media are prohibited.
- Assessments delivered to learners separate hidden answer logic from public prompts and do not trust the client for authoritative scoring or credential claims.

Validation and rendering do not establish factual correctness. Standard lessons require academic and learning-design review inputs plus a separate fresh finalizer who performs the final accessibility-and-rights audit; other risk tiers follow their declared role quorum.

## 13. Specimen versus contribution

`examples/lesson-pack/` is the gold-standard **format specimen**. It may demonstrate every scene kind, source adapter, accessibility alternative, claim map, and export fixture. It is deliberately excluded from:

- the `lessons/` production namespace;
- open pull-request and merged-lesson indexes;
- review or adjudication counts;
- covered-outcome and progress totals;
- claims that Premed has a published lesson.

A contributed lesson lives under `lessons/` or in an open pull request proposing that path, maps to real graph outcomes, and follows the governance protocol. It becomes `published open` only after validation, independent quorum, separate adjudication, merge, public availability, and confirmed CC BY 4.0 status. Copying the specimen into `lessons/` does not confer review or publication state.

## 14. Author preflight

Before requesting review:

- [ ] Read the current schema and quorum policy; build against their version.
- [ ] Complete the learner/outcome authoring brief and every applicable acceptance gate in `CONTENT-STANDARD.md`.
- [ ] Prototype the opening, central explanation, and one worked use; pass the first-read gate before completing the pack.
- [ ] Confirm every outcome and prerequisite ID exists in the current graph.
- [ ] Validate scene order, kinds, links, directives, math, chemistry, structured visuals, and pack-local paths.
- [ ] Check all lesson must-haves and item-to-outcome mappings.
- [ ] Resolve every claim ID to a supporting reference and every glossary use to a term record.
- [ ] Run accessibility checks on semantic HTML and export fixtures, including keyboard, screen reader, reflow, contrast, captions, MathML, and long descriptions.
- [ ] Confirm rights and attribution for every asset and dataset.
- [ ] Confirm no executable content, raw HTML, unsafe SVG, remote embed, secret, personal data, or unapproved experiment appears.
- [ ] Disclose accountable principals and material agent provenance.
- [ ] Run `npm ci`, `npm run source:preflight -- --strict lessons/<lesson-pack>`, and `npm run verify` sequentially.
- [ ] Freeze one candidate commit before collecting review artifacts.

The review record determines whether the pack is fit to merge. The specimen and automated checks are necessary implementation references, not substitutes for independent judgment.


---

## Repository file: course/PREMED-SYLLABUS.md

# Premed: Foundations for Medical Study

**Curriculum version:** 1.1 candidate · 20 July 2026<br>
**Provider:** EmbeddedKnowledge<br>
**Language:** English<br>
**Level:** university preparatory; introductory undergraduate science with a medicine-facing bridge<br>
**Format:** mastery-based, online-first, asynchronous with optional live/tutorial and supervised practical components<br>
**Nominal workload:** 1,440 hours for the common core and medicine bridge, plus 120–240 hours for one pathway<br>
**Typical pace:** 72 weeks at 22–24 hours/week; 96 weeks at 17–18 hours/week part-time; no fixed expiry<br>
**Credential:** an EmbeddedKnowledge record of demonstrated outcomes, **not** a university degree, academic credit, professional license, medical-school admission guarantee, or substitute for institution-specific prerequisites

> **Course promise:** finishers should be able to explain, model, investigate, and integrate the science that a well-prepared entrant commonly brings to medical study. The target is demonstrated readiness, not merely exposure and not an admissions credential.

## 1. Place in the curriculum

Premed is EmbeddedKnowledge's first course and the prerequisite for later medical-science courses. It deliberately combines the portable scientific spine found across major preparation systems with a clearly labelled medicine-facing bridge and route-specific pathways.

It does **not** claim that every medical entrant in every country has taken all of these subjects. Admissions requirements are decentralized in the United States, institution-specific in the United Kingdom, and partly determined by annual university weightings in Spain. AP, IB, A-level, Bachillerato, Abitur, and Baccalaureat subject lists are choices or curriculum unions, not a single global entrant minimum. A learner applying to a real institution must check that institution's current rules.

The course is broader than the MCAT where medical readiness benefits from anatomy, physiology, immunity, laboratory practice, and scientific communication. It is narrower than a biology degree: ecology, plant biology, calculus-based physics, and advanced organic synthesis are pathway options rather than universal requirements.

## 2. Audience

Premed is designed for:

- curious adults who want genuine premedical knowledge without seeking a title;
- prospective medical applicants who need a structured foundation or route-specific review;
- nurses, EMTs, biomedical engineers, health-technology workers, and other health-adjacent learners filling gaps;
- access/foundation learners rebuilding mathematics, chemistry, biology, or academic-English confidence.

No prior degree is required. The normal entry floor is upper-secondary reading and algebra. Prior science courses may shorten the route through diagnostic placement; they never exempt a learner from demonstrating the mapped outcomes.

## 3. Entry, diagnostic placement, and on-ramp

### 3.1 Expected entry capabilities

Before the common core, learners should be able to:

- read a 1,000-word expository text and state its claim and supporting evidence;
- rearrange a one-variable equation, use fractions, ratios, percentages, and powers of ten;
- interpret axes, slope, and a simple data table;
- use a calculator and basic spreadsheet cells without copying sensitive data to a public service;
- write a short explanation in their own words and distinguish a quotation from a paraphrase;
- follow basic digital-safety and laboratory-safety instructions.

### 3.2 Diagnostic

The untimed entry diagnostic has five 30–45 minute stations: `DIA-MATH`, `DIA-READ`, `DIA-CHEM`, `DIA-BIO`, and `DIA-DATA`. It samples prerequisites only; it is not graded. Each station produces a concept-level placement report and one of three actions:

| Result | Action |
|---|---|
| Secure (at least 85%, no critical misconception) | Begin the linked core node; on-ramp practice remains available. |
| Developing (60–84%) | Complete only the prescribed on-ramp lessons and retry the station. |
| Foundational (below 60% or a critical safety/numeracy error) | Complete the full on-ramp below with tutor check-ins. |

### 3.3 On-ramp `ONR-000` (0–120 hours, as needed)

| Topic ID | Scope | Exit evidence |
|---|---|---|
| `ONR-001` Scientific language | variables, units, prefixes, notation, estimation, significant figures | annotate and calculate from a short scientific passage |
| `ONR-002` Algebra | equations, inequalities, ratios, proportionality, exponents, scientific notation | solve a mixed, contextual problem set with units |
| `ONR-003` Graphs and functions | linear, inverse, exponential, logarithmic relationships; interpolation | construct and interpret three graph types |
| `ONR-004` Chemical language | element symbols, formulas, ions, moles as counting units | translate words, formulas, and particle diagrams |
| `ONR-005` Cell and gene primer | cells, chromosomes, genes, proteins, energy as orienting models | produce a correct concept map without memorized detail |
| `ONR-006` Evidence and reading | claims, evidence, assumptions, correlation, causation, source types | critique a short popular-health claim |
| `ONR-007` Study system | retrieval practice, spacing, interleaving, error logs, planning | run a two-week study cycle and reflect on its data |
| `ONR-008` Academic English | paragraph logic, disciplinary vocabulary, paraphrase, discussion norms | submit a cited 500-word explanation |

Learners may use another language to reason or annotate, but assessed scientific terms and final explanations are in English. Accommodations are described in section 16.

## 4. Curriculum philosophy

1. **A knowledge graph, not a content pile.** Every topic has a stable ID, prerequisites, observable outcomes, evidence, and downstream uses.
2. **Reasoning is the curriculum.** Recall supports modelling, experimental design, data analysis, and argument. It never substitutes for them.
3. **Mastery is revisable.** Gates are untimed and retakeable after targeted practice. A failed attempt is diagnostic evidence, not a verdict.
4. **Medicine supplies context, not false authority.** Cases make basic science consequential, but learners do not diagnose, prescribe, or simulate clinical competence.
5. **Representations must agree.** Learners move among words, equations, graphs, mechanisms, diagrams, and physical models.
6. **Practical evidence matters.** Simulations can teach design and interpretation; they cannot prove hands-on dexterity. The record names which modality was actually completed.
7. **Sources outrank confidence.** Claims are traceable. Uncertainty, limitations, and correction are normal parts of scholarship.
8. **Motivation is infrastructure.** Visible node-level progress, short feedback loops, learner choice, recovery weeks, and meaningful cases are core design requirements.

## 5. Program-level learning outcomes

By completion, a learner can:

### Knowledge and explanation

1. `PLO-01` Explain how atomic and molecular interactions give rise to biological structure, reactivity, energy transfer, and measurement.
2. `PLO-02` Explain cells as regulated, evolving systems that exchange matter, energy, and information.
3. `PLO-03` Trace genetic information from inheritance through gene expression and connect variation to phenotype.
4. `PLO-04` Model enzyme function, bioenergetics, and the integration and regulation of major metabolic pathways.
5. `PLO-05` Apply introductory organic chemistry to biologically important functional groups, stereochemistry, mechanisms, and analytical evidence.
6. `PLO-06` Apply algebra-based mechanics, fluids, thermal physics, electricity, waves, optics, and atomic/nuclear physics to biological and medical contexts.
7. `PLO-07` Explain homeostasis through integrated organ-system structure, transport, control, and feedback.
8. `PLO-08` Explain host defence, microbial and viral strategies, biotechnology, and the limits and ethics of intervention.

### Inquiry and quantitative reasoning

9. `PLO-09` Form a testable question, identify variables and controls, select measurements, anticipate confounding, and distinguish exploratory from confirmatory work.
10. `PLO-10` Calculate with units, uncertainty, ratios, logarithms, probability, descriptive statistics, confidence intervals, and simple inferential tests at the course's stated level.
11. `PLO-11` Create and interpret tables, graphs, images, spectra, and simple models without overstating what the data establish.
12. `PLO-12` Evaluate validity, reliability, bias, effect size, statistical versus practical significance, and alternative explanations.
13. `PLO-13` Keep a reproducible record, handle data ethically, and communicate methods sufficiently for another learner to audit the work.

### Integration, communication, and self-regulation

14. `PLO-14` Integrate chemistry, biology, physics, and behaviour in unfamiliar medicine-facing problems while distinguishing mechanism from association.
15. `PLO-15` Read a primary research article at an introductory level and reconstruct its question, method, principal result, limitations, and warranted conclusion.
16. `PLO-16` Explain a scientific idea accurately to expert and public audiences, cite sources, disclose tool use, and revise after critique.
17. `PLO-17` Identify the boundary between education and medical advice and recognize when a claim requires a qualified professional.
18. `PLO-18` Use retrieval data, confidence calibration, an error taxonomy, and a study plan to remediate weak nodes independently.

Route-specific outcomes (`PLO-US`, `PLO-UK`, `PLO-ES`, `PLO-BIO`, and `PLO-QNT`) are defined in section 12.

## 6. Architecture and completion rules

### 6.1 Components

| Component | Modules | Nominal hours | Required for EmbeddedKnowledge Premed completion? |
|---|---:|---:|---|
| Course welcome | `WEL-000` | 1 | Yes |
| On-ramp | `ONR-000` | 0–120 | Only when prescribed by diagnostic |
| Inquiry and quantitative spine | `QRS-100`, embedded studios | 180 | Yes |
| General chemistry | `CHE-110`–`CHE-140` | 220 | Yes |
| Biology and genetics | `BIO-110`–`BIO-150` | 250 | Yes |
| Physics | `PHY-110`–`PHY-150` | 210 | Yes |
| Organic chemistry | `ORG-210`–`ORG-230` | 150 | Yes |
| Biochemistry | `BCH-210`–`BCH-240` | 200 | Yes |
| Medicine-facing bridge | `MED-310`–`MED-330` | 170 | Yes, but labelled as EmbeddedKnowledge design scope rather than universal entrant minimum |
| Integration and capstone | `INT-300`, `CAP-400` | 60 | Yes |
| One route/pathway | section 12 | 120–240 | Yes; learner may complete more than one |

Hours overlap because inquiry studios, practical work, retrieval, and cases are embedded in disciplinary modules. The common-core plus bridge total is capped at **1,440 nominal hours**, not the sum of every possible activity in every row.

### 6.2 Completion standard

Completion requires all of the following:

- every required outcome at `M` (mastered) or above in the mastery ledger;
- all safety-critical outcomes at `M` with no compensation by average;
- twelve practical investigations, including at least four using physical equipment if the learner wants a **blended-practical** notation;
- the research-reading portfolio and quantitative portfolio;
- one completed route/pathway (see the instrumentation limitation in section 12: in curriculum version 1.1 only `PW-US`, `PW-BIO` and `PW-QNT` can satisfy this requirement, because `PW-UK` and `PW-ES` are not yet represented in the knowledge graph);
- an accepted integrative capstone and oral defence;
- a final calibration review showing that confidence judgments track actual performance.

No time limit applies. Institution-specific applications may require formally accredited laboratory courses regardless of this record.

## 7. Stable mastery language

Each outcome is stored independently; module averages cannot hide a gap.

| State | Meaning | Minimum evidence |
|---|---|---|
| `N` Not yet encountered | No interpretable evidence | none |
| `A` Acquiring | Can follow a worked example; recall is fragile | guided checks |
| `P` Practising | Can solve familiar problems with limited prompts | two successful practice occasions |
| `M` Mastered | Can explain and apply independently in a new but bounded context | gate at ≥80%, all critical items correct, plus transfer task |
| `R` Retained | Re-demonstrated after spacing in a different context | two delayed checks, one at least 21 days later |
| `T` Transfer | Uses the outcome appropriately in a cross-domain case or investigation | rubric-scored authentic task |

Confidence is logged before feedback on selected tasks. A high-confidence incorrect response triggers misconception remediation even if the surrounding module score is adequate.

## 8. Module sequence at a glance

Modules are shown in the recommended order. `||` indicates modules that can run concurrently after their shared prerequisites.

```text
WEL-000
  -> QRS-100 (begins with Scientific Inquiry)
  -> CHE-110 || BIO-110
  -> CHE-120 || BIO-120 || PHY-110
  -> CHE-130 || BIO-130 || PHY-120
  -> CHE-140 || BIO-140 || PHY-130
  -> ORG-210 || BIO-150 || PHY-140
  -> ORG-220 || BCH-210 || PHY-150
  -> ORG-230 || BCH-220
  -> BCH-230 -> BCH-240
  -> MED-310 || MED-320 -> MED-330
  -> INT-300 + one PATHWAY
  -> CAP-400
```

`WEL-000` is the only lesson before the scientific curriculum. It previews the complete journey and the capabilities learners are working toward. `QRS-100` then opens with Scientific Inquiry; its practices continue through every later module. Diagnostic-prescribed `ONR-000` refreshers are optional support, not a separate opening content block. The curriculum map in `PREMED-CURRICULUM-MAP.md` gives the cross-domain edges and misconception gates.

## 9. Detailed common-core modules

### `WEL-000` Welcome to Premed — 1 hour

**Purpose.** Welcome learners into the scientific journey ahead before asking them to master any new content. The lesson previews the questions each subject helps answer, shows how the subjects connect, and makes the destination concrete: by the end, learners should be able to explain, model, investigate, and integrate the science that supports later medical study.

The welcome introduces the course in plain language:

- **Scientific inquiry and laboratory practice:** how we turn a question into a fair test, decide what evidence can support, and make work reproducible.
- **Mathematics, statistics, chemistry, and physics:** how we measure the world and explain matter, energy, forces, fluids, electricity, waves, and uncertainty.
- **Biology, genetics, organic chemistry, and biochemistry:** how cells are built, use energy, store information, respond, reproduce, and change.
- **Human systems, immunity, microbiology, behaviour, and society:** how those foundations connect across a whole person and population without pretending that premedical study is clinical practice.
- **Integration and capstones:** how learners will combine the subjects to reason through unfamiliar problems rather than merely repeat isolated facts.

**Completion evidence.** The learner can describe the subject journey, explain why its major concepts matter for medical study, and identify what they should know and be able to do by course completion. This is a motivating orientation, not a prior-knowledge test or a graded science gate.

### `QRS-100` Quantitative and scientific reasoning — 180 hours distributed

**Purpose.** Begin with Scientific Inquiry—questions, hypotheses, variables, evidence, causal limits, uncertainty, and reproducibility—then establish the mathematical and statistical practices used everywhere else.

Instruction begins with `QRS-105`, `QRS-110`, and `QRS-112`; the numeric IDs below are stable references, not a required teaching order. Laboratory and measurement work then make the inquiry cycle concrete before the remaining quantitative sequence expands what learners can test and interpret.

| Topic ID | Topic-level scope and measurable outcome |
|---|---|
| `QRS-101` | **Dimensions and estimation:** convert units, check dimensional consistency, propagate powers of ten, estimate order of magnitude, and explain whether a result is physically plausible. |
| `QRS-102` | **Algebraic models:** solve and rearrange linear, inverse, proportional, power, exponential, and logarithmic relationships; state domain assumptions. |
| `QRS-103` | **Functions and graphs:** derive slope and area meanings in context; recognize linearized relationships; choose axes and scales that do not mislead. |
| `QRS-104` | **Measurement:** distinguish accuracy, precision, resolution, random error, systematic error, calibration, and uncertainty; report significant figures appropriately. |
| `QRS-105` | **Study design:** distinguish observational, experimental, cross-sectional, case-control, cohort, and randomized designs at an introductory level; identify exposure, outcome, controls, blinding, randomization, and confounding. |
| `QRS-106` | **Descriptive statistics:** compute and interpret mean, median, variance, standard deviation, percentile, distribution shape, outlier, and standardization. |
| `QRS-107` | **Probability:** use complements, conditional probability, independence, counting, Bayes reasoning with natural frequencies, sensitivity, specificity, and predictive values. |
| `QRS-108` | **Inference:** interpret sampling variability, confidence intervals, null hypotheses, p-values, Type I/II errors, power, effect sizes, correlation, simple linear regression, chi-square, and a two-group comparison without treating a p-value as truth. Calculation emphasis is limited to simple hand/spreadsheet cases. |
| `QRS-109` | **Data representation:** construct and critique tables, scatterplots, line/bar plots, histograms, box plots, uncertainty bars, log plots, and image quantification. |
| `QRS-110` | **Causal and model reasoning:** distinguish association from causation, mechanism from prediction, model from reality, and interpolation from extrapolation; compare plausible alternatives. |
| `QRS-111` | **Evidence appraisal:** reconstruct a paper's question, population, design, result, limitation, funding/conflict disclosure, and warranted claim. |
| `QRS-112` | **Reproducible practice:** maintain a dated notebook, data dictionary, formulas or scripts, provenance, change log, and an auditable figure. |

**Activities and evidence.** Weekly mixed quantitative sets; misleading-graph repair; diagnostic-test natural-frequency simulation; paper figure reconstruction; preregistration of one miniature investigation; reproducibility audit by a peer or tutor.

**Mastery gate.** An unseen multi-figure study requiring calculations, design critique, uncertainty analysis, and a 700-word evidence judgment. Safety-critical items are unit conversion, denominator choice, and unsupported causal claims.

**Deliberate exclusions.** Calculus, matrix algebra, ANOVA derivations, multivariable regression, survival analysis, and programming proficiency are not in the common core. They appear only in `PW-QNT` or later medical research courses.

---

### `CHE-110` Matter, measurement, atoms, and periodicity — 50 hours

**Prerequisites:** `QRS-101`–`QRS-104`.

| Topic ID | Scope and outcome |
|---|---|
| `CHE-111` | Classify matter, physical/chemical properties, and changes; select separation methods from particle-level reasoning. |
| `CHE-112` | Use SI quantities, density, temperature scales, uncertainty, and significant figures in laboratory calculations. |
| `CHE-113` | Relate protons, neutrons, electrons, isotopes, ions, atomic mass, and mole-scale measurements. |
| `CHE-114` | Use electromagnetic radiation and quantized-energy evidence to describe atomic orbitals and electron configurations at introductory depth. |
| `CHE-115` | Predict periodic trends in radius, ionization energy, electron affinity, and electronegativity and justify exceptions cautiously. |
| `CHE-116` | Name and formulate common ionic and molecular compounds; move among symbolic, macroscopic, and particulate representations. |

**Practice/practical.** Density and calibration investigation; isotope-abundance problem set; flame/emission-spectrum simulation or supervised demonstration.

**Gate.** Identify an unknown material from measurement and spectroscopic evidence and defend the uncertainty budget.

**Exclusions.** Full quantum-mechanical derivations and many-electron wavefunctions.

### `BIO-110` Chemistry of life and cellular organization — 55 hours

**Prerequisites:** `CHE-111`–`CHE-116` may be co-studied.

| Topic ID | Scope and outcome |
|---|---|
| `BIO-111` | Relate water polarity, hydrogen bonding, ions, hydrophobic effects, pH, and buffering to biological environments. |
| `BIO-112` | Compare carbohydrates, lipids, proteins, and nucleic acids by monomer, bond, structure, function, and chemical behavior. |
| `BIO-113` | Explain prokaryotic and eukaryotic cell organization, organelles, surface-area constraints, and endosymbiotic evidence. |
| `BIO-114` | Explain membrane structure, fluidity, diffusion, osmosis, channels, carriers, pumps, and electrochemical gradients. |
| `BIO-115` | Predict effects of extracellular tonicity and transport disruption from concentration and permeability data. |
| `BIO-116` | Explain cytoskeleton, extracellular matrix, junctions, adhesion, and cell motility as dynamic systems. |
| `BIO-117` | Relate microscopy choice, scale bars, magnification, resolution, staining, and artifacts to a biological claim. |

**Practice/practical.** Membrane permeability investigation; microscopy/image-scale lab; osmotic model; biomolecule evidence stations.

**Gate.** Explain an unfamiliar transport disorder from molecular, cellular, and quantitative evidence.

**Exclusions.** Named rare storage diseases and exhaustive organelle protein inventories.

### `CHE-120` Bonding, reactions, and stoichiometry — 55 hours

**Prerequisites:** `CHE-110`, `QRS-102`.

| Topic ID | Scope and outcome |
|---|---|
| `CHE-121` | Draw and evaluate Lewis structures, formal charge, resonance, VSEPR geometry, polarity, and hybridization as useful models. |
| `CHE-122` | Relate ionic, covalent, metallic bonding and intermolecular forces to structure and measurable properties. |
| `CHE-123` | Balance equations and classify precipitation, acid-base, gas-forming, redox, and combustion reactions. |
| `CHE-124` | Convert among mass, moles, particles, concentration, gas volume, and equivalents with correct units. |
| `CHE-125` | Solve limiting-reagent, theoretical/percent-yield, empirical/molecular-formula, and mixture problems. |
| `CHE-126` | Write complete and net ionic equations and predict solubility using supplied rules and particle reasoning. |

**Practice/practical.** Gravimetric or microscale precipitation investigation; solution preparation; reaction-yield audit.

**Gate.** Plan, calculate, execute or simulate, and audit a microscale synthesis with limiting reagent and uncertainty.

**Exclusions.** Industrial process engineering and advanced crystallography.

### `BIO-120` Cell energy, enzymes, signalling, and division — 55 hours

**Prerequisites:** `BIO-110`, `CHE-120` co-study recommended.

| Topic ID | Scope and outcome |
|---|---|
| `BIO-121` | Distinguish energy, free-energy change, activation energy, equilibrium, and coupling in cellular terms. |
| `BIO-122` | Explain enzyme active sites, specificity, saturation, environmental effects, inhibition, allostery, and feedback regulation using rate data. |
| `BIO-123` | Trace cellular respiration conceptually through glycolysis, pyruvate oxidation, citric-acid cycle, electron transport, chemiosmosis, and fermentation. Detailed integration follows in biochemistry. |
| `BIO-124` | Compare receptor classes and trace a signal through transduction, amplification, response, termination, and cross-talk. |
| `BIO-125` | Explain cell-cycle control, mitosis, checkpoints, apoptosis, stem-cell renewal, and how loss of regulation contributes to cancer. |
| `BIO-126` | Compare meiosis and mitosis and connect chromosome behavior to variation and aneuploidy. |

**Practice/practical.** Enzyme kinetics investigation; mitotic-index image analysis; signalling-pathway perturbation model.

**Gate.** Interpret enzyme and cell-cycle experiments, then propose a discriminating follow-up.

**Exclusions.** Memorization of every signalling intermediate and clinical oncology management.

### `PHY-110` Motion, forces, energy, and momentum — 55 hours

**Prerequisites:** `QRS-101`–`QRS-103`.

| Topic ID | Scope and outcome |
|---|---|
| `PHY-111` | Represent vectors and one-/two-dimensional motion using diagrams, equations, and position/velocity/acceleration graphs. |
| `PHY-112` | Construct free-body diagrams and apply Newton's laws, friction, tension, normal force, gravity, and circular-motion relations. |
| `PHY-113` | Apply work, kinetic/potential energy, power, conservation, and efficiency to bounded systems. |
| `PHY-114` | Apply impulse and linear-momentum conservation; distinguish elastic and inelastic collisions. |
| `PHY-115` | Use torque, centre of mass, equilibrium, and mechanical advantage in simple biomechanical models. |
| `PHY-116` | Evaluate assumptions in idealized movement and injury models rather than mistaking the model for clinical prediction. |

**Practice/practical.** Video motion analysis; force/lever model of a limb; conservation simulation; uncertainty in derived acceleration.

**Gate.** Build and validate an algebra-based mechanical model from an unfamiliar movement dataset.

**Exclusions.** Calculus derivations, rigid-body tensors, continuum mechanics, and diagnosis of injury.

### `CHE-130` States, gases, solutions, and thermochemistry — 55 hours

**Prerequisites:** `CHE-120`, `PHY-110` may be concurrent.

| Topic ID | Scope and outcome |
|---|---|
| `CHE-131` | Use kinetic-molecular theory and phase diagrams to explain states, phase change, vapour pressure, and deviations from idealization. |
| `CHE-132` | Apply gas laws, partial pressures, mole fractions, and gas stoichiometry to laboratory and respiratory contexts. |
| `CHE-133` | Prepare and dilute solutions; calculate molarity, molality, mass fraction, and selected colligative effects. |
| `CHE-134` | Explain dissolution using intermolecular interactions and energy/entropy tradeoffs; predict qualitative solubility. |
| `CHE-135` | Apply system/surroundings, heat, work, calorimetry, enthalpy, Hess's law, and bond-energy estimates. |
| `CHE-136` | Distinguish enthalpy, entropy, and Gibbs free energy and connect spontaneity to conditions without implying rate. |

**Practice/practical.** Calorimetry; dilution and standard-curve preparation; gas-law simulation with residual analysis.

**Gate.** Quantitatively explain a coupled gas/solution/thermal process and evaluate model failure.

**Exclusions.** Real-gas equations beyond qualitative deviations and statistical thermodynamics.

### `BIO-130` Heredity, chromosomes, and populations — 45 hours

**Prerequisites:** `BIO-120`, probability portions of `QRS-107`.

| Topic ID | Scope and outcome |
|---|---|
| `BIO-131` | Use Mendelian models, pedigrees, test crosses, and probability while recognizing model assumptions. |
| `BIO-132` | Analyze incomplete dominance, codominance, multiple alleles, penetrance, expressivity, linkage, recombination, and sex linkage. |
| `BIO-133` | Explain chromosomal segregation, nondisjunction, structural variation, genomic imprinting at survey depth, and karyotype evidence. |
| `BIO-134` | Apply Hardy-Weinberg as a null model and reason about mutation, selection, drift, migration, and non-random mating. |
| `BIO-135` | Explain natural selection, adaptation, common ancestry, and speciation using multiple lines of evidence without teleological language. |
| `BIO-136` | Distinguish individual risk, family recurrence, population frequency, and causal determinism in genetic communication. |

**Practice/practical.** Model-organism or simulated crosses; chi-square analysis; pedigree critique; allele-frequency simulation.

**Gate.** Reconcile molecular, pedigree, and population data for an unfamiliar trait and communicate uncertainty.

**Exclusions.** Clinical genetic counselling, polygenic risk-score construction, ecology, and detailed phylogenetic computation.

### `PHY-120` Fluids, solids, and transport — 40 hours

**Prerequisites:** `PHY-110`, `CHE-130` recommended.

| Topic ID | Scope and outcome |
|---|---|
| `PHY-121` | Apply density, pressure, hydrostatic pressure, Pascal's principle, buoyancy, and continuity. |
| `PHY-122` | Use Bernoulli reasoning with explicit assumptions and recognize where viscosity invalidates it. |
| `PHY-123` | Relate laminar flow, resistance, radius, viscosity, and pressure gradient qualitatively and with supplied equations. |
| `PHY-124` | Explain surface tension, capillarity, adhesion/cohesion, and their biological implications. |
| `PHY-125` | Interpret stress, strain, elastic modulus, and failure in simplified tissue/material models. |

**Practice/practical.** Flow-resistance investigation with tubing; pressure sensor or simulation; capillary measurement.

**Gate.** Evaluate a model of flow through a branching system, including assumptions and sensitivity to radius.

**Exclusions.** Full Navier-Stokes derivation, turbulence modelling, and hemodynamic diagnosis.

### `CHE-140` Kinetics, equilibrium, acids/bases, and electrochemistry — 60 hours

**Prerequisites:** `CHE-130`, `QRS-102`.

| Topic ID | Scope and outcome |
|---|---|
| `CHE-141` | Determine and interpret rate, rate law, reaction order, integrated-rate graphs at introductory depth, half-life, collision factors, catalysts, and multi-step energy profiles. |
| `CHE-142` | Apply dynamic equilibrium, reaction quotient, equilibrium constants, and Le Chatelier reasoning while distinguishing kinetics from position. |
| `CHE-143` | Define acids and bases using Brønsted-Lowry and Lewis models; identify conjugate pairs, amphoterism, and relative strength. |
| `CHE-144` | Calculate pH/pOH for strong species and bounded weak-acid/base systems; use `Ka`, `Kb`, `pKa`, percent ionization, and approximations responsibly. |
| `CHE-145` | Explain buffers, Henderson-Hasselbalch use and limits, titration curves, indicators, equivalence, and physiological acid-base context without clinical interpretation. |
| `CHE-146` | Assign oxidation states, balance introductory redox reactions, and relate galvanic/electrolytic cells, potential, free energy, and concentration qualitatively or with supplied equations. |
| `CHE-147` | Apply solubility equilibria, common-ion effects, complexation at survey depth, and competing equilibria in qualitative analysis. |

**Practice/practical.** Reaction-rate investigation; pH-metric titration and buffer-capacity comparison; electrochemical cell or simulation; open-response equilibrium defence.

**Gate.** Investigate an unknown acid/base system and integrate rate, equilibrium, uncertainty, and molecular explanation.

**Exclusions.** Full activity-coefficient treatment, advanced electroanalytical chemistry, and clinical management of acid-base disorders.

### `BIO-140` Molecular biology and control of gene expression — 55 hours

**Prerequisites:** `BIO-130`, `CHE-140` may be concurrent.

| Topic ID | Scope and outcome |
|---|---|
| `BIO-141` | Explain DNA/RNA structure, antiparallel organization, base pairing, chromatin, and genome organization. |
| `BIO-142` | Trace replication, repair, mutation, telomere function, and evidence for semiconservative replication at introductory depth. |
| `BIO-143` | Trace transcription, RNA processing, the genetic code, translation, protein targeting, and degradation. |
| `BIO-144` | Compare prokaryotic and eukaryotic gene regulation and reason from promoter, enhancer, transcription-factor, epigenetic, and RNA-level perturbations. |
| `BIO-145` | Predict molecular and phenotypic consequences of substitution, insertion/deletion, splice, regulatory, and copy-number changes. |
| `BIO-146` | Explain PCR, electrophoresis, restriction analysis, sequencing, cloning, recombinant expression, and CRISPR-Cas systems by purpose, inputs, outputs, controls, and limitations. |
| `BIO-147` | Evaluate the evidence and ethical dimensions of a bounded genetic-technology proposal without assuming that technical feasibility determines acceptability. |

**Practice/practical.** DNA extraction; PCR/electrophoresis simulation or supervised wet lab; sequence alignment; gene-expression dataset; CRISPR claim audit.

**Gate.** Select and defend a molecular strategy to test an unfamiliar gene-function claim, including controls and ethical limits.

**Exclusions.** Production-scale bioinformatics, genome assembly, clinical variant classification, and therapeutic recommendations.

### `PHY-130` Oscillations, waves, sound, and optics — 45 hours

**Prerequisites:** `PHY-110`, `QRS-102`–`QRS-103`.

| Topic ID | Scope and outcome |
|---|---|
| `PHY-131` | Relate period, frequency, amplitude, wavelength, speed, phase, energy, and superposition in oscillations and waves. |
| `PHY-132` | Explain interference, standing waves, resonance, damping, reflection, refraction, and diffraction with diagrams and bounded calculations. |
| `PHY-133` | Apply intensity, inverse-square behavior, decibel scale, pitch, timbre, Doppler effect, and ultrasound principles. |
| `PHY-134` | Use ray models for mirrors, thin lenses, focal length, image formation, and magnification. |
| `PHY-135` | Explain refraction, total internal reflection, dispersion, polarization at survey depth, and optical instrument resolution. |
| `PHY-136` | Connect wave/optics models to hearing, vision, microscopy, fibre optics, and imaging while identifying where physiology or engineering adds complexity. |

**Practice/practical.** Resonance measurement; decibel/logarithm task; lens focal-length and uncertainty investigation; optical-system ray tracing.

**Gate.** Design or evaluate a measurement system using wave and optical evidence.

**Exclusions.** Fourier analysis, wave equations by differential calculus, diagnostic interpretation, and detailed imaging reconstruction.

### `BIO-150` Microorganisms, viruses, and biological evolution — 40 hours

**Prerequisites:** `BIO-140`.

| Topic ID | Scope and outcome |
|---|---|
| `BIO-151` | Compare bacteria, archaea, microbial eukaryotes, and viruses in organization, replication, metabolism, and dependence on hosts. |
| `BIO-152` | Explain bacterial growth, horizontal gene transfer, viral cycles, mutation, selection, and the evolution of resistance. |
| `BIO-153` | Distinguish sterilization, disinfection, antisepsis, asepsis, and antimicrobial selectivity at principle level. |
| `BIO-154` | Use growth curves, plaques, serial dilution, and culture-independent evidence to make bounded quantitative claims. |
| `BIO-155` | Explain microbiomes as ecological communities while avoiding simplistic healthy/unhealthy labels and causal overreach. |
| `BIO-156` | Apply evolutionary reasoning to host-pathogen interaction, virulence tradeoffs, immune escape, and comparative evidence. |

**Practice/practical.** Aseptic-technique demonstration or supervised practice; safe environmental culture only under approved protocol; serial-dilution simulation; outbreak phylogeny exercise.

**Gate.** Explain changing population evidence in an antimicrobial-resistance scenario and design a safe discriminating investigation.

**Exclusions.** Culturing unknown high-risk organisms, pathogen identification for personal health, exhaustive taxonomy, and clinical antimicrobial selection.

### `PHY-140` Electricity, fields, and circuits — 45 hours

**Prerequisites:** `PHY-110`, `CHE-140` recommended.

| Topic ID | Scope and outcome |
|---|---|
| `PHY-141` | Apply charge, Coulomb force, electric field, potential, potential energy, and equipotential reasoning in simple geometries. |
| `PHY-142` | Relate current, voltage, resistance, resistivity, power, and energy; solve series/parallel and simple Kirchhoff circuits. |
| `PHY-143` | Explain capacitance, dielectric behavior, charging/discharging qualitatively, and membranes as capacitive structures with explicit model limits. |
| `PHY-144` | Explain conductivity in metals and ionic solutions and connect electrical measurement to safety and electrode interfaces at survey depth. |
| `PHY-145` | Describe magnetic force, fields around currents, induction, and electromagnetic devices at an introductory, algebra-based level. |
| `PHY-146` | Identify electrical hazards, grounding, isolation, current paths, and the difference between a measurement model and human exposure. |

**Practice/practical.** Build or simulate series/parallel circuits; determine an unknown resistance; RC trace investigation; electrical-safety case.

**Gate.** Diagnose a non-working measurement circuit from quantitative evidence and document safe correction.

**Exclusions.** Maxwell-equation derivations, AC phasors, electrophysiological diagnosis, and device repair beyond supervised low-voltage work.

### `ORG-210` Organic structure, stereochemistry, and reactivity — 50 hours

**Prerequisites:** `CHE-140` and molecular portions of `BIO-110`.

| Topic ID | Scope and outcome |
|---|---|
| `ORG-211` | Interpret line-angle, condensed, wedge-dash, Newman, and skeletal representations; identify constitutional and conformational isomers. |
| `ORG-212` | Name and recognize alkanes, alkenes, alkynes, aromatic systems, halides, alcohols, ethers, thiols, amines, aldehydes, ketones, carboxylic acids and derivatives. |
| `ORG-213` | Determine formal charge, resonance, aromatic stabilization at survey depth, hybridization, geometry, dipoles, and intermolecular interactions. |
| `ORG-214` | Assign stereocentres and `R/S`, alkene `E/Z`, enantiomer/diastereomer relationships, meso forms, optical activity, and biological consequences of stereochemistry. |
| `ORG-215` | Rank acidity/basicity and nucleophile/electrophile behavior from charge, electronegativity, resonance, induction, orbital, solvent, and steric effects. |
| `ORG-216` | Use curved-arrow electron flow, reaction coordinate diagrams, kinetic/thermodynamic control at introductory depth, and equilibrium reasoning. |

**Practice/practical.** Molecular-model portfolio; stereochemical medicine case; acid/base extraction plan; mechanism error correction.

**Gate.** Infer physical and reactive properties of unfamiliar organic structures and justify each prediction mechanistically.

**Exclusions.** Comprehensive IUPAC edge cases, pericyclic orbital theory, and memorization of named reactions without mechanism.

### `PHY-150` Thermal, atomic, nuclear, and radiation physics — 25 hours

**Prerequisites:** `CHE-130`, `PHY-130`.

| Topic ID | Scope and outcome |
|---|---|
| `PHY-151` | Apply temperature, thermal expansion, heat capacity, phase change, conduction, convection, radiation, and the first law to bounded systems. |
| `PHY-152` | Relate photons, spectra, quantized transitions, photoelectric effect, and matter-wave ideas at conceptual/introductory calculation depth. |
| `PHY-153` | Explain nuclear composition, binding, isotopes, alpha/beta/gamma processes, half-life, activity, attenuation, and inverse-square behavior. |
| `PHY-154` | Distinguish ionizing from non-ionizing radiation, absorbed/equivalent dose concepts, stochastic risk, and basic protection principles. |
| `PHY-155` | Compare the physical basis—not clinical interpretation—of radiography, CT, nuclear medicine, ultrasound, and MRI at survey depth. |

**Practice/practical.** Cooling curve or thermal-transfer investigation; half-life simulation; shielding/attenuation dataset; modality comparison.

**Gate.** Calculate and communicate a radiation/thermal exposure scenario with assumptions, units, uncertainty, and risk-language discipline.

**Exclusions.** Quantum derivations, radiation treatment planning, image diagnosis, and operation of regulated radiation sources.

### `ORG-220` Organic reaction families and biological mechanisms — 60 hours

**Prerequisites:** `ORG-210`.

| Topic ID | Scope and outcome |
|---|---|
| `ORG-221` | Compare substitution and elimination (`SN1`, `SN2`, `E1`, `E2`) by substrate, nucleophile/base, solvent, stereochemical, kinetic, and product evidence. |
| `ORG-222` | Explain electrophilic addition to alkenes/alkynes, hydration, oxidation/reduction, and regio-/stereochemical outcomes at introductory depth. |
| `ORG-223` | Explain nucleophilic addition to carbonyls and relate aldehyde/ketone chemistry to biological molecules. |
| `ORG-224` | Explain nucleophilic acyl substitution and compare carboxylic-acid derivatives, hydrolysis, condensation, and biological acyl transfer. |
| `ORG-225` | Explain alcohol, amine, thiol, phosphate, and redox transformations relevant to biomolecules. |
| `ORG-226` | Plan short reaction sequences from a supplied reaction set, tracking chemoselectivity, stereochemistry, yield, purification, and hazards. |

**Practice/practical.** Safe esterification or hydrolysis; reaction-mechanism studio; product-distribution dataset; green-chemistry comparison.

**Gate.** Predict products and mechanisms in an unfamiliar but bounded network, then select evidence that distinguishes alternatives.

**Exclusions.** Unbounded multistep synthesis, organometallic breadth beyond selected examples, and advanced synthesis strategy.

### `BCH-210` Amino acids, proteins, and enzymes — 55 hours

**Prerequisites:** `BIO-120`, `CHE-140`, `ORG-210`.

| Topic ID | Scope and outcome |
|---|---|
| `BCH-211` | Draw amino-acid charge states, estimate predominant form from pH/pKa, classify side-chain chemistry, and calculate peptide net charge in bounded cases. |
| `BCH-212` | Explain peptide-bond properties and primary through quaternary structure, folding forces, disorder, chaperones, denaturation, and aggregation. |
| `BCH-213` | Relate structure to binding, cooperativity, allostery, and the oxygen-binding behavior of myoglobin/haemoglobin as models. |
| `BCH-214` | Apply Michaelis-Menten concepts, `Km`, `Vmax`, turnover, catalytic efficiency, and common inhibition patterns to data without overinterpreting simplified kinetics. |
| `BCH-215` | Explain catalytic strategies, cofactors/coenzymes, zymogens, covalent and allosteric regulation, and feedback. |
| `BCH-216` | Select chromatography, electrophoresis, centrifugation, immunodetection, activity assay, and mass-spectrometry evidence by question and limitation. |

**Practice/practical.** Protein standard curve and unknown; enzyme kinetics/inhibition; chromatography or electrophoresis simulation/physical lab; haemoglobin curve case.

**Gate.** Infer protein function and regulation from sequence/structure, purification, binding, and kinetic evidence.

**Exclusions.** Structural determination pipelines, enzyme mechanism catalogues, and clinical protein-disease management.

### `ORG-230` Separation, spectroscopy, and structure evidence — 40 hours

**Prerequisites:** `ORG-220`, optics portions of `PHY-130`.

| Topic ID | Scope and outcome |
|---|---|
| `ORG-231` | Select extraction, recrystallization, distillation, filtration, and chromatography from physical-property reasoning. |
| `ORG-232` | Interpret TLC and introductory column/GC/HPLC outputs, retention, resolution, standards, and purity limitations. |
| `ORG-233` | Interpret characteristic IR absorptions and connect vibrations to functional groups. |
| `ORG-234` | Interpret introductory proton NMR chemical shift, integration, splitting, equivalence, and exchange using provided ranges. |
| `ORG-235` | Interpret molecular ion, fragments at survey depth, isotope patterns, and accurate-mass/formula evidence in mass spectra. |
| `ORG-236` | Integrate formula, reactivity, separation, IR, NMR, and MS evidence to discriminate plausible structures. |

**Practice/practical.** Pigment or analgesic chromatography; virtual spectra; extraction/purification design; blind structure-elucidation set.

**Gate.** Produce an auditable structure argument from a multi-technique evidence packet.

**Exclusions.** Carbon NMR depth, multidimensional NMR, spectral prediction algorithms, and instrument maintenance.

### `BCH-220` Carbohydrates, lipids, membranes, and nucleotides — 45 hours

**Prerequisites:** `BCH-210`, `ORG-220`.

| Topic ID | Scope and outcome |
|---|---|
| `BCH-221` | Relate monosaccharide stereochemistry, ring formation, glycosidic bonds, storage/structural polysaccharides, and glycoconjugates to function. |
| `BCH-222` | Compare fatty acids, triacylglycerols, phospholipids, sphingolipids, steroids, eicosanoids at survey depth, and their physical/biological roles. |
| `BCH-223` | Explain membrane asymmetry, fluidity, curvature, rafts as a qualified model, transport proteins, gradients, and membrane potential foundations. |
| `BCH-224` | Explain nucleotide chemistry, phosphodiester bonds, base pairing, nucleic-acid stability, ATP/GTP roles, and selected nucleotide-derived cofactors/signals. |
| `BCH-225` | Connect chemical structure to digestion, transport, storage, signalling, and experimentally observable behavior without turning nutrient chemistry into diet advice. |

**Practice/practical.** Lipid-membrane model; carbohydrate stereochemistry; partitioning/chromatography dataset; nutrition-claim evidence audit.

**Gate.** Explain a membrane transport and storage problem from molecular structures through system consequences.

**Exclusions.** Personalized nutrition advice, complete lipidomics, and specialized glycobiology.

### `BCH-230` Bioenergetics and central metabolism — 60 hours

**Prerequisites:** `BCH-210`, `BCH-220`, `CHE-140`.

| Topic ID | Scope and outcome |
|---|---|
| `BCH-231` | Relate free energy, equilibrium, mass action, coupling, phosphorylation potential, redox carriers, and compartmentation. |
| `BCH-232` | Trace glycolysis and gluconeogenesis by carbon, energy, redox, irreversible control points, reciprocal regulation, and tissue context at first-semester depth. |
| `BCH-233` | Explain glycogen synthesis/breakdown and the pentose-phosphate pathway by purpose, regulation, and products. |
| `BCH-234` | Trace pyruvate dehydrogenase, citric-acid cycle, anaplerosis/cataplerosis at survey depth, and their regulation. |
| `BCH-235` | Explain electron transport, proton-motive force, ATP synthase, oxidative phosphorylation, uncoupling, reactive oxygen species, and inhibitor evidence. |
| `BCH-236` | Trace fatty-acid mobilization, transport, beta-oxidation, synthesis, and ketone-body production/use at first-semester depth. |
| `BCH-237` | Explain amino-acid carbon/nitrogen handling at overview depth; the **urea cycle is enrichment, not an MCAT-derived core claim**. |
| `BCH-238` | Compare aerobic/anaerobic yields using clearly declared conventions and reconcile fed, fasting, exercise, and selected organ-context signals. |

**Practice/practical.** Respirometry or simulation; pathway perturbation maps; comparative ATP accounting; multi-tissue fasting case.

**Gate.** Trace labelled atoms and energy/redox consequences through an unfamiliar metabolic perturbation and predict discriminating measurements.

**Exclusions.** Memorizing every enzyme structure, exhaustive inborn errors, urea-cycle detail as a common-core requirement, and treatment planning.

### `BCH-240` Integrated information flow and biochemical regulation — 40 hours

**Prerequisites:** `BIO-140`, `BCH-230`.

| Topic ID | Scope and outcome |
|---|---|
| `BCH-241` | Integrate DNA replication, transcription, translation, turnover, compartmentation, and post-translational modification in a regulated system. |
| `BCH-242` | Explain receptor binding, second messengers, phosphorylation, gene regulation, amplification, feedback, and desensitization across time scales. |
| `BCH-243` | Interpret omics-style heat maps, fold change, normalization, replicates, multiple-testing caution, and pathway enrichment only at conceptual depth. |
| `BCH-244` | Connect molecular perturbations to cellular phenotype while separating loss of function, gain of function, dominant-negative, compensation, and correlation. |
| `BCH-245` | Select molecular and biochemical assays that jointly test a mechanism and specify positive, negative, loading, and process controls. |

**Practice/practical.** Integrated signalling dataset; blot/assay control audit; gene-to-metabolite causal diagram; research-figure journal club.

**Gate.** Defend a causal molecular mechanism using a multi-panel dataset and propose the most informative next experiment.

**Exclusions.** Independent analysis of patient omics, advanced causal inference, and drug-selection recommendations.

## 10. Medicine-facing bridge

These modules are required for the EmbeddedKnowledge Premed completion because they make the foundational sciences usable at the threshold of medical study. The source adjudication does **not** establish them all as universal entrant requirements. They are marked `EK-BRIDGE` in the knowledge graph and on learner records.

### `MED-310` Human structure, transport, and homeostasis — 75 hours

**Prerequisites:** all 100-level Biology; `PHY-120`; `BCH-220`.<br>
**Scope rule:** structure-function and homeostatic reasoning, not clinical diagnosis or memorization of exhaustive anatomical detail.

| Topic ID | Scope and outcome |
|---|---|
| `MED-311` | **Organization and tissues:** orient anatomical planes/terms; compare epithelial, connective, muscle, and nervous tissue; infer function from microstructure. |
| `MED-312` | **Homeostasis:** construct negative/positive feedback models; distinguish regulated variable, sensor, integrator, effector, set range, compensation, and failure. |
| `MED-313` | **Nervous system:** relate membrane potentials, synapses, neurotransmission, sensory/motor organization, autonomic control, and neural integration at introductory depth. |
| `MED-314` | **Endocrine system:** compare peptide/steroid signalling, axes, feedback, receptor regulation, and time course. |
| `MED-315` | **Musculoskeletal system:** connect bone, joint, muscle microstructure, excitation-contraction coupling, force-length/velocity ideas, levers, and repair at survey depth. |
| `MED-316` | **Cardiovascular system:** trace circulation; relate cardiac cycle, electrical conduction, pressure-flow-resistance, vessels, exchange, and regulation. |
| `MED-317` | **Respiratory system:** relate ventilation, compliance, diffusion, perfusion, haemoglobin transport, acid-base contribution, and control of breathing. |
| `MED-318` | **Renal and fluid balance:** trace filtration, reabsorption, secretion, concentration/dilution, electrolytes, osmolarity, volume, and acid-base regulation at introductory depth. |

**Practice/practical.** Histology image atlas; feedback-loop perturbations; ECG principles simulation (not interpretation for care); spirometry demonstration/data; pressure-flow model; urinalysis simulation with privacy-safe synthetic data.

**Gate.** Explain a multi-system homeostatic perturbation from molecular transport through feedback, identifying observations that the model does and does not explain.

**Exclusions.** Named-disease catalogues, diagnostic thresholds, physical examination credentialing, treatment, and patient-specific inference.

### `MED-320` Digestion, reproduction, development, and integrated physiology — 45 hours

**Prerequisites:** `MED-310`, `BCH-230`.

| Topic ID | Scope and outcome |
|---|---|
| `MED-321` | **Digestive system:** trace motility, secretion, digestion, absorption, portal transport, liver/pancreas roles, and regulation by structure and molecular process. |
| `MED-322` | **Energy balance:** distinguish energy intake/expenditure, storage/mobilization, hormonal signals, and population association from individual causation. |
| `MED-323` | **Reproductive systems:** explain gametogenesis, endocrine cycles, fertilization, and reproductive anatomy at introductory, inclusive depth. |
| `MED-324` | **Development:** trace cleavage, implantation, germ layers, morphogenetic patterning at survey depth, placental exchange, and major developmental principles. |
| `MED-325` | **Exercise and environmental integration:** integrate cardiovascular, respiratory, neural, muscular, renal, endocrine, and metabolic responses over time. |
| `MED-326` | **Life-course physiology:** distinguish growth, maturation, ageing, reserve, adaptation, and pathological claims with appropriate uncertainty. |

**Practice/practical.** Digestion/absorption model; endocrine-cycle data; development concept map; exercise-response investigation with opt-out/equivalent dataset.

**Gate.** Construct and defend a time-resolved, cross-system model of feeding, fasting, exercise, pregnancy, or thermal stress from an unseen evidence set.

**Exclusions.** Fertility advice, obstetric management, diet prescription, developmental diagnosis, and detailed embryological anatomy.

### `MED-330` Immunity, infection, biotechnology, and public responsibility — 50 hours

**Prerequisites:** `BIO-150`, `BCH-240`; `MED-310` may be concurrent.

| Topic ID | Scope and outcome |
|---|---|
| `MED-331` | Distinguish barriers, innate sensing, inflammation, complement, phagocytes, antigen presentation, and the logic of rapid defence. |
| `MED-332` | Explain B/T lymphocyte development at survey depth, clonal selection, antibodies, cell-mediated responses, tolerance, contraction, and memory. |
| `MED-333` | Compare active/passive and natural/artificial immunity and explain vaccination using individual and population evidence without making personal recommendations. |
| `MED-334` | Explain infection as an interaction among agent, host, environment, transmission, virulence, dose, immunity, and time—not as a property of a microbe alone. |
| `MED-335` | Explain hypersensitivity, autoimmunity, immune deficiency, transplant rejection, and immune evasion as mechanism classes at survey depth. |
| `MED-336` | Evaluate diagnostic-assay concepts using sensitivity, specificity, likelihood information at introductory depth, predictive value, reference standards, and spectrum bias. |
| `MED-337` | Evaluate PCR, sequencing, CRISPR, recombinant products, immunoassays, cell technologies, and antimicrobial interventions by mechanism, evidence, uncertainty, access, dual-use, and ethical limits. |
| `MED-338` | Apply biosafety, data privacy, informed consent, community impact, antimicrobial stewardship principles, and responsible public communication to a bounded scenario. |

**Practice/practical.** Immunoassay simulation; outbreak model; vaccine-evidence appraisal; diagnostic-test natural frequencies; biotechnology ethics deliberation.

**Gate.** Respond to an evolving infectious/immune evidence packet: explain mechanisms, calculate test implications, critique a study, and produce a public-facing correction that stays outside medical advice.

**Exclusions.** Diagnosis, vaccination or antimicrobial advice, culturing pathogens, advanced immunophenotyping, and clinical immunology management.

## 11. Integration and capstone

### `INT-300` Longitudinal integration studios — 30 hours accounted separately

Integration studios occur at six transition points; they do not wait until the end.

| Studio ID | Trigger | Integrative problem | Required artefact |
|---|---|---|---|
| `INT-301` | after first 100-level cluster | Membrane transport and solution chemistry | annotated model + calculation audit |
| `INT-302` | after `CHE-140`/`BIO-140` | Enzyme, pH, gene regulation, and experimental controls | multi-panel evidence explanation |
| `INT-303` | after Physics core | Pressure, flow, gas exchange, optics, and measurement | model comparison + uncertainty memo |
| `INT-304` | after Organic/Biochem core | Molecular structure to metabolism | atom/energy trace + mechanism defence |
| `INT-305` | during Medicine bridge | Multi-system homeostasis over time | causal graph + narrated explanation |
| `INT-306` | before capstone | Behaviour, society, biology, and ethical evidence (uses pathway material or supplied primer) | stakeholder/evidence brief |

At least four studios must reach `T` (transfer) and the remainder `M`. An integrative answer earns no credit for naming many facts without specifying causal or evidential relationships.

### `CAP-400` Premed readiness capstone — 30 hours

**Prerequisites:** all common-core and bridge gates at `M`; pathway substantially complete.

Learners choose an unfamiliar, non-diagnostic problem from an approved bank or propose one. Examples include oxygen delivery at altitude, antibiotic resistance in a community, an enzyme variant and metabolism, optical screening technology, or stress and health inequality. Personal medical cases are not accepted.

**Required products:**

1. `CAP-401` a one-page question, scope, assumptions, safety/ethics, and evidence plan;
2. `CAP-402` a reproducible analysis package containing source ledger, data dictionary, calculations, figures, and change log;
3. `CAP-403` a 2,500–3,500 word synthesis connecting at least four domains and evaluating at least one primary study;
4. `CAP-404` an 8–10 minute public explanation or accessible equivalent;
5. `CAP-405` a 20-minute oral defence or text-based synchronous defence with two assessors where feasible;
6. `CAP-406` a post-defence correction memo identifying what changed and why.

The capstone cannot compensate for missing module outcomes. Its purpose is transfer, synthesis, provenance, and intellectual honesty.

## 12. Route and breadth pathways

Every learner completes at least one pathway. Pathways are additions to the portable core, not claims about universal preparation. Institution-facing pathways require annual review because admissions rules change.

> **Instrumentation limitation (curriculum version 1.1).** Only `PW-US`, `PW-BIO` and `PW-QNT` are instrumented in the v1 knowledge graph (`site/data/premed-graph.json`) as trackable atomic outcomes. **`PW-UK` and `PW-ES` are documented routes that are NOT yet instrumented.** Their route-defining outcomes — UCAT familiarization (`UK-240`), the QAA Access baseline (`UK-230`), the Spanish admission audit (`ES-211`), and the open-response studio (`ES-250`) — are not among the graph's 388 atomic outcomes, so a learner cannot currently satisfy the section 6.2 "one completed route/pathway" requirement through `PW-UK` or `PW-ES`, and contributors cannot yet author lessons against those outcomes. Learners on those routes should complete `PW-US`, `PW-BIO` or `PW-QNT` for the completion requirement and treat the UK/Spain material below as an uninstrumented reading and audit guide. Instrumenting both routes is planned for a later curriculum version.

### `PW-US` United States graduate-entry / MCAT — 200 hours

**Status:** route-specific, not global core. Always verify each target school's current coursework and laboratory policy.

| Module | Topic-level outcomes |
|---|---|
| `PSY-211` Mind and behaviour | Explain biological bases of behaviour, sensation/perception, consciousness, learning, memory, language, cognition, motivation, emotion, stress, development, personality, psychological disorder concepts, and treatment-evidence categories at first-semester depth. |
| `SOC-221` Social structure and health | Explain socialization, interaction, groups, organizations, institutions, culture, deviance, demographics, migration, urbanization, stratification, inequality, race/ethnicity as social processes, gender, age, and social determinants using non-essentialist language. |
| `BHV-230` Person in context | Integrate attitudes, attribution, identity, self, prejudice, conformity, social support/capital, access, environment, behaviour change, and biopsychosocial evidence. |
| `CAR-240` Critical analysis | Identify argument structure, assumptions, tone, evidence, analogy, contradiction, author perspective, and implications in humanities/social-science passages without relying on outside content. |
| `MCAT-250` Format transfer | Apply common-core knowledge under representative passage/data conditions; build timing only after untimed accuracy; use official-style practice to locate—not conceal—node gaps. |

**Route outcome `PLO-US`.** Solve integrated MCAT-level science and behavioural passages at the AAMC-stated introductory depths and create an institution-specific prerequisite audit. EmbeddedKnowledge does not promise a score.

**Route gate.** Three spaced, mixed-form readiness sets plus an oral error analysis. Official AAMC materials should be used under their terms, not reproduced into the course.

### `PW-UK` UK direct-entry / Access to HE orientation — 180 hours

**Status:** route-specific and institution-dependent. The UCAT is an aptitude test rather than a science syllabus. **Not instrumented in curriculum version 1.1:** no module of the v1 knowledge graph maps to `PW-UK`, and none of the outcomes below exist as trackable atomic outcomes, so this route cannot yet satisfy the section 6.2 pathway requirement and cannot yet be authored against. Planned for a later curriculum version.

| Module | Topic-level outcomes |
|---|---|
| `UK-211` Qualification audit | Map target-course A-level/IB/Access subject and grade rules from current primary admissions pages; identify gaps without treating sector summaries as final authority. |
| `UK-220` Scientific practical portfolio | Demonstrate planning, safe implementation, observation, quantitative treatment, evaluation, and independent reporting across biology and chemistry; where relevant, align evidence to common practical-assessment concepts. |
| `UK-230` Access to HE medicine baseline | Demonstrate the QAA descriptor's breadth: chemistry and biology/human biology, an additional science/mathematics component, numerical data, study skills, and professional behaviours. |
| `UK-240` Aptitude familiarization | Practise current UCAT verbal, decision, quantitative, and situational-judgement constructs using authorized materials; do not invent science content for UCAT. |
| `UK-250` Constructed scientific communication | Complete open-response explanation, practical/research/statistical report, oral presentation, reflective work, and representative selected-response tasks. |

The [QAA Access to HE Diploma Subject Descriptor for Medicine](https://www.qaa.ac.uk/docs/qaa/access-to-he/ahe-subject-descriptor-medicine-25.pdf?sfvrsn=30cfa581_2) is an important access baseline, not a substitute qualification here. The actual diploma has 60 credits: 45 graded Level 3 credits, including at least 15 Chemistry, 15 Biology/Human Biology, and 9 other science/mathematics; its ungraded Level 3 minimum includes 3 credits each in numerical data, study skills, and professional behaviours. EmbeddedKnowledge mirrors the outcomes—atoms, bonds, formulae, periodicity, moles, acids/bases, rates/equilibria, enthalpy, organic chemistry, cells, biomolecules, gene expression, heredity, cardiovascular/respiratory/digestive and further systems, plus laboratory design/performance/interpretation—and its assessment mix spans unseen work, selected response, essays, practical/research/statistical reports, oral presentation, and reflection. The gates remain retakeable. EmbeddedKnowledge does not award QAA credit.

**Route outcome `PLO-UK`.** Produce a current application requirement map and a QAA-informed portfolio spanning scientific knowledge, practical work, data, study, communication, and professional behaviour.

### `PW-ES` Spain / continental direct-entry orientation — 180 hours

**Status:** route-specific. PAU rules and university weighting tables must be versioned by year and jurisdiction. **Not instrumented in curriculum version 1.1:** no module of the v1 knowledge graph maps to `PW-ES`, and none of the outcomes below exist as trackable atomic outcomes, so this route cannot yet satisfy the section 6.2 pathway requirement and cannot yet be authored against. (The graph's `spain-bach-pau` *filter tag* marks Spanish-curriculum content inside other modules; it is not the same thing as an instrumented `PW-ES` route.) Planned for a later curriculum version.

| Module | Topic-level outcomes |
|---|---|
| `ES-211` Admission audit | Distinguish the national PAU access formula and obligatory exercises from university/regional subject weightings; build a current, source-linked target table. |
| `ES-220` Biology emphasis | Consolidate biomolecules, molecular genetics, cell biology, metabolism including autotrophic anabolism (photosynthesis and chemosynthesis), biotechnology, and immunity; distinguish second-year national minimum from broader biology. RD 243/2022 specifies autotrophic anabolic pathways, so photosynthesis belongs to this route; it contains no ecology block and no broad plant-anatomy unit. |
| `ES-230` Chemistry/physics/math emphasis | Solve open and semi-constructed problems at the selected jurisdiction's current specification depth and explain reasoning in the required assessment language. |
| `ES-240` Metabolism and biotechnology depth | Quantitatively compare glycolysis, fermentation, beta-oxidation, citric-acid cycle, electron transport, oxidative phosphorylation; explain PCR, restriction methods, cloning, and CRISPR. |
| `ES-250` Open-response studio | Construct a defensible response, expose intermediate reasoning, interpret novel evidence, and self-audit against published criteria. |

**Route outcome `PLO-ES`.** Demonstrate the selected specification outcomes and maintain a versioned PAU/university-weighting audit without presenting one university's table as a national rule.

### `PW-BIO` Broad biological literacy — 120 hours

For learners targeting AP/A-level/IB/Abitur/Baccalaureat breadth or a later bioscience degree. This pathway includes topics important to biology but weak as a universal premedical minimum.

| Module | Topic-level outcomes |
|---|---|
| `ECO-211` Ecology | Model populations, communities, competition, predation, succession, biodiversity, nutrient cycles, energy flow, and anthropogenic change; analyze field evidence. |
| `PLT-220` Plant biology | Explain plant form, meristems, transport, mineral nutrition, hormones/tropisms, reproduction, and adaptation. |
| `PLT-230` Photosynthesis | Explain light capture, electron transport, chemiosmosis, carbon fixation, photorespiration, and environmental response, linked to but not mislabeled as MCAT content. |
| `EVO-240` Evolutionary breadth | Extend population genetics to phylogenetics, evo-devo survey, coevolution, and macroevolutionary evidence. |
| `BIO-250` Field/practical inquiry | Design and report an ethical observational or manipulative ecology/plant investigation with sampling and statistical reasoning. |

**Route outcome `PLO-BIO`.** Explain and investigate biological systems beyond medicine while clearly distinguishing this breadth from MCAT requirements.

### `PW-QNT` Calculus and extended quantitative science — 160 hours

For mathematics-rich national routes, quantitative bioscience, physics preparation, or learner interest. Calculus is explicitly outside the MCAT mathematics specification and is not a portable entrant minimum.

| Module | Topic-level outcomes |
|---|---|
| `QNT-211` Calculus | Interpret limits, derivatives, and integrals; solve basic optimization, rate, accumulation, and separable growth/decay problems. |
| `QNT-220` Vectors and multivariable ideas | Work with vector components, dot products, partial-change intuition, gradients at survey depth, and parameter sensitivity. |
| `QNT-230` Differential models | Analyze simple exponential/logistic, compartment, oscillatory, and pharmacokinetic teaching models without clinical dosing use. |
| `QNT-240` Extended inference | Use simulation, resampling, multiple-comparison awareness, simple multivariable models, and reproducible code or spreadsheet workflows. |
| `QNT-250` Quantitative project | Build, validate, and communicate one model using a public or synthetic biological dataset. |

**Route outcome `PLO-QNT`.** Use calculus, simulation, and extended statistics to construct and criticize a biological model.

### Custom institution pathway `PW-CUS`

A learner may propose a 120–240 hour pathway mapped to a named, current primary-source specification. Approval requires: stable topic IDs, source versions, measurable outcomes, no collision with the common-core claims, comparable evidence volume, and a route gate. This mechanism supports IB HL, particular A-level boards, Abitur Länder, French specialties, India/NEET, GAMSAT, or an individual school's prerequisites without pretending they are interchangeable.

## 13. Practical and laboratory curriculum

### 13.1 Principle and notation

Practical work has three valid but non-equivalent modalities:

- `P` **physical:** learner handles equipment/materials under an approved risk assessment;
- `S` **simulation:** learner makes experimental choices in a model and analyzes generated data;
- `D` **dataset/observation:** learner analyzes documented real-world or recorded experimental evidence.

The transcript reports the modality for every investigation. A simulation may satisfy design and analysis outcomes but never certifies pipetting, microscopy setup, aseptic handling, dissection, electrical construction, chemical handling, or other physical skill. External programs may reject all remote practical evidence.

### 13.2 Required investigation families

Complete at least twelve, including one from every row. For the blended-practical notation, at least four must be physical, at least one each in Biology and Chemistry, under qualified local supervision where risk requires it.

| Lab ID | Investigation family | Minimum assessable evidence | Permitted modality |
|---|---|---|---|
| `LAB-01` | Measurement, calibration, density | instrument choice, calibration, repeated measures, uncertainty | P/S/D |
| `LAB-02` | Solution preparation and reaction quantity | preparation record, stoichiometry, yield, waste/safety | P/S |
| `LAB-03` | Calorimetry or thermal transfer | energy model, controls, loss analysis | P/S/D |
| `LAB-04` | Kinetics or equilibrium | variable manipulation, rate/equilibrium distinction, graph/model | P/S/D |
| `LAB-05` | Acid-base titration/buffer | standardization or calibration, curve, equivalence, uncertainty | P/S/D |
| `LAB-06` | Membranes/osmosis | biological material/model, controls, quantitative response | P/S/D |
| `LAB-07` | Microscopy/image analysis | scale, sampling, artifact, reproducible measurement | P/S/D |
| `LAB-08` | Enzyme activity | initial-rate logic, replicates, environmental/inhibitor effect | P/S/D |
| `LAB-09` | Genetics/molecular method | cross or PCR/electrophoresis evidence, controls, inference | P/S/D |
| `LAB-10` | Motion/force/energy | measurement model, graph, conservation or force analysis | P/S/D |
| `LAB-11` | Fluid/circuit/wave/optics | model prediction, safe apparatus, parameter estimate | P/S/D |
| `LAB-12` | Separation/spectroscopy | method selection, standards, evidence-based identification | P/S/D |
| `LAB-13` | Physiology | consent/opt-out, non-diagnostic measurement, variation, privacy | P/S/D |
| `LAB-14` | Independent inquiry | question, preregistered plan, risk review, data, revision, limitations | P/S/D |

### 13.3 Practical report standard

Each investigation record contains: question and rationale; method and deviations; hazard/ethics check; raw data with units and provenance; transformations/formulas; visualizations; uncertainty; results; interpretation; alternatives and limitations; reproducibility package; and a correction log. Stock “method/results/conclusion” prose without raw evidence is insufficient.

### 13.4 Safety limits

At-home work is limited to an approved low-risk list. It excludes regulated radiation sources, mains electricity construction, unknown human samples, blood/body fluids, vertebrate dissection, culturing unknown or potentially pathogenic organisms, prescription/controlled substances, unsupervised corrosive/flammable/toxic reagents, self-experimentation that changes medication/diet/sleep, and any experiment presented as personal medical testing. A learner can always choose a simulation or supplied dataset without penalty.

## 14. Learning design and weekly experience

Every standard study week uses a common cycle. Learners may change media, but not the outcome or evidence standard.

1. **Orient (30–45 min):** inspect the node map, prerequisite check, why-it-matters case, and explicit exclusions.
2. **Acquire (4–7 h):** read, watch, or listen to source-linked explanations; complete worked-example pauses and representation checks.
3. **Retrieve (1–2 h):** spaced short-answer and recognition prompts selected by node state, including earlier domains.
4. **Model and solve (4–6 h):** quantitative problems, diagrams, mechanisms, graph reading, and explanation comparison.
5. **Investigate (2–4 h average):** practical, simulation, dataset, or research-method studio.
6. **Discuss (1 h):** AI Socratic dialogue and, where available, peer/tutorial discussion. The learner submits the conversation's correction note, not a transcript dump.
7. **Apply (2–3 h):** medicine-facing or society-facing case that requires transfer but no personal medical advice.
8. **Reflect and plan (30 min):** classify errors, compare confidence with performance, schedule remediation and delayed retrieval.

### Required recurring activities

- one cumulative mixed set every week;
- one graph/table/figure interpretation every week;
- one substantive explanation from memory every week;
- one primary-source figure or method excerpt every two weeks after `QRS-111`;
- one integration studio at each curricular transition;
- one source/provenance and accessibility audit each term;
- one protected recovery week after every 10–12 teaching weeks.

The system must offer a “minimum viable study day” of 15–25 minutes—one retrieval set, one worked example, and the next explicit action—so a disrupted week does not become abandonment.

## 15. Assessment system

### 15.1 Evidence categories

The course reports mastery outcomes, but the following conventional weighting is retained for audit and for learners who need a familiar summary. No category can compensate for an unmastered required node.

| Category | Audit weight | Typical evidence |
|---|---:|---|
| Spaced retrieval and calibration | 10% | delayed checks, confidence accuracy, correction quality |
| Quantitative/problem-solving portfolio | 15% | mixed sets, complete units, model assumptions, alternative strategies |
| Scientific reasoning and source appraisal | 15% | design critiques, paper analyses, causal/evidence judgments |
| Practical/investigation portfolio | 20% | twelve investigations and reproducibility audits |
| Module mastery gates | 20% | retakeable unseen tasks, oral/written explanation, transfer item |
| Integration studios | 10% | cross-domain causal models and evidence briefs |
| Capstone and defence | 10% | synthesis, reproducibility, public explanation, correction memo |

### 15.2 Assessment layers

1. **Checks:** 2–10 minute, low stakes, immediate explanatory feedback. Unlimited attempts; later attempts vary the surface form.
2. **Practice sets:** untimed by default, hints fade, answer explanations require error classification. Practice completion never proves mastery by itself.
3. **Socratic probes:** tutor asks for prediction, mechanism, evidence, boundary case, and self-correction. A human-auditable summary is stored.
4. **Performances:** practicals, data analyses, paper critiques, models, and explanations scored with analytic rubrics.
5. **Mastery gates:** unseen, bounded, retakeable only after a targeted prescription; forms are equivalent by blueprint rather than repeated.
6. **Retention checks:** short delayed probes at 7–14 days and at least 21 days; downstream tasks can also supply retention evidence.
7. **Capstone:** integrated transfer with defence and revision.

### 15.3 Gate blueprint

Every disciplinary gate samples:

- 20–30% accurate explanation and retrieval;
- 25–35% quantitative, symbolic, or representational reasoning;
- 20–30% novel data or experimental-design reasoning;
- 15–25% transfer to a medicine-facing or authentic scientific context;
- at least one limitation/uncertainty item and one misconception trap.

Gate forms are untimed unless a route explicitly requires pacing practice. The normal pass rule is ≥80% overall, ≥70% in every sampled dimension, and 100% on named safety/integrity items. Oral verification is used when authorship or conceptual ownership is unclear, never as punishment.

### 15.4 Common analytic rubric (4 levels)

| Dimension | 4 — Transfer | 3 — Mastery | 2 — Developing | 1 — Insufficient |
|---|---|---|---|---|
| Accuracy and scope | Accurate, appropriately qualified, connects boundaries | Accurate on required scope; no material misconception | Mostly accurate but omission or imprecision changes part of conclusion | Major misconception, contradiction, or unsafe claim |
| Mechanistic reasoning | Coherent multilevel causal chain; tests alternatives | Correct causal/mechanistic chain with relevant steps | Names factors but leaves causal gaps or confuses levels | Restates outcome, uses teleology, or asserts causation without mechanism/evidence |
| Quantitative work | Correct model, units, uncertainty, sensitivity, plausibility | Correct setup/calculation/units and interpretation | Minor procedural errors or weak interpretation | Wrong model/denominator/units or uninterpretable result |
| Evidence and design | Triangulates evidence; identifies bias and decisive follow-up | Interprets data and controls; states warranted conclusion/limits | Partial interpretation or generic limitations | Ignores controls, confounds, or overclaims |
| Representation | Moves consistently among words, equations, graphs, diagrams | Uses the requested representation correctly | Understandable but incomplete/inconsistent | Representation conflicts with claim or hides reasoning |
| Communication and provenance | Audience-fit, concise, accessible, fully traceable | Clear, organized, sources/tool use disclosed | Meaning recoverable; citation/accessibility gaps | Misleading, copied, inaccessible, or source provenance absent |
| Reflection and correction | Diagnoses root cause, generalizes repair, verifies it | Corrects error and states prevention strategy | Corrects answer without explaining cause | Defends error, edits invisibly, or cannot reproduce work |

A score of 3 on all dimensions required by the task constitutes mastery. A 4 is evidence of transfer, not extra decorative detail.

### 15.5 Practical rubric additions

Practical work also requires a 3 or 4 in: safety/ethics; method fitness; raw-record quality; calibration/controls; data treatment; uncertainty/limitations; reproducibility. A serious safety breach ends the procedure safely and moves the learner to remediation; it does not permanently bar a retry.

### 15.6 Capstone rubric

The capstone uses the common rubric plus four weighted dimensions:

| Dimension | Weight | Mastery anchor |
|---|---:|---|
| Cross-domain integration | 25% | at least four domains linked by explicit, correct relationships rather than juxtaposed sections |
| Evidence and reproducibility | 25% | source/data provenance, auditable analysis, result reproduced by assessor |
| Scientific judgment | 20% | assumptions, alternatives, uncertainty, ethics, and boundary of claim are material to conclusion |
| Communication and defence | 20% | expert paper and public artefact are accurate; answers reveal ownership and allow correction |
| Revision | 10% | correction memo makes substantive, traceable changes after critique |

Acceptance requires ≥3/4 on each dimension. Learners revise and defend again when needed.

### 15.7 Record issued

The public/exportable record states:

- curriculum and source-alignment version;
- each module and outcome state (`M`, `R`, or `T` only; earlier states remain private by default);
- dates of first mastery and most recent retention evidence;
- practical titles and modality (`P`, `S`, `D`), never implying hands-on work where none occurred;
- pathway(s) and capstone title;
- accommodations used only when the learner elects disclosure;
- an explicit non-credit/non-licensure notice.

## 16. Accessibility, inclusion, and learner safety

The course follows an accommodation-by-design principle:

- all essential video has captions and a transcript; audio has a text equivalent;
- diagrams include alt text and a structured description; color is never the only code;
- math is supplied in accessible notation plus a plain-language reading;
- interactions are keyboard-operable; timed animation can pause; flashing content is excluded;
- PDFs are not the sole presentation of required material where a structured HTML alternative can be provided;
- learners can replace speaking/listening performances with live text or equivalent modes while preserving the reasoning outcome;
- extended time is unnecessary for untimed gates; scheduled defences allow rest breaks and flexible time zones;
- practicals have non-penalizing simulation/data alternatives for disability, location, cost, pregnancy, allergy, religious/ethical concerns, or simple preference;
- physiology activities never require disclosure of health information, sex characteristics, disability, weight, medication, or diagnosis;
- examples use diverse bodies and populations without presenting a demographic average as a biological norm;
- sex, gender, race, ancestry, disability, and socioeconomic position are represented with scientifically appropriate constructs and explicit limits;
- distressing topics carry advance notices and equivalent cases where possible.

Accessibility changes the path, medium, or timing—not the scientific outcome being demonstrated. When an outcome genuinely depends on a physical action, the record accurately marks whether that action was performed.

## 17. Academic integrity, AI, and privacy

### 17.1 Integrity standard

Learners may collaborate, consult sources, and use tools on learning tasks unless a gate is marked closed-resource. They must preserve authorship and provenance. Prohibited conduct includes fabricated data or citations, undisclosed copying, submitting another person's reasoning, changing raw data without a log, using personal health data without approval, or having an AI generate evidence the learner cannot explain and defend.

### 17.2 AI use

AI is permitted for questioning, alternative explanations, feedback, language support, code/spreadsheet debugging, and formative generation when:

- the learner discloses tool/model and material use in assessed work;
- all factual and bibliographic claims are checked against accessible sources;
- no confidential, personal-health, or third-party data are uploaded;
- the learner retains drafts, prompts or an equivalent use log when AI materially shaped the work;
- the learner can reproduce calculations and defend conclusions without the tool.

AI output is never a source solely because it is fluent. Mastery gates may require controlled conditions and a brief authorship verification.

### 17.3 Citation and licenses

Use a consistent citation system (APA, Vancouver, or a discipline-appropriate equivalent). Cite exact edition/version and retrieval date for mutable web sources. Open licenses are asset-specific: provider names such as “OpenStax” or “LibreTexts” do not establish one blanket license. Before adapting content, record title, edition, URL/format, displayed license, attribution text, modification, retrieval date, and intended use in the rights ledger.

## 18. Mastery recovery and remediation

### 18.1 Error taxonomy

Every substantive miss is classified as one or more of:

- `RCL` retrieval gap;
- `VOC` terminology/reading gap;
- `REP` representation translation gap;
- `MOD` wrong or misapplied model;
- `QNT` algebra/unit/denominator/numeracy gap;
- `MEC` causal/mechanistic gap;
- `EVD` data/design/evidence gap;
- `PRE` missing prerequisite;
- `CAL` confidence-calibration problem;
- `SAF` safety, ethics, privacy, or scope problem.

### 18.2 Recovery protocol

1. Stop and name the failing node and error class.
2. Check prerequisite nodes; reopen the earliest insecure dependency, not the whole module.
3. Complete a contrasting example and explain why the original strategy failed.
4. Produce a fresh representation from memory.
5. Solve two varied problems without hints, separated in time.
6. Retry an equivalent gate no earlier than the prescribed interval (normally 24 hours; longer for repeated high-confidence misconceptions).
7. Schedule delayed retention evidence.

After two unsuccessful gate attempts, a tutor or structured diagnostic reviews the prerequisite map. After three, the learner receives a rewritten route through the node, an alternate modality, and a live/human review where available. The standard does not fall; the route changes.

### 18.3 Lapse policy

A later failure does not erase historical evidence. A mastered node moves to “review due,” and high-risk downstream work is paused only if the lapse is material. One successful bounded refresh restores active mastery; recurrent lapses trigger a broader prerequisite audit.

### 18.4 Motivation protection

Progress views show secure nodes, next reachable nodes, hours invested, and recovered misconceptions—not rank, streak shame, or peer comparison. A learner can declare a recovery week, reduce to the minimum viable study day, or switch media without resetting progress. Inactivity triggers a compassionate re-entry diagnostic, not a backlog dump.

## 19. Suggested schedules and workload

### 19.1 Standard 72-week route (22–24 hours/week)

| Block | Weeks | Main modules | Milestone |
|---|---:|---|---|
| Welcome | Before week 1 | `WEL-000` | understand the journey and destination |
| Placement/on-ramp | 0–6 as needed | diagnostics, `ONR-000` | individualized starting map |
| A. Scientific inquiry and language | 1–8 | `QRS-100`, `CHE-110`, `BIO-110` | `INT-301` |
| B. Particles to cells | 9–16 | `CHE-120`, `BIO-120`, `PHY-110` | first practical review |
| C. Matter, inheritance, transport | 17–24 | `CHE-130`, `BIO-130`, `PHY-120` | cumulative gate + recovery week |
| D. Regulation and measurement | 25–32 | `CHE-140`, `BIO-140`, `PHY-130` | `INT-302`/`INT-303` |
| E. Reactivity and systems | 33–40 | `BIO-150`, `PHY-140`/`PHY-150`, `ORG-210` | six-practical checkpoint |
| F. Molecules of life | 41–48 | `ORG-220`/`ORG-230`, `BCH-210`/`BCH-220` | structure-evidence gate |
| G. Integrated biochemistry | 49–54 | `BCH-230`, `BCH-240` | `INT-304` |
| H. Medicine-facing bridge | 55–62 | `MED-310`–`MED-330` | `INT-305`, twelve-practical checkpoint |
| I. Route pathway | 63–68 | one `PW-*` (some pathways begin earlier) | route gate, `INT-306` |
| J. Capstone | 69–72 | `CAP-400`, retention audit | defence and completion review |

Pathways with 180–240 hours normally start during blocks F–H or extend the route by 4–8 weeks. Recovery weeks are included approximately after weeks 12, 24, 36, 48, and 60 and should contain no new required concepts.

### 19.2 Part-time 96-week route (17–18 hours/week)

Use the same dependencies, stretching each eight-week block to roughly eleven weeks and reserving every twelfth week for recovery/integration. Complete at least two short retrieval contacts per week; avoid a single long weekly session. Practicals may be batched into supervised weekends.

### 19.3 Accelerated route

An accelerated 48–56 week route is suitable only after diagnostics demonstrate secure prior science and the learner can sustain 28–34 hours/week. Acceleration removes repeated instruction, never outcomes, practical evidence, spaced retention, or capstone defence.

### 19.4 Workload model

The 1,440-hour common-core/bridge estimate is approximately:

| Activity | Hours | Share |
|---|---:|---:|
| Source study and worked examples | 360 | 25% |
| Retrieval and deliberate practice | 230 | 16% |
| Quantitative/problem sets | 260 | 18% |
| Practical/simulation/data investigations | 230 | 16% |
| Cases, discussion, and integration | 150 | 10% |
| Assessments, revision, and capstone | 150 | 10% |
| Planning, feedback, and reflection | 60 | 4% |

These are planning estimates, not seat-time requirements. Fast completion of a mastered node does not require artificial time filling; struggling learners receive extra practice without penalty.

## 20. Teaching team and delivery requirements

### 20.1 Learner equipment and accounts

Required access is limited to a current standards-compliant browser; stable enough connectivity to download text/low-bandwidth packages; an email or privacy-preserving notification channel; a scientific calculator; paper or digital notebook; and spreadsheet software capable of formulas, plots, and CSV import/export. A local/offline package must preserve readings, transcripts, prompts, and progress sync where feasible. Programming is not required in the common core.

Physical practicals use a published, low-cost kit list and local household/school equipment only after risk review. No learner must purchase a microscope, sensor, chemical kit, commercial textbook, test-prep subscription, or AI subscription to earn the standard online completion. Costs and supervision requirements are disclosed before pathway selection.

### 20.2 Course operations

- There are no late penalties in the self-paced route. Cohort deadlines organize feedback access, but an equivalent later window remains available.
- Learner questions receive an acknowledgement within two working days and substantive instructional feedback within five working days when human support is included.
- Assessed artefacts receive rubric-linked feedback. A gate result must identify the exact weak nodes and a feasible next action.
- Planned service outages, source withdrawals, and substantive corrections are shown in the learner record; progress data can be exported before migration.
- Learners can report content, safety, accessibility, privacy, or conduct concerns without exposing the report publicly.
- Harassment, discriminatory conduct, unsafe practical pressure, falsification, and retaliation are incompatible with participation. Sanctions require a reviewable process and preserve access to the open learning content where safety permits.

### 20.3 Teaching roles

A credible implementation requires more than generated lessons:

- **curriculum lead:** owns outcome map, versioning, assessment blueprint, and cross-domain coherence;
- **domain editors:** Biology/Biochemistry, Chemistry/Organic, Physics/Quantitative, Behaviour/Society, and Human Physiology;
- **practical lead:** approves risks, modalities, local-supervision standards, equipment lists, and safety-critical gates;
- **assessment editor:** maintains parallel forms, rubrics, item statistics, bias review, and standard setting;
- **accessibility reviewer:** audits content and assessments before release;
- **source/rights steward:** maintains provenance, license ledger, source freshness, and correction trail;
- **learner support/tutors:** perform misconception diagnosis, capstone supervision, authorship checks, and re-entry planning;
- **medical/ethics reviewer:** checks medicine-facing contexts for scope, harm, stereotyping, and accidental advice.

AI may draft or review at scale, but high-risk safety, medical framing, assessment standards, and disputed claims require named human accountability before publication.

Recommended cohort support is one facilitated tutorial of 8–16 learners per week, plus monthly practical clinics. Fully self-paced learners receive equivalent asynchronous discussion and scheduled assessor access.

## 21. Learning resources and adoption rules

### 21.1 Primary alignment resources

- [AAMC, *What's on the MCAT Exam?* full content outline](https://students-residents.aamc.org/media/9261/download) — principal public taxonomy and depth statement for the US pathway; not a global admissions mandate.
- [AAMC Premed Competencies](https://students-residents.aamc.org/real-stories-demonstrating-premed-competencies/premed-competencies-entering-medical-students) — scientific inquiry, quantitative reasoning, living systems, human behaviour, communication, teamwork, and reliability framing.
- [AAMC admission requirements overview](https://students-residents.aamc.org/medical-school-admission-requirements/admission-requirements) and [school-specific directory](https://students-residents.aamc.org/medical-school-admission-requirements/required-premedical-coursework-and-competencies) — evidence that US requirements are school-controlled.
- [QAA Access to HE Subject Descriptor: Medicine](https://www.qaa.ac.uk/docs/qaa/access-to-he/ahe-subject-descriptor-medicine-25.pdf?sfvrsn=30cfa581_2) — medicine-access breadth, laboratory, data, study-skill, professional-behaviour, and assessment baseline.
- [Spain RD 243/2022](https://www.boe.es/buscar/act.php?id=BOE-A-2022-5521) and [RD 534/2024](https://www.boe.es/buscar/act.php?id=BOE-A-2024-11858) — national Bachillerato/PAU floor and access mechanics, distinct from university weights.
- [AQA A-level Biology specification](https://filestore.aqa.org.uk/resources/biology/specifications/AQA-7401-7402-SP-2015.PDF), [OCR Biology A specification](https://www.ocr.org.uk/Images/687834-download-a-level-specification.pdf), and [Pearson Edexcel Biology A specification](https://qualifications.pearson.com/content/dam/pdf/A%20Level/biology-a/2015/specification-and-sample-assessment-materials/9781446930885-gce2015-a-bioa-spec.pdf) — practical and quantitative expectations plus broad biology.
- [AP Biology Course and Exam Description](https://apcentral.collegeboard.org/media/pdf/ap-biology-course-and-exam-description.pdf) — science practices, inquiry laboratory emphasis, and broad biology; AP is optional, not a national premed requirement.
- [IB Diploma sciences](https://ibo.org/programmes/diploma-programme/curriculum/sciences/) — interdisciplinary science framework; one Group 4 choice does not imply all sciences.
- [French terminale Physics-Chemistry](https://www.education.gouv.fr/bo/19/Special8/MENE1921249A.htm) and [SVT](https://www.education.gouv.fr/bo/19/Special8/MENE1921252A.htm) — route-specific secondary science scope.
- [KMK educational standards](https://www.kmk.org/themen/qualitaetssicherung-in-schulen/bildungsstandards/bildungsstandards-und-allgemeine-hochschulreife.html) — German cross-Land reference; not a substitute for each Land and course choice.

### 21.2 Teaching-material candidates

- [OpenStax subjects](https://openstax.org/subjects), especially versioned Biology, Chemistry, Organic Chemistry, Physics, Anatomy & Physiology, Psychology, and Sociology titles;
- [LibreTexts Chemistry](https://chem.libretexts.org/) and relevant Biology/Physics libraries, with page-level provenance;
- [MIT OpenCourseWare 5.111 Principles of Chemical Science](https://ocw.mit.edu/courses/5-111-principles-of-chemical-science-fall-2008/);
- [Khan Academy MCAT collection](https://www.khanacademy.org/prep/mcat), still linked by the [AAMC free resources page](https://students-residents.aamc.org/prepare-mcat-exam/free-planning-and-study-resources) as of this curriculum version;
- official AAMC free practice and outline-mapping resources under their published terms.

No resource is “the course.” Editors select at node level, verify currency and accessibility, declare what a resource does **not** cover, and supply an alternative modality. Resource availability is checked at least annually and before a high-stakes route cohort begins.

### 21.3 Resource-selection checklist

Adopt or link an asset only when it has: exact scope match; named author/provider; stable URL or archived/versioned identifier where lawful; publication/update date; source citations; accessible alternative; no uncorrected material error; appropriate depth; acceptable privacy/tracking behavior; and license/terms compatible with the intended use. “Free to view” is not permission to adapt or redistribute.

## 22. Source alignment and design rationale

| Curriculum choice | Evidence basis | Design judgment / boundary |
|---|---|---|
| Biology, general chemistry, organic chemistry, algebra-based physics, first-semester biochemistry | Reconciled 3/3 report spine; official MCAT depth; QAA access breadth | Required common core; does not imply every system requires every course. |
| Quantitative reasoning, experimental design, data/statistical reasoning | MCAT science-skills weighting; AP practices; A-level practical/quantitative work; QAA numerical-data outcomes | Distributed spine and 20% practical portfolio, not a one-off methods unit. |
| Statistics before calculus | Repeated reasoning evidence; MCAT mathematics excludes calculus | Intro probability/inference required; calculus pathway optional. |
| Human physiology/homeostasis | Strong medicine-facing value, MCAT organ-system content, QAA Human Biology breadth | Required `EK-BRIDGE`; explicitly not labelled universal entrant minimum. |
| Immunity and biotechnology | MCAT/Spanish Biology and medicine-facing relevance | Required bridge at introductory depth; no clinical recommendations. |
| Psychology/sociology | Full MCAT domain but weak as global requirement | `PW-US`, not portable common core; supplied behavioural primer supports `INT-306` for others. |
| Ecology | Broad school Biology (AQA/AP); no ecology block in Spain's second-year Biology; absent from MCAT and weak as premedical minimum | `PW-BIO` only; not attributed to MCAT and not attributed to `spain-bach-pau`. |
| Plant structure, transport, signalling and reproduction | Broad school Biology (AQA/AP); RD 243/2022 has no broad plant unit; absent from MCAT | `PW-BIO` only; not attributed to MCAT and not attributed to `spain-bach-pau`. |
| Photosynthesis | Broad school Biology **and** RD 243/2022 autotrophic anabolism; absent from MCAT | `PW-BIO` plus `spain-bach-pau`/`ES-220`; photosynthesis not falsely attributed to MCAT. |
| Evolution | Broad Biology and present in MCAT; absent from Spain's second-year slice | Required within Biology, with route-specific review. |
| Organic mechanisms/synthesis/spectroscopy | Cross-system but depth uneven | Common core bounded to biologically relevant mechanisms, short supplied sequences, and introductory evidence; advanced synthesis excluded. |
| Physical laboratory evidence | A-level practical endorsements, AP inquiry emphasis, US/QAA laboratory expectations | Twelve investigations; modality disclosed; online simulation never presented as hands-on equivalence. |
| Retakeable gates | EmbeddedKnowledge anti-exam mission | Retains unseen transfer and safety standards while removing one-shot judgment. |
| Route audits | Admissions volatility and jurisdictional divergence | Learner verifies named current institutions; course never promises eligibility. |

## 23. Quality assurance and revision

### 23.1 Release gates

A module moves from draft to learner-facing only when:

1. every outcome maps to at least one teaching asset, practice item, gate item, prerequisite, and stated exclusion;
2. a domain editor verifies factual accuracy against primary or high-quality scholarly sources;
3. a separate reviewer checks the gate against the outcome and difficulty blueprint;
4. accessibility and bias checks pass;
5. practical hazards and data/privacy implications are approved;
6. all third-party assets have a rights-ledger entry;
7. at least one novice and one prepared learner complete a usability pilot;
8. corrections have public version history.

### 23.2 Monitoring

Monitor node-level success, hint dependence, time-on-task ranges, high-confidence misconceptions, gate-form comparability, retention, accessibility failures, adverse/safety reports, and withdrawal points. Do not optimize for clicks, streak length, or rapid completion at the expense of retention or well-being.

Suggested triggers for review are: >20 percentage-point performance difference between equivalent gate forms; >30% of prepared learners failing the same non-intentional item; a subgroup gap unexplained by prerequisite evidence; any credible safety issue; a primary source/version change; or a substantiated content flag.

### 23.3 Correction severity

- `C1 editorial`: wording, accessibility, broken link; no scientific conclusion changes.
- `C2 substantive`: explanation or assessment ambiguity; affected learners are notified and evidence re-evaluated if needed.
- `C3 safety/factual`: could cause unsafe action or materially false learning; content is withdrawn, linked nodes frozen, learners notified, and expert/source adjudication required before restoration.

Annual review covers admissions links, source versions, route definitions, resource availability/licenses, and aggregate evidence. A generational AI re-audit may propose changes but cannot silently overwrite published content or learner records.

## 24. Boundaries and completion interpretation

Completion means the learner has demonstrated this version's outcomes under the stated conditions. It does not establish:

- eligibility for any particular medical school;
- equivalence to named school qualifications, university credits, laboratory endorsements, QAA credit, or an MCAT/UCAT/PAU score;
- clinical competence, permission to treat, or reliability of personal medical decisions;
- mastery of ecology/plant science, calculus, advanced synthesis, programming, or another optional pathway unless recorded;
- permanent retention without continued practice.

The honest next step after Premed is a learner-specific readiness review: inspect retained nodes, verify the current target institution or next-course requirements, and prescribe only the missing bridge—not repeat the entire syllabus.

## Appendix A — Misconceptions that must be directly tested

| ID | Misconception | Disconfirming performance |
|---|---|---|
| `MIS-01` | heavier/faster objects experience a different gravitational acceleration under the same ideal conditions | predict and explain controlled motion data |
| `MIS-02` | force is required to maintain constant velocity | free-body/model comparison |
| `MIS-03` | pressure and flow are interchangeable | reason through a changed resistance/radius scenario |
| `MIS-04` | equilibrium means equal concentrations or stopped reactions | particle/model and `Q` versus `K` task |
| `MIS-05` | a catalyst changes equilibrium or reaction free energy | energy diagram plus equilibrium evidence |
| `MIS-06` | spontaneity means fast | contrast thermodynamic and kinetic data |
| `MIS-07` | pH is linear and buffers keep pH perfectly fixed | logarithmic and buffer-capacity calculations |
| `MIS-08` | oxidation always adds oxygen; reduction always removes it | electron/oxidation-state account |
| `MIS-09` | dominant means common, stronger, or better | pedigree/population counterexample |
| `MIS-10` | one gene normally determines one trait or one protein | regulatory/alternative processing and polygenic examples |
| `MIS-11` | evolution is purposeful improvement of individuals | population-level selection explanation |
| `MIS-12` | enzymes create energy or make impossible reactions favourable | free-energy/coupling and rate distinction |
| `MIS-13` | membranes are static walls and diffusion stops at equal amount | dynamic/electrochemical equilibrium model |
| `MIS-14` | ATP is long-term stored energy or has uniquely “high-energy bonds” in isolation | coupled-reaction and storage comparison |
| `MIS-15` | respiration means only breathing | distinguish cellular and organismal processes and connect them |
| `MIS-16` | all microbes are pathogens; antibiotics treat viruses | classification/mechanism scenario |
| `MIS-17` | immunity is simply stronger or weaker | response-type, specificity, tolerance, timing model |
| `MIS-18` | a sensitive test makes a positive result probably true regardless of prevalence | natural-frequency predictive-value task |
| `MIS-19` | statistical significance proves importance, causation, or truth | effect/interval/design critique |
| `MIS-20` | no significant result proves no effect | power/interval and equivalence-language task |
| `MIS-21` | correlation, a plausible mechanism, or an animal experiment alone establishes human causation | evidence-chain comparison |
| `MIS-22` | uncertainty is a mistake that should be hidden | measurement and conclusion audit |
| `MIS-23` | an anatomical or population average defines what is normal for every person | variation and construct critique |
| `MIS-24` | a fluent AI answer or prestigious source is self-validating | source triangulation and citation verification |

## Appendix B — Minimal learner portfolios

By completion, the private learner workspace contains:

- a mastery ledger and dependency view;
- a dated error log with at least five documented misconception recoveries;
- a quantitative portfolio covering units, graphs, probability, diagnostic tests, inference, and model critique;
- twelve practical reports with raw evidence and modality labels;
- six integration-studio artefacts;
- six introductory primary-paper appraisals across at least four domains;
- an institution/route audit with retrieval dates;
- a capstone reproducibility package and correction memo;
- a final 30-day study-maintenance plan.

## Appendix C — Versioning convention

Topic IDs remain stable when wording is clarified. A changed scientific outcome receives a new ID or version suffix and a migration note. This syllabus uses semantic versions:

- patch (`1.0.x`) for links/editorial/accessibility changes without outcome changes;
- minor (`1.x.0`) for added optional topics, resources, or assessment forms;
- major (`x.0.0`) when required outcomes, pathways, or mastery standards change.

Learner records always retain the curriculum version under which evidence was earned.


---

## Repository file: course/PREMED-CURRICULUM-MAP.md

# Premed Curriculum Implementation Map

**Companion to:** `PREMED-SYLLABUS.md` version 1.0 candidate<br>
**Purpose:** registrar/editor crosswalk for sequencing, assessment coverage, and content production<br>
**Not the visual knowledge graph:** the site graph may render these stable IDs and relationships differently.

> **Boundary.** This is educational preparation only. It is **not** a degree, academic credit, admission guarantee, medical qualification, or medical advice. Enumerating a medical curriculum does not confer clinical authority; nothing here should be used to make a decision about a real patient, including yourself.

## 1. Node contract

Every implemented topic node must contain:

| Field | Requirement |
|---|---|
| Stable ID | Exact syllabus topic ID; never recycle an ID for a different outcome. |
| Title and one-sentence purpose | Learner-readable, medicine relevance accurate and non-sensational. |
| Status | `PORTABLE-CORE`, `EK-BRIDGE`, `PATHWAY-US`, `PATHWAY-UK`, `PATHWAY-ES`, `PATHWAY-BIO`, `PATHWAY-QNT`, or `ONRAMP`. |
| Prerequisites | Topic or module IDs that must be at least `P`; mark `hard` or `recommended`. |
| Measurable outcome | Observable verb plus scope, representations, and expected depth. |
| Exclusions | At least one boundary when learners may reasonably overgeneralize the node. |
| Representations | Applicable set: verbal, symbolic, equation, graph, table, particle, mechanism, pathway, image, causal model. |
| Learning evidence | acquisition resource, retrieval prompts, worked examples, independent problems, transfer task. |
| Assessment evidence | gate item family, rubric dimensions, misconception traps, retention task. |
| Practical link | relevant `LAB-*` family and valid modality, or `none`. |
| Source alignment | exact source/version and whether it supports scope, depth, practice, or route only. |
| Rights and accessibility | asset-level rights entry; transcript/alt text/math/keyboard status. |
| Stewardship | named domain owner, last review, next review, open flags, version. |

The graph may show progress by node. It must not infer prerequisite mastery from time spent, resource completion, or confidence alone.

## 2. Module registry

| Order | Module | Status | Hard prerequisites | May co-study | Direct downstream modules | Principal gate evidence |
|---:|---|---|---|---|---|---|
| 0 | `ONR-000` | ONRAMP | entry diagnostic prescription | — | all 100-level | station retry and two-week study cycle |
| 1 | `QRS-100` | PORTABLE-CORE | selected `ONR-*` | all disciplines after foundations | every later module | multi-figure study critique |
| 2 | `CHE-110` | PORTABLE-CORE | `QRS-101`–`104` | `BIO-110` | `CHE-120`, `BIO-110` | unknown-material evidence argument |
| 3 | `BIO-110` | PORTABLE-CORE | entry bio primer or diagnostic | `CHE-110` | `BIO-120`, `BCH-220` | unfamiliar transport disorder |
| 4 | `CHE-120` | PORTABLE-CORE | `CHE-110`, `QRS-102` | `BIO-120`, `PHY-110` | `CHE-130`, `BIO-120` | microscale synthesis audit |
| 5 | `BIO-120` | PORTABLE-CORE | `BIO-110` | `CHE-120` | `BIO-130`, `BCH-210` | enzyme/cell-cycle experiment |
| 6 | `PHY-110` | PORTABLE-CORE | `QRS-101`–`103` | `CHE-120`, `BIO-120` | `PHY-120`–`150`, `MED-315` | movement model validation |
| 7 | `CHE-130` | PORTABLE-CORE | `CHE-120` | `BIO-130`, `PHY-120` | `CHE-140`, `PHY-150` | gas/solution/thermal model |
| 8 | `BIO-130` | PORTABLE-CORE | `BIO-120`, `QRS-107` | `CHE-130` | `BIO-140`, pathways | multi-level trait evidence |
| 9 | `PHY-120` | PORTABLE-CORE | `PHY-110` | `CHE-130` | `MED-316`–`318` | branching-flow model |
| 10 | `CHE-140` | PORTABLE-CORE | `CHE-130`, `QRS-102` | `BIO-140`, `PHY-130` | `ORG-210`, `BCH-210` | unknown acid/base system |
| 11 | `BIO-140` | PORTABLE-CORE | `BIO-130` | `CHE-140` | `BIO-150`, `BCH-240` | molecular strategy/design |
| 12 | `PHY-130` | PORTABLE-CORE | `PHY-110`, `QRS-102/103` | `CHE-140` | `ORG-230`, medicine bridge | wave/optical measurement system |
| 13 | `BIO-150` | PORTABLE-CORE | `BIO-140` | `PHY-140` | `MED-330` | resistance scenario investigation |
| 14 | `PHY-140` | PORTABLE-CORE | `PHY-110`, recommended `CHE-140` | `BIO-150`, `ORG-210` | medicine bridge | circuit diagnosis and safety |
| 15 | `ORG-210` | PORTABLE-CORE | `CHE-140`, `BIO-110` | `PHY-140/150` | `ORG-220`, `BCH-210` | structure/reactivity inference |
| 16 | `PHY-150` | PORTABLE-CORE | `CHE-130`, `PHY-130` | `ORG-210` | medicine bridge | radiation/thermal communication |
| 17 | `ORG-220` | PORTABLE-CORE | `ORG-210` | `BCH-210` | `ORG-230`, `BCH-220` | bounded reaction network |
| 18 | `BCH-210` | PORTABLE-CORE | `BIO-120`, `CHE-140`, `ORG-210` | `ORG-220` | `BCH-220/230` | protein-function evidence |
| 19 | `ORG-230` | PORTABLE-CORE | `ORG-220`, `PHY-130` | `BCH-220` | research evidence nodes | structure elucidation |
| 20 | `BCH-220` | PORTABLE-CORE | `BCH-210`, `ORG-220` | `ORG-230` | `BCH-230`, `MED-310` | membrane/storage problem |
| 21 | `BCH-230` | PORTABLE-CORE | `BCH-210/220`, `CHE-140` | — | `BCH-240`, `MED-320` | metabolic perturbation |
| 22 | `BCH-240` | PORTABLE-CORE | `BIO-140`, `BCH-230` | — | `MED-330` | multi-panel causal mechanism |
| 23 | `MED-310` | EK-BRIDGE | Biology 100-level, `PHY-120`, `BCH-220` | pathways | `MED-320`, `INT-305` | homeostatic perturbation |
| 24 | `MED-320` | EK-BRIDGE | `MED-310`, `BCH-230` | `MED-330` | `INT-305` | time-resolved integration |
| 25 | `MED-330` | EK-BRIDGE | `BIO-150`, `BCH-240` | `MED-320` | `INT-305/306` | infection/immune evidence packet |
| 26 | `INT-300` | PORTABLE-CORE | staged triggers | all stages | `CAP-400` | six cross-domain artefacts |
| 27 | one `PW-*` | PATHWAY | pathway-dependent; most core completed | medicine bridge | `CAP-400` | route-specific gate and current audit |
| 28 | `CAP-400` | PORTABLE-CORE | all required gates, pathway near complete | — | later medical sciences | reproducible synthesis and defence |

## 3. Hard dependency chains

These chains are used by the mastery tracker to prevent superficially advanced progress from hiding a foundational gap.

```text
units/algebra -> stoichiometry -> solutions -> equilibrium/pH -> amino-acid charge -> enzyme/metabolic regulation

bonding/IMFs -> membrane chemistry -> transport/gradients -> membrane potential -> neural/cardiovascular/renal physiology

cell division -> meiosis -> heredity -> DNA information flow -> molecular methods -> biochemical regulation

energy/forces -> fluids/pressure -> cardiovascular/respiratory models -> integrated homeostasis

waves/optics -> spectroscopy/microscopy -> structure evidence -> research-figure appraisal

probability -> genetic inference -> diagnostic test reasoning -> evidence communication

experimental design -> practical controls -> causal inference -> integration studios -> capstone
```

“Hard” means later assessment assumes the earlier representation or operation. Editors can introduce motivating previews before mastery, but a learner cannot earn downstream `M` while the hard dependency remains below `P`.

## 4. Cross-domain integration edges

| From | To | Relationship learners must demonstrate |
|---|---|---|
| `CHE-122` intermolecular forces | `BIO-114`, `BCH-223` membranes | molecular interaction explains assembly, fluidity, and permeability |
| `CHE-133` solutions | `BIO-115`, `MED-318` | concentration/osmolarity models predict water and solute movement |
| `CHE-144/145` pH/buffers | `BCH-211/214`, `MED-317/318` | proton state alters proteins; organ systems contribute to acid-base regulation |
| `CHE-136` free energy | `BIO-121`, `BCH-231` | spontaneity, coupling, and cellular work share one thermodynamic account |
| `CHE-146` redox | `BCH-231/235` | electron potentials and carriers underlie respiratory energy conversion |
| `BIO-114` gradients | `PHY-143`, `MED-313` | membranes combine chemical and electrical driving forces and capacitance |
| `BIO-122` enzymes | `BCH-214/215` | qualitative rate/regulation becomes quantitative biochemical evidence |
| `BIO-126` meiosis | `BIO-131/133` | chromosome behavior produces inheritance patterns and aneuploidy |
| `BIO-143/144` expression | `BCH-241/244` | information processing and regulation link variant to cellular phenotype |
| `PHY-115` torque | `MED-315` | mechanical advantage and force apply to muscles and joints with model limits |
| `PHY-123` fluid resistance | `MED-316/317` | radius, pressure, viscosity, and branching constrain transport models |
| `PHY-133` sound | `MED-313`, pathway behaviour | waves become sensory transduction only after physiological processing |
| `PHY-134/135` optics | `BIO-117`, `ORG-233` | instrument evidence depends on light-matter interaction and resolution |
| `PHY-153/154` radiation | `MED` imaging contexts | physical signal/dose must be separated from clinical interpretation |
| `ORG-214` stereochemistry | `BCH-211/221` | three-dimensional molecular identity affects biochemical recognition |
| `ORG-224/225` mechanisms | `BCH-221/222/232` | biomolecular bonds and metabolic reactions are organic chemistry in context |
| `BCH-213` oxygen binding | `MED-316/317` | molecular binding, transport, partial pressure, and flow jointly determine delivery |
| `BCH-230` metabolism | `MED-320` | tissue/organ context regulates fuel selection rather than merely listing pathways |
| `BIO-150` evolution | `MED-334` | pathogen/resistance change is population evolution, not individual adaptation |
| `QRS-107` Bayes reasoning | `MED-336` | test meaning depends on pretest prevalence and conditional probabilities |
| `QRS-105/108` design/inference | every bridge case | the strength of a health claim depends on design and uncertainty, not plausibility alone |

## 5. Program outcome coverage

`I` = introduced, `D` = developed, `M` = principal mastery evidence, `T` = transfer evidence.

| PLO | Primary modules | Coverage progression |
|---|---|---|
| `PLO-01` molecular interactions | `CHE-110`–`140`, `ORG-210`, `BCH-220` | I → D → M → T in `INT-301/304` |
| `PLO-02` cells as systems | `BIO-110/120/150`, `BCH-240` | I → D → M → T in `MED-330` |
| `PLO-03` genetic information | `BIO-130/140`, `BCH-240` | I → D/M → T in `INT-304` or capstone |
| `PLO-04` enzymes/metabolism | `BIO-120`, `BCH-210/230` | I → D → M → T in `MED-320` |
| `PLO-05` organic chemistry | `ORG-210/220/230`, `BCH-210/220` | I → D/M → T in `INT-304` |
| `PLO-06` physics | `PHY-110`–`150` | I/D/M across sequence → T in `INT-303` |
| `PLO-07` homeostasis | `MED-310/320` | I/D/M → T in `INT-305` |
| `PLO-08` immunity/biotech | `BIO-140/150`, `MED-330` | I → D → M/T |
| `PLO-09` study design | `QRS-105`, every lab, `CAP-401` | I → D → M in `LAB-14` → T in capstone |
| `PLO-10` quantitative reasoning | `QRS-100`, all science modules | I/D distributed → M in QRS gate → T in capstone |
| `PLO-11` representations | all modules, `ORG-230` | I/D distributed → M in gate portfolio → T in studios |
| `PLO-12` evidence quality | `QRS-108/110/111`, labs | I/D → M in paper portfolio → T in capstone |
| `PLO-13` reproducibility | `QRS-112`, labs, capstone | I → D → M/T in capstone package |
| `PLO-14` integration | `INT-300`, bridge, capstone | I/D in studios → M/T in capstone |
| `PLO-15` paper reading | `QRS-111`, six appraisals | I → D → M in portfolio → T in capstone |
| `PLO-16` communication | every gate, studios, capstone | I/D distributed → M/T in dual-audience capstone |
| `PLO-17` advice boundary | bridge cases and public artefacts | I/D → M in `MED-330` → T in capstone |
| `PLO-18` self-regulation | `ONR-007`, weekly cycle, remediation | I → D/M in error log → T in final maintenance plan |

## 6. Practical-to-content crosswalk

| Lab family | Primary content nodes | QRS nodes | Downstream use |
|---|---|---|---|
| `LAB-01` calibration/density | `CHE-112`, `PHY` measurements | `QRS-101/104/109` | every quantitative practical |
| `LAB-02` solution/reaction | `CHE-124/125` | `QRS-101/112` | `CHE-133`, biochemistry standards |
| `LAB-03` calorimetry | `CHE-135`, `PHY-151` | `QRS-103/104` | energy accounting |
| `LAB-04` kinetics/equilibrium | `CHE-141/142`, `BIO-122` | `QRS-103/105/108` | enzyme/metabolism interpretation |
| `LAB-05` acid/base | `CHE-143`–`145` | `QRS-102/104/109` | protein charge, physiology |
| `LAB-06` membranes | `BIO-114/115`, `BCH-223` | `QRS-105/108` | renal/neural homeostasis |
| `LAB-07` microscopy | `BIO-113/117` | `QRS-104/109/112` | tissue/histology evidence |
| `LAB-08` enzyme | `BIO-122`, `BCH-214/215` | `QRS-103/105/108` | metabolic regulation |
| `LAB-09` molecular/genetic | `BIO-131/146` | `QRS-107/108/112` | biotech/immunity |
| `LAB-10` mechanics | `PHY-111`–`115` | `QRS-101/103/104` | musculoskeletal bridge |
| `LAB-11` physical system | `PHY-121`–`145` | `QRS-102/104/110` | organ transport/instrumentation |
| `LAB-12` separation/spectra | `ORG-231`–`236`, `BCH-216` | `QRS-109/111` | molecular evidence |
| `LAB-13` physiology | `MED-310/320` | `QRS-104/105/108`, privacy | integrated homeostasis |
| `LAB-14` inquiry | learner-selected | all `QRS` | capstone readiness |

## 7. Assessment minimum map

Every required topic must be sampled at least once independently and once after spacing. For content production, each topic-node package requires at minimum:

- 8 retrieval prompts: 3 direct, 2 discrimination, 2 representation translation, 1 misconception check;
- 4 worked examples with faded support;
- 6 independent problems: 2 familiar, 2 varied, 1 data/design, 1 transfer;
- 2 delayed variants that cannot be solved by remembering surface wording;
- 1 gate item family with at least 3 equivalent forms;
- 1 explanation rubric anchor at mastery level and 1 authentic non-example;
- practical linkage where material, with a safe non-physical alternative;
- 1 downstream case or explicit statement that the node is terminal within Premed.

The item bank should not be auto-generated into production without content review, answer verification, accessibility review, and form-difficulty monitoring.

## 8. Editorial production order

Build in dependency slices so early learners can begin studying before the whole program exists:

1. `QRS-101`–`104`, `CHE-110`, `BIO-110`, `INT-301`.
2. `CHE-120`, `BIO-120`, `PHY-110`, Labs 1/2/6/8/10.
3. `CHE-130/140`, `BIO-130/140`, `PHY-120/130`, `INT-302/303`.
4. `BIO-150`, `PHY-140/150`, `ORG-210/220`.
5. `BCH-210/220`, `ORG-230`, `BCH-230/240`, `INT-304`.
6. `MED-310/320/330`, `INT-305`.
7. Highest-priority learner pathway, then `INT-306` and `CAP-400`.

Each slice ships only after its gates, source ledger, progress events, and remediation branches work end to end. A visible graph with empty content is a planning artefact; it is not a teachable module.

## 9. Definition of curriculum-complete

The syllabus skeleton is curriculum-complete when every required module and pathway has outcomes, order, depth, exclusions, assessment, workload, and source alignment. An implemented course is complete only when every required topic satisfies the node contract, all gates and practical alternatives pass review, learner progress can be exported, and a pilot cohort has produced evidence that the prerequisite and remediation paths function.


---

## Repository file: course/PREMED-KNOWLEDGE-GRAPH.md

# Premed Knowledge Graph

**Version:** 1.1 candidate · 20 July 2026
**Dataset:** [`site/data/premed-graph.json`](../site/data/premed-graph.json)
**Canonical builder:** [`scripts/build-premed-graph.mjs`](../scripts/build-premed-graph.mjs)
**Validator:** [`scripts/validate-premed-graph.mjs`](../scripts/validate-premed-graph.mjs)

> **Boundary.** This is educational preparation only. It is **not** a degree, academic credit, admission guarantee, medical qualification, or medical advice. Enumerating a medical curriculum does not confer clinical authority; nothing here should be used to make a decision about a real patient, including yourself.

The Premed knowledge graph is the machine-readable companion to [`PREMED-SYLLABUS.md`](PREMED-SYLLABUS.md). It turns the syllabus into navigable domains, fine-grained modules, observable topic outcomes, prerequisites, conceptual cross-links, route labels, workload estimates, and evidence tags.

It is deliberately more granular than the teaching syllabus. The syllabus's stable IDs (`CHE-110`, `BIO-120`, `MED-310`, and so on) identify assessable teaching modules; graph IDs such as `module-acids-bases` and `topic-acids-bases-buffers` identify concepts that can be visualized and tracked independently. Every graph module and topic carries `syllabusModuleIds` to connect the two layers.

## 1. What the graph contains

| Layer | Count | Purpose |
|---|---:|---|
| Domains | 10 | Broad visual and curricular regions |
| Modules | 46 | Coherent concept clusters, finer than syllabus modules |
| Topics | 388 | Observable learning outcomes and mastery units |
| All nodes | 444 | Domains + modules + topics |
| Hierarchy links | 434 | Domain → module and module → topic containment |
| Prerequisite links | 592 | Directed learning dependencies |
| Cross-links | 34 | High-value conceptual connections that do not impose order |
| All links | 1,060 | The complete interactive graph |

The sum of every topic estimate is **1,551 hours**. That number is an all-branches graph inventory, not a completion promise: it includes the welcome, every route-specific topic, broad biology, calculus, and the entire medicine bridge. A learner follows a selected subgraph. The syllabus defines the official workload and completion rules; the graph's `estimatedHours` is for planning, progress, and relative scope.

Coverage by requirement label:

| Requirement | Topics | Graph hours | Meaning |
|---|---:|---:|---|
| `portable-core` | 213 | 810.5 | Course welcome plus the high-confidence cross-system academic spine |
| `medicine-bridge` | 127 | 560 | Deliberate EmbeddedKnowledge bridge toward medical science; not a universal entrant minimum |
| `pathway` | 48 | 180.5 | Route-specific or broad-literacy content |

## 2. Data contract

The top-level object is stable at schema version `2.0.0`:

```json
{
  "schemaVersion": "2.0.0",
  "course": {},
  "schema": {},
  "pathways": [],
  "sources": [],
  "nodes": [],
  "links": [],
  "metrics": {}
}
```

The landing page can depend on the deliberately small common surface:

```json
{
  "nodes": [
    {
      "id": "topic-acids-bases-buffers",
      "title": "Buffers",
      "domain": "domain-general-chemistry",
      "summary": "Explain buffer action and calculate composition, pH, capacity, and response to added acid or base.",
      "level": "college-intro",
      "module": "module-acids-bases",
      "estimatedHours": 5,
      "pathway": ["portable-core", "us-mcat"]
    }
  ],
  "links": [
    {
      "source": "topic-acids-bases-weak-acid-base",
      "target": "topic-acids-bases-buffers",
      "type": "prerequisite"
    }
  ]
}
```

### Node kinds

- `domain` is a visual grouping. A domain contains modules but carries no mastery state of its own.
- `module` is a graph-scale concept cluster. It has a stable `PREM-##` code and one or more mappings to the syllabus.
- `topic` is the atomic mastery unit. Its `summary` and `outcome` are the same observable learning statement in UI-friendly and curriculum-friendly fields.

All nodes expose `id`, `kind`, `title`, `label`, `summary`, `domain`, `subject`, `estimatedHours`, and `pathway`. Modules and topics also expose:

| Field | Meaning |
|---|---|
| `code` | Human-readable stable code; topic codes extend their graph module code |
| `module` / `moduleId` | Parent module ID for topics; self ID for modules |
| `syllabusModuleIds` | Corresponding assessable module(s) in `PREMED-SYLLABUS.md` |
| `level` | `foundation`, `secondary`, `college-intro`, `first-semester`, or `bridge` |
| `requirement` | `portable-core`, `medicine-bridge`, `pathway`, `on-ramp`, or `enrichment` |
| `core` | Convenience boolean; true only for `portable-core` |
| `evidenceConfidence` | Confidence in curricular placement, not confidence that every institution requires it |
| `sourceTags` | Keys into the top-level source registry |
| `prerequisites` | Ordered-learning dependency IDs, mirrored as links |
| `crossLinks` | Related topic IDs, mirrored as symmetric semantic links |

### Link semantics

| Type | Direction | Interpretation |
|---|---|---|
| `contains` | Parent → child | Domain contains module, or module contains topic |
| `prerequisite` | Prerequisite → dependent | The source should normally be mastered before the target |
| `cross-link` | Canonicalized but conceptually undirected | The concepts illuminate one another; no study order is implied |

Prerequisite links form a directed acyclic graph. Cross-links may close loops and are intentionally excluded from topological ordering.

## 3. Identity and versioning rules

IDs are lowercase kebab-case and describe curricular meaning, not screen position. Once published, an ID must not be silently reused for another concept.

1. A wording improvement keeps the existing ID.
2. A concept split creates new child IDs and records the retired ID in a migration note.
3. A concept merge chooses one surviving ID only when the learning outcome is materially identical; otherwise it creates a new ID.
4. Renaming a domain or moving a topic between modules does not require changing the topic ID unless the ID itself becomes false.
5. Breaking structural or semantic changes increment `schemaVersion`; ordinary topic additions increment the curriculum version, not the JSON schema.
6. Mastery records should key on topic `id`, never array index, title, or visual coordinates.

The builder is canonical because it keeps compact declarations, derived hierarchy, aliases, totals, and links consistent. Do not hand-edit the generated JSON. Edit the builder, regenerate, and validate.

```bash
node scripts/build-premed-graph.mjs
node scripts/validate-premed-graph.mjs
```

## 4. Domain and syllabus mapping

The graph's 46 modules give the interface enough resolution to show meaningful dependencies. They map onto the syllabus's larger teaching blocks as follows.

| Graph code | Graph module | Syllabus module(s) |
|---|---|---|
| `PREM-01` | Welcome to Premed | `WEL-000` |
| `PREM-02` | Scientific Inquiry | `QRS-100` |
| `PREM-03` | Quantitative Foundations | `QRS-100` |
| `PREM-04` | Statistics & Data | `QRS-100` |
| `PREM-05` | Cell Biology | `BIO-110`, `BIO-120` |
| `PREM-06` | Molecular Genetics | `BIO-130`, `BIO-140`, `BCH-240` |
| `PREM-07` | Evolution | `BIO-150`, `PW-BIO` |
| `PREM-08` | Ecology & Plant Biology | `PW-BIO` |
| `PREM-09` | Biomolecules | `BIO-110`, `BCH-210`, `BCH-220` |
| `PREM-10` | Enzymes | `BIO-120`, `BCH-210` |
| `PREM-11` | Metabolism & Bioenergetics | `BIO-120`, `BCH-230`, `BCH-240` |
| `PREM-12` | Atomic Structure & Periodicity | `CHE-110` |
| `PREM-13` | Bonding & Molecular Structure | `CHE-120` |
| `PREM-14` | Stoichiometry & Reactions | `CHE-120` |
| `PREM-15` | States, Gases & Solutions | `CHE-130` |
| `PREM-16` | Thermodynamics, Kinetics & Equilibrium | `CHE-130`, `CHE-140` |
| `PREM-17` | Acids, Bases & Buffers | `CHE-140` |
| `PREM-18` | Redox & Electrochemistry | `CHE-140` |
| `PREM-19` | Organic Foundations | `ORG-210` |
| `PREM-20` | Organic Reactivity | `ORG-220` |
| `PREM-21` | Carbonyls & Biological Organic Chemistry | `ORG-220`, `BCH-220` |
| `PREM-22` | Separation & Structure Analysis | `ORG-230` |
| `PREM-23` | Laboratory Practice | `QRS-100`, `INT-300` |
| `PREM-24` | Mechanics | `PHY-110` |
| `PREM-25` | Fluids | `PHY-120` |
| `PREM-26` | Thermal Physics & Gases | `PHY-150` |
| `PREM-27` | Electricity & Magnetism | `PHY-140` |
| `PREM-28` | Waves & Sound | `PHY-130` |
| `PREM-29` | Optics | `PHY-130` |
| `PREM-30` | Atomic & Nuclear Physics | `PHY-150` |
| `PREM-31` | Calculus Extension | `PW-QNT` |
| `PREM-32` | Anatomy, Homeostasis & Tissues | `MED-310` |
| `PREM-33` | Nervous System | `MED-310` |
| `PREM-34` | Endocrine System | `MED-310`, `MED-320` |
| `PREM-35` | Cardiovascular System | `MED-310` |
| `PREM-36` | Respiratory System | `MED-310` |
| `PREM-37` | Renal System | `MED-310` |
| `PREM-38` | Digestive System | `MED-320` |
| `PREM-39` | Musculoskeletal System | `MED-310`, `MED-320` |
| `PREM-40` | Reproduction & Development | `MED-320` |
| `PREM-41` | Immunity | `MED-330` |
| `PREM-42` | Microbiology | `BIO-150`, `MED-330` |
| `PREM-43` | Psychology | `PW-US`, `INT-300` |
| `PREM-44` | Sociology & Health | `PW-US`, `INT-300` |
| `PREM-45` | Critical Reading & Argument | `PW-US`, `QRS-100` |
| `PREM-46` | Integrative Capstones | `INT-300`, `CAP-400` |

This is a many-to-many mapping by design. For example, the graph's `module-molecular-genetics` spans heredity, molecular control, and biochemical information flow, which the teaching sequence revisits in `BIO-130`, `BIO-140`, and `BCH-240`.

### 4.1 Version 1.1 opening-sequence migration

The version 1.1 candidate retires `module-orientation`, `module-study-research-literacy`, `module-professional-behaviours`, and their 17 topic IDs. No production lesson or repository learner-mastery record used those IDs when they were retired. The focused `module-welcome` and `topic-welcome-course-journey` now precede `module-scientific-inquiry`; Scientific Inquiry is the first instructional module. Because `PREM-##` is a human-readable display sequence rather than the mastery key, retained modules receive new sequential codes while their stable `module-*` and `topic-*` IDs remain unchanged. Removing published graph identities is a breaking semantic change, so the graph schema version advances to `2.0.0` even though the JSON field structure is unchanged.

## 5. Pathways and honest scope

Pathway labels are filters, not admissions claims. A topic can belong to more than one pathway.

### 5.1 The route resolution rule

There is exactly one rule for deciding which topics a route contains:

> **A route's effective topic set is the union of the topics carrying that route tag and every `portable-core` topic.** `portable-core` resolves to exactly its own tagged set.

Two corollaries follow, and both are load-bearing:

1. **`sourceTags` never determine route membership.** They record which evidence families justify a topic's placement. A topic can carry `AAMC-MCAT` because the MCAT outline covers that material while its `pathway` array omits `us-mcat` because it already arrives through `portable-core` — 101 of the 388 topics are in exactly that position, including all of `PREM-12` Atomic Structure & Periodicity. Filtering on source tags instead of the rule above would hide general chemistry from an MCAT learner. Conversely, the ten `PREM-03` Quantitative Foundations topics sit in `us-mcat` under the `AAMC-MATH` source rather than `AAMC-MCAT`; a missing `AAMC-MCAT` tag is not an exclusion. One topic — `topic-organic-reactivity-multistep-synthesis` — inherits its module's `AAMC-MCAT` tag but is deliberately excluded from `us-mcat`, because extended synthesis is outside the MCAT; the rule above resolves this correctly and source tags must not override it.
2. The builder publishes the rule and its results so no consumer has to re-derive them: `schema.routeResolutionRule`, `pathways[].includesPortableCore`, and `metrics.routeTopicCounts` (with `tagged`, `effective`, and `estimatedHours` per route).

### 5.2 Route table

| Graph pathway | Syllabus route | Scope rule |
|---|---|---|
| `portable-core` | Common core | Adjudicated high-confidence academic spine; not a claim that every school legally requires every item |
| `medicine-bridge` | `MED-310`–`MED-330`, `INT-300` | EmbeddedKnowledge design choice for readiness; anatomy and full organ integration are not universal entrant requirements |
| `us-mcat` | `PW-US` | AAMC MCAT content and reasoning alignment; individual medical schools retain their own prerequisites |
| `uk-direct-entry` | `PW-UK` (**tag only — route not instrumented**) | A-level and Access orientation; UK subject rules remain institution-specific |
| `spain-bach-pau` | `PW-ES` (**tag only — route not instrumented**) | Bachillerato science coverage and PAU orientation; weightings remain annual and university-specific |
| `continental-science` | `PW-ES` (**tag only — route not instrumented**) | French/German and related secondary-science extensions; no claim that all entrants took every science |
| `access-foundation` | `ONR-000`, `PW-UK` | Novice entry, scientific inquiry, laboratory literacy, and numeracy |
| `broad-biology` | `PW-BIO` | Evolution, ecology, and plant science retained for biological literacy |
| `advanced-quantitative` | `PW-QNT` | Calculus and mathematical extensions explicitly outside the MCAT minimum |

> **Instrumentation limitation (curriculum version 1.1).** The syllabus defines five routes; only `PW-US`, `PW-BIO` and `PW-QNT` are instrumented here. **`PW-UK` and `PW-ES` are documented routes that are NOT yet instrumented in the v1 graph.** No graph module maps to them, and their route-defining outcomes — UCAT familiarization, the QAA Access baseline, the Spanish admission audit, and the open-response studio — are not among the 388 atomic outcomes. Consequently they cannot yet satisfy the syllabus section 6.2 pathway-completion requirement, and contributors cannot yet author lessons against them. The `uk-direct-entry` and `spain-bach-pau` **pathway tags below are content filters** marking A-level/Access-shaped and Bachillerato-shaped material inside other modules; they are not implementations of `PW-UK` or `PW-ES`. Instrumenting both routes is planned for a later curriculum version.

Important adjudicated boundaries are represented directly:

- MCAT-aligned evolution is present, but ecology, ecosystems, plant biology, and photosynthesis are **not** tagged as MCAT content.
- Photosynthesis is the **only** topic of `PREM-08` Ecology & Plant Biology carrying `spain-bach-pau` and the `RD243` source, because [RD 243/2022](https://www.boe.es/buscar/act.php?id=BOE-A-2022-5521) specifies autotrophic anabolism while Spain's six second-year Biology blocks contain no ecology block and no broad plant unit (see `research/ADJUDICATION-2026-07-19.md` section 1.8). The module's other six topics carry only `broad-biology` and `uk-direct-entry` with `AQA-BIO`/`AP-BIO` evidence.
- Human physiology is a medicine-facing bridge and MCAT-aligned survey, not mislabeled as universal upper-level physiology or a universal entrant prerequisite.
- MCAT metabolism includes glycolysis, gluconeogenesis, pentose phosphate, the citric acid cycle, fatty-acid oxidation, ketone bodies, and oxidative phosphorylation. Nitrogen-disposal principles appear as biochemistry, but the graph does not claim the urea cycle is an explicit MCAT leaf topic.
- Statistics and experimental reasoning are core. Calculus is a separate optional path and is explicitly excluded from the MCAT mathematical floor.
- Organic functional groups and biological reactivity are portable core; extended mechanisms, synthesis, and spectroscopy carry pathway or medicine-bridge scope where appropriate.

## 6. Source and evidence model

`sourceTags` answer, “Which evidence families justify this topic's placement or provide an aligned learning resource?” They do not mean that every source contains the exact wording of the topic, and a learning-resource tag is not normative authority.

The registry separates:

- `primary-normative` and `primary-normative-directory`: requirements, legal curricula, or institution-level directories;
- `primary-descriptive`: test content and assessed competency outlines;
- `primary-competency`: broad readiness competencies;
- `primary-secondary`: upper-secondary specifications;
- `learning-resource`: teachable materials, whose licensing must be checked per title and edition;
- `local-analysis`: the dated adjudication that reconciles conflicting reports.

`evidenceConfidence` concerns curricular placement:

- `high`: directly in the adjudicated portable core;
- `medium-high`: strongly medicine-facing or repeatedly supported, but not globally universal;
- `medium`: pathway placement or breadth choice whose exact depth varies.

It must never be rendered as “probability this fact is true” or “percentage of medical schools requiring this topic.” Requirements change, and US, UK, Spanish, French, German, IB, and AP structures are not interchangeable.

## 7. Recommended use in the product

### Learner view

1. Begin with a diagnostic and select one route.
2. Resolve the route's topic set with the rule in section 5.1 — route tag **union** `portable-core` — and hide only what falls outside that set unless the learner asks to explore. Never filter by `sourceTags`.
3. Topologically sort unmet prerequisite links, then rank available nodes by the learner's study plan and spaced-review needs.
4. Roll topic mastery up to modules and domains for display only; never let an average conceal an unmastered safety-critical or prerequisite topic.
5. Render `cross-link` edges on demand to prevent a 1,058-edge hairball.
6. Show `requirement`, `evidenceConfidence`, and `sourceTags` in detail views rather than implying that all visible content is compulsory.

### Suggested graph states

- `locked`: an essential prerequisite is not yet demonstrated;
- `available`: prerequisites are met and no evidence exists yet;
- `acquiring`, `practising`, `mastered`, `retained`, `transfer`: use the mastery language in the syllabus;
- `review-due`: previously demonstrated but scheduled for retrieval;
- `route-hidden`: outside the learner's selected route, still searchable;
- `needs-correction`: source or content issue flagged for review.

The static graph contains curriculum facts only. Learner state, confidence, last-touched time, review schedule, attempts, and accommodations belong in a separate per-user record keyed by topic ID.

## 8. Integrity guarantees and current validation

The validator checks:

- valid JSON and required node fields;
- unique node and link IDs;
- valid source tags and pathway IDs;
- valid endpoints for all 1,058 links;
- exact agreement between topic `prerequisites` arrays and prerequisite links;
- topic-only prerequisite endpoints;
- absence of prerequisite self-links and cycles;
- consistency of topic, module, and course workload totals.

Current result:

```text
Premed graph valid: 444 nodes, 1058 links, 388 topics, 1551 estimated hours.
```

The validator intentionally does not forbid cycles made only of `cross-link` edges, because those encode reciprocal conceptual relationships rather than study order.

## 9. Maintainer checklist

Before publishing a graph revision:

1. Confirm the change is within the syllabus promise and label medicine-bridge or pathway scope honestly.
2. Attach at least the adjudication and the strongest applicable primary source tag.
3. Add prerequisites only when earlier mastery materially reduces conceptual load; do not encode mere teaching preference as dependency.
4. Use cross-links sparingly for relationships worth surfacing in an interactive explanation or transfer task.
5. Check source URLs and curriculum volatility, especially annual MSAR, UK institution rules, and Spanish university weightings.
6. Regenerate the JSON and run the validator.
7. Review the affected route subgraph visually and verify that the landing page still degrades cleanly without coordinates.
8. Add an ID migration note when retiring or splitting published topics so mastery history remains interpretable.


---

## Repository file: course/PSYCHIATRY-SYLLABUS.md

# Psychiatry: Mind, Brain, Person, Society

**Curriculum version:** 0.2 graph-backed academic curriculum · 23 July 2026<br>
**Provider:** EmbeddedKnowledge<br>
**Language:** English<br>
**Level:** postgraduate medical knowledge; residency-aligned academic study with a separate supervised-practice crosswalk<br>
**Format:** mastery-based, online-first, case- and evidence-centred; optional tutorials and simulation; real clinical work only through an independently accredited and licensed host<br>
**Nominal workload:** 1,440 academic hours for the common curriculum, plus 120–240 hours for one area of depth<br>
**Typical pace:** four academic stages over 24–48 months; no fixed expiry<br>
**Credential:** an EmbeddedKnowledge record of demonstrated academic outcomes, **not** a medical degree, residency, board eligibility, continuing-professional-development credit, clinical placement, licence, specialist title, or authorization to diagnose or treat

> **Course promise:** finishers should be able to reason about psychiatric knowledge as an integrated study of mind, brain, person, relationship, culture, and system; appraise evidence and uncertainty; and explain how safe, humane, collaborative psychiatric care is designed. Clinical competence requires supervised work with patients in an accredited program and cannot be earned from this book.

## 1. Place in the library

Psychiatry is EmbeddedKnowledge's second book and its first postgraduate, clinically adjacent curriculum. It is a modern academic syllabus, not a universal residency prescription. It synthesizes durable common ground across current United States, United Kingdom, Canadian, Australian and Aotearoa New Zealand, European, and global frameworks while keeping jurisdiction-specific law, certification, service organization, and scope of practice visible.

The design is intentionally broader than a diagnostic manual or board-review outline. Neuroscience, medicine, phenomenology, formulation, psychotherapy, pharmacotherapy, recovery, public mental health, human rights, culture, evidence, and systems practice remain co-equal strands. Classification is taught as one useful tool inside a larger clinical argument, not as the definition of psychiatry.

The syllabus has a public knowledge graph and focused production-lesson intake. Its lesson set, assessment bank, optional area-of-depth graph, and clinical-host standard grow only through separate reviewed pull requests. Graph availability does not create clinical authority.

## 2. Audience and routes

The common academic curriculum supports three routes without pretending they are equivalent:

| Route | Intended learner | What EmbeddedKnowledge can evidence | What remains outside this course |
|---|---|---|---|
| Open scholar | graduates and advanced learners studying psychiatry without seeking clinical authority | knowledge, case reasoning with fictional material, evidence appraisal, communication, reflection, and capstone work | patient care, licensure, protected titles, prescribing, detention decisions, and professional credentialing |
| Physician learner | medical graduates, residents, and practising clinicians using the book alongside a recognized program | the same academic outcomes, plus a portfolio that a host may choose to inspect | the host alone supervises patients, judges clinical entrustment, verifies identity, and awards any formal credit or progression |
| Educator or system learner | teachers, researchers, peer specialists, service designers, and health leaders | curriculum mapping, evidence synthesis, quality-improvement simulation, teaching and service-design work | local employment privileges, regulatory authority, and independent clinical practice |

People with lived and living experience are not treated as teaching objects. They may participate as learners, paid educators, authors, reviewers, or governance partners under explicit role, consent, accessibility, attribution, compensation, withdrawal, and support agreements.

## 3. Entry and on-ramp

### 3.1 Expected entry capabilities

The normal physician-learner entry point is a medical degree or the biomedical, behavioural, and research-methods knowledge expected at graduation from medical school. Open scholars may enter without a medical degree, but should be able to:

- explain introductory neuroanatomy, physiology, pharmacology, genetics, and general pathology;
- take apart a research paper's question, population, design, main estimate, uncertainty, and limitations;
- reason with probability, diagnostic-test performance, risk, and absolute as well as relative effects;
- communicate about distress, disability, identity, and disagreement without stigmatizing language;
- distinguish education, simulation, supervision, and licensed clinical practice;
- protect personal and health information in notes, discussion, collaboration, and tool use.

### 3.2 Diagnostic and bridge `PSY-ONR-000` (0–120 hours, as needed)

The untimed diagnostic samples six prerequisite domains. It produces a concept-level study prescription, never a clinical judgment about the learner or another person.

| Bridge | Scope | Exit evidence |
|---|---|---|
| `PSY-ONR-001` Biomedical bridge | nervous-system organization, endocrine and immune signalling, pharmacokinetics, pharmacodynamics, common medical illness | mechanism-and-mimic case map |
| `PSY-ONR-002` Behavioural science bridge | learning, cognition, emotion, development, relationships, social determinants | multi-level explanation of a fictional presentation |
| `PSY-ONR-003` Evidence bridge | study designs, bias, causal inference, effect measures, diagnostic accuracy, qualitative evidence | structured appraisal of one study |
| `PSY-ONR-004` Communication bridge | person-first and identity-respecting language, listening, interpreters, uncertainty, boundaries | recorded standardized-person conversation and self-critique |
| `PSY-ONR-005` Ethics and rights bridge | autonomy, capacity, consent, confidentiality, disability and human-rights models | jurisdiction-labelled ethics analysis |
| `PSY-ONR-006` Learning and safety bridge | retrieval, feedback, reflective practice, trauma-aware participation, privacy, crisis boundaries | personal learning and support plan |

Bridge completion is support, not admission or evidence of medical training.

## 4. Curriculum commitments

1. **The person exceeds the diagnosis.** Symptoms, strengths, goals, development, relationships, body, culture, spirituality, material conditions, and power belong in the formulation.
2. **Lived experience changes the curriculum.** Paid contributors with lived and living experience help set outcomes, author cases, teach, review language, and evaluate shared decision-making. Consultation after decisions are made is not co-production.
3. **Medicine and meaning stay connected.** Biological mechanisms and medical mimics are taught alongside phenomenology, narrative, psychology, and social context; no single level is treated as a complete explanation.
4. **Uncertainty is an assessed capability.** Learners calibrate confidence, compare alternatives, state missing evidence, update formulations, and name when a classification or intervention may harm.
5. **Safety includes rights.** Immediate protection, therapeutic alliance, supported decision-making, least-restrictive care, safeguarding, staff safety, disclosure, and learning after harm are taught together.
6. **Treatment is collaborative and longitudinal.** Benefits, burdens, alternatives, interactions, monitoring, withdrawal, access, preference, and recovery goals accompany every intervention.
7. **Psychotherapy is core clinical science.** Relationship, formulation, common factors, model-specific methods, supervision, process, outcome, and adverse effects recur throughout the course.
8. **Culture is not a module-sized add-on.** Language, racism, migration, Indigeneity, gender, sexuality, disability, poverty, housing, religion, family, and local knowledge shape every clinical question.
9. **Assessment samples performance.** Retrieval supports learning, but progression depends on observed reasoning, communication, transfer, feedback uptake, and work across cases and contexts.
10. **Open does not mean unsupervised.** The book can teach knowledge and simulation. Real-patient assessment, prescribing, psychotherapy, compulsory care, and procedures require an authorized host.
11. **Sources and rights outrank convenience.** Every production lesson must support material claims, track version and jurisdiction, and clear text, instrument, case, image, and data rights.
12. **The curriculum remains corrigible.** Diagnostic categories, treatment evidence, law, technology, and preferred language change. Dated audits and reversible versioning are part of the design.

## 5. Graduate capabilities

By academic completion, a learner can:

### Person-centred assessment and reasoning

1. `PSY-PLO-01` Establish a respectful, boundaried, trauma-aware conversation with a standardized person and adapt it for language, disability, development, culture, and setting.
2. `PSY-PLO-02` Elicit and organize a psychiatric, medical, neurologic, developmental, medication, substance, family, social, cultural, spiritual, and functional history using fictional or consented simulation material.
3. `PSY-PLO-03` Describe a mental-status examination as time-bound observation, distinguish observation from inference, and communicate uncertainty without dehumanizing shorthand.
4. `PSY-PLO-04` Produce and revise a concise formulation that integrates predisposing, precipitating, perpetuating, protective, biological, psychological, relational, cultural, and structural factors.
5. `PSY-PLO-05` Construct a prioritized differential that includes medical, neurologic, substance-, medication-, sleep-, developmental-, and context-related explanations and states what evidence would change it.
6. `PSY-PLO-06` Use dimensional measures and classification systems critically, respecting intended use, validity, copyright, version, culture, and the difference between a code and a person.

### Safety, ethics, and partnership

7. `PSY-PLO-07` Reason through fictional urgent presentations involving self-harm, suicide, violence, abuse, exploitation, intoxication, withdrawal, delirium, catatonia, agitation, or severe medication effects and identify when emergency escalation is required.
8. `PSY-PLO-08` Explain capacity, consent, confidentiality, safeguarding, supported decision-making, advance planning, and least-restrictive practice while labelling jurisdiction-specific law.
9. `PSY-PLO-09` Build a collaborative safety and recovery plan in simulation that includes the person's priorities, strengths, supporters, access barriers, warning signs, and local crisis pathways.
10. `PSY-PLO-10` Recognize diagnostic overshadowing, coercion, discrimination, epistemic injustice, conflicts of interest, and iatrogenic harm and propose accountable repair.

### Treatment and longitudinal care

11. `PSY-PLO-11` Compare reasonable intervention options by indication, mechanism, expected benefit, uncertainty, burden, contraindication, interaction, monitoring, access, preference, and stopping plan.
12. `PSY-PLO-12` Explain the principles and evidence of psychopharmacology, deprescribing, psychotherapy, social interventions, peer support, rehabilitation, physical-health care, and interventional treatments without issuing personal treatment advice.
13. `PSY-PLO-13` Build a measurement-informed longitudinal plan that tracks symptoms, function, quality of life, physical health, adverse effects, alliance, preference, equity, and recovery rather than a symptom score alone.
14. `PSY-PLO-14` Explain psychotherapy formulation and common factors; compare supportive, cognitive-behavioural, psychodynamic, interpersonal, dialectical-behavioural, family, group, and motivational approaches at a level appropriate to academic study.
15. `PSY-PLO-15` Coordinate a fictional care pathway across primary care, emergency, inpatient, outpatient, community, rehabilitation, housing, education, employment, peer, and specialist services.

### Scholarship, systems, and professional formation

16. `PSY-PLO-16` Appraise diagnostic, prognostic, therapeutic, qualitative, implementation, and harms evidence; detect spin, selective outcome reporting, confounding, and clinically trivial effects.
17. `PSY-PLO-17` Design an ethical quality-improvement or service-evaluation proposal with stakeholder governance, balancing measures, equity analysis, privacy, and an explicit plan to stop ineffective practice.
18. `PSY-PLO-18` Evaluate telepsychiatry, clinical software, generative AI, predictive models, and digital phenotyping for validity, bias, explainability, privacy, security, workflow effects, and failure modes.
19. `PSY-PLO-19` Communicate with patients, families, interpreters, peers, multidisciplinary teams, services, and public audiences; disclose uncertainty, conflicts, errors, evidence limits, and tool use.
20. `PSY-PLO-20` Teach, supervise in simulation, seek help, use feedback, maintain boundaries, respond to impairment or mistreatment, and construct a sustainable plan for continued learning.
21. `PSY-PLO-21` Compare mental-health systems across resource settings and explain how financing, workforce, law, housing, inequality, colonization, climate, conflict, and policy shape access and outcomes.
22. `PSY-PLO-22` Integrate the curriculum in an independently authored capstone and defend its decisions before academic, clinical, and lived-experience perspectives.

## 6. Architecture and workload

The common curriculum contains 1,440 nominal academic hours. Hours include guided study, tutorials, standardized-person simulation, deliberate practice, assessment, and reflection. They **exclude** real patient care, on-call work, clinical-service delivery, and the protected supervision an accredited residency must provide.

| Stage | Modules | Academic hours | Developmental purpose |
|---|---|---:|---|
| I. Person, mind, brain, and method | `PSY-101`–`PSY-140` | 320 | build a humane, medically grounded language for observing, listening, explaining, and evaluating evidence |
| II. Formulation, safety, and presentations | `PSY-210`–`PSY-280` | 480 | reason from presentation to alternatives, urgency, formulation, and an initial collaborative plan |
| III. Treatment, lifespan, and continuity | `PSY-310`–`PSY-370` | 400 | compare treatments, follow outcomes, practise psychotherapy reasoning, and coordinate care across ages and settings |
| IV. Rights, systems, and independent scholarship | `PSY-410`–`PSY-460` | 240 | lead, improve, teach, evaluate technology, and integrate the curriculum under uncertainty |
| One area of depth | section 12 | 120–240 | extend a capability without implying subspecialty certification |

Recommended sequence:

```text
PSY-101 -> PSY-110 || PSY-120 || PSY-140
  -> PSY-130
  -> PSY-210 || PSY-220
  -> PSY-230
  -> PSY-240 || PSY-250 || PSY-260 || PSY-270 || PSY-280
  -> PSY-310 || PSY-320
  -> PSY-330
  -> PSY-340 || PSY-350 || PSY-360 || PSY-370
  -> PSY-410 || PSY-420 || PSY-430 || PSY-440 || PSY-450
  -> one AREA OF DEPTH -> PSY-460
```

The order is spiral rather than strictly linear. The longitudinal strands in section 7 must appear in every stage. Learners revisit common presentations with increasing complexity instead of completing one disorder and treating it as closed.

## 7. Longitudinal strands

Every module must visibly map to at least four strands; every assessment cycle samples all ten.

| Strand | Required recurrence |
|---|---|
| Lived experience, recovery, and rights | first-person epistemology, goals, strengths, supported decisions, peer roles, coercion alternatives, service co-design |
| Therapeutic relationship and psychotherapy | alliance, boundaries, formulation, process, rupture and repair, model-specific reasoning, supervision |
| Physical health and clinical neuroscience | medical and neurologic alternatives, mechanisms, disability, prevention, medication burden, mortality gap |
| Development and lifespan | attachment, family, schools, transitions, reproductive life, ageing, cognition, carers |
| Culture and structural context | language, explanatory models, racism, migration, Indigeneity, gender, sexuality, poverty, housing, spirituality, local knowledge |
| Safety, ethics, and uncertainty | risk formulation, safeguarding, capacity, least-restrictive response, clinician safety, confidence calibration, adverse events |
| Evidence and measurement | causal inference, diagnostic and prognostic evidence, outcomes that matter, qualitative knowledge, uncertainty and harms |
| Systems and public mental health | access, prevention, primary and community care, crisis systems, financing, workforce, policy, global resource variation |
| Communication and teamwork | standardized people, families, interpreters, peer specialists, multidisciplinary teams, handovers, public explanation |
| Professional formation | reflection, supervision, feedback, teaching, conflicts, boundaries, wellbeing, help-seeking, lifelong learning |

## 8. Detailed modules

### Stage I — Person, mind, brain, and method (320 hours)

#### `PSY-101` What psychiatry is — 40 hours

**Scope.** Competing histories and models of mental distress; the medical specialty and its limits; phenomenology; disability, neurodiversity, recovery, survivor and service-user movements; institutional power; racism, colonialism, gender and sexuality in diagnosis and care; rights; stigma; professional roles; the uses and harms of classification.

**Evidence.** A source-traced position paper compares three explanatory models, includes a lived-experience critique, states where they conflict, and proposes safeguards against overreach.

#### `PSY-110` Clinical neuroscience and mind–brain models — 80 hours

**Scope.** Functional neuroanatomy; neural development; signalling, networks, plasticity, stress systems, sleep and circadian biology; genetics and epigenetics; immune, endocrine, metabolic and gut–brain claims; neuropsychology; lesion and stimulation evidence; imaging; computational models; inference from association; replication and translation limits.

**Evidence.** An unfamiliar multi-method study is reconstructed from molecule to circuit to behaviour, with causal claims downgraded where the design cannot support them.

#### `PSY-120` Development, relationships, and the life course — 60 hours

**Scope.** Prenatal development through older age; attachment and caregiving; temperament; cognition, language, identity and moral development; family and peer systems; adversity and resilience; education and work; migration; reproductive transitions; ageing; developmental psychopathology; cohort and intergenerational evidence.

**Evidence.** A fictional longitudinal case is reformulated at four ages without treating childhood experience as destiny.

#### `PSY-130` Interview, observation, alliance, and communication — 80 hours

**Scope.** Preparing the setting; introductions and role clarity; open and focused inquiry; narrative and chronology; symptoms, strengths, goals and function; collateral sources; mental-status examination; interpreters and communication access; trauma-aware practice; tele-interviewing; boundaries; documentation; uncertainty; feedback; difficult conversations.

**Evidence.** Three standardized-person encounters across different communication needs, followed by transcript annotation, observer feedback, self-calibration, and repair.

#### `PSY-140` Epidemiology, evidence, and measurement — 60 hours

**Scope.** Incidence, prevalence, burden and service use; sampling; measurement validity and invariance; diagnostic accuracy; prognosis; trials; observational and qualitative designs; systematic reviews; heterogeneity; causal diagrams; missing data; effect size; minimal important change; harms; conflicts; open science; measurement-based care; instrument rights.

**Evidence.** A multi-study evidence brief answers a bounded question and reports absolute effects, uncertainty, applicability, harms, funding, and what remains unknown.

### Stage II — Formulation, safety, and presentations (480 hours)

#### `PSY-210` Formulation, classification, and diagnostic reasoning — 60 hours

**Scope.** Problem lists; syndromic, dimensional and narrative descriptions; ICD-11 literacy; DSM literacy without reproducing proprietary criteria; differential diagnosis; base rates; thresholds; comorbidity; heterogeneity; cultural formulation; functional analysis; causal and maintenance hypotheses; protective factors; Bayesian updating; diagnostic error; communicating a working formulation.

**Evidence.** A timed case conference compares at least four alternatives, explicitly tests medical and contextual explanations, and revises the formulation after new data.

#### `PSY-220` Medicine, neurology, investigations, and physical health — 60 hours

**Scope.** Delirium; seizures; movement, autoimmune, endocrine, infectious, nutritional, toxic, medication-related and sleep-related presentations; neurologic and physical examination reasoning; laboratory, toxicology, ECG, EEG, imaging and cognitive testing; pregnancy; multimorbidity; pain; health promotion; diagnostic overshadowing; physical-health inequity.

**Evidence.** A medical-mimic objective structured simulation requires escalation, investigation choices, interpretation, and a plain-language explanation of uncertainty.

#### `PSY-230` Urgency, safety, and crisis care — 60 hours

**Scope.** Suicide and self-harm; violence and victimization; abuse, neglect and exploitation; intoxication and withdrawal; agitation; catatonia; delirium; severe adverse effects; crisis communication; de-escalation; safety planning; means safety; safeguarding; handover; observation; emergency pathways; post-incident care; restraint and seclusion alternatives; limits of prediction.

**Evidence.** A circuit of fictional emergencies scored on recognition, immediate priorities, help-seeking, rights, communication, and avoidance of false numerical certainty. Any missed emergency action is non-compensable.

#### `PSY-240` Psychosis and unusual experiences — 60 hours

**Scope.** Hallucinations, delusions, disorganization, negative symptoms, catatonia and cognitive change; trauma, culture and spirituality; first episode; affective, substance-, medication-, neurologic and medical alternatives; early intervention; family work; antipsychotic evidence and burden; physical health; rehabilitation; hearing-voices and recovery perspectives; refractory illness.

**Evidence.** A first-episode transfer case integrates urgency, differential, meaning, shared decisions, family context, monitoring, and a five-year recovery horizon.

#### `PSY-250` Depression, bipolarity, and mood variation — 60 hours

**Scope.** Depressive and manic presentations; mixed features; grief; trauma; perinatal and seasonal contexts; bipolar differential; medical and medication-related causes; suicide; longitudinal course; psychosocial, pharmacologic and interventional options; activation, withdrawal and relapse; function and recovery.

**Evidence.** A longitudinal case requires phase-specific treatment comparisons and revision after an adverse effect and a change in the person's goals.

#### `PSY-260` Anxiety, obsessive-compulsive, trauma-related, and dissociative presentations — 60 hours

**Scope.** Fear and avoidance; panic; generalized anxiety; obsessions and compulsions; traumatic stress; dissociation; somatic arousal; developmental and cultural expression; medical and substance alternatives; exposure principles; cognitive, behavioural, pharmacologic and social interventions; stabilization and the risks of premature trauma processing.

**Evidence.** A mechanism map links learning processes to a graded plan while preserving consent, pacing, comorbidity, and alternative explanations.

#### `PSY-270` Substance use and behavioural addictions — 60 hours

**Scope.** Intoxication, withdrawal, overdose and interactions; alcohol, sedatives, opioids, stimulants, cannabis, nicotine and emerging substances; pain; gambling and debated behavioural addictions; harm reduction; motivational communication; pharmacotherapy; recovery communities; trauma; housing and criminalization; family effects; dual diagnosis.

**Evidence.** A non-judgmental care plan handles acute risk, readiness, harm reduction, medication options, social needs, recurrence, and local service constraints.

#### `PSY-280` Relational, behavioural, and bodily presentations — 60 hours

**Scope.** Personality functioning and enduring patterns; self-injury; eating and feeding disorders; somatic symptom and functional presentations; sleep disorders; sexual function and concerns; gender-related care without pathologizing identity; impulse-control presentations; factitious illness; overlap, stigma, medical risk, therapeutic stance, and iatrogenic harm.

**Evidence.** Comparative cases test medical risk, formulation, language, boundaries, team dynamics, and selection of specialist care without stereotype-based shortcuts.

### Stage III — Treatment, lifespan, and continuity (400 hours)

#### `PSY-310` Psychopharmacology and deprescribing — 70 hours

**Scope.** Decision thresholds; placebo and expectancy; pharmacokinetics, pharmacodynamics, interactions and pharmacogenomic limits; antidepressant, antipsychotic, mood-stabilizing, anxiolytic, stimulant, cognitive and addiction pharmacotherapies; age, pregnancy, organ function and comorbidity; monitoring; adherence; polypharmacy; shortages and access; adverse effects; withdrawal; deprescribing; regulatory evidence and conflicts.

**Evidence.** An oral medication conference compares options with absolute benefit and harm, monitoring and stopping plans, interaction checks, person preferences, cost, and uncertainty. It is not a prescribing authorization exercise.

#### `PSY-320` Psychotherapy science and core models — 70 hours

**Scope.** Alliance, empathy, boundaries, common and contextual factors; case formulation; goals and contracting; supportive, cognitive-behavioural, psychodynamic, interpersonal, dialectical-behavioural, motivational, family, couples and group approaches; trauma-informed care; process and outcome research; cultural adaptation; rupture and repair; ending; adverse effects; supervision and therapist development.

**Evidence.** Standardized-person micro-skills, model comparisons, session formulation, observed role-play, outcome interpretation, and supervision response. Reading or role-play never counts as delivering psychotherapy to patients.

#### `PSY-330` Recovery, rehabilitation, social and interventional treatments — 50 hours

**Scope.** Shared decision-making; peer support; family intervention; supported education, employment and housing; occupational and social recovery; physical activity, sleep and nutrition; ECT, TMS and other neuromodulation; ketamine and emerging interventions; service eligibility; evidence, consent, anaesthetic and medical considerations; maintenance; access and hype.

**Evidence.** A recovery portfolio contrasts a social, psychological, pharmacologic, and interventional option against goals chosen by a fictional person.

#### `PSY-340` Child, adolescent, and family psychiatry — 55 hours

**Scope.** Developmental assessment; family and school systems; neurodevelopmental presentations; mood, anxiety, psychosis, eating, trauma and substance use in young people; safeguarding; consent and assent; transition; medicine and psychotherapy evidence; caregiver partnership; online environments; anti-pathologizing and neurodiversity-affirming care.

**Evidence.** A multi-informant case reconciles disagreement among a young person, family, school, and clinician while centring safety, development, autonomy, and context.

#### `PSY-350` Older-adult, neurocognitive, and neuropsychiatric care — 45 hours

**Scope.** Normal ageing; delirium; mild and major neurocognitive disorders; depression, psychosis and anxiety in later life; neurologic disease; cognition; sensory and communication access; polypharmacy; capacity; carers; abuse; loneliness; palliative goals; services and transitions.

**Evidence.** A fluctuating-cognition case distinguishes delirium, dementia and depression and includes carers without erasing the older person's voice.

#### `PSY-360` Consultation-liaison and integrated care — 55 hours

**Scope.** Psychiatric questions in medical, surgical, obstetric, neurologic, pain, oncology, transplant, intensive, palliative and primary-care settings; reproductive and perinatal mental health; functional symptoms; capacity; treatment refusal; medication interactions; collaborative care; communication between teams.

**Evidence.** A consultation note answers the referrer's question, reframes it when necessary, integrates medicine and context, and provides prioritized, feasible recommendations.

#### `PSY-370` Community, public, rural, and global mental health — 55 hours

**Scope.** Prevention; population need; severe and persistent illness; crisis/home treatment; assertive and early-intervention models; primary care; homelessness; migration and displacement; correctional interfaces; rural and remote services; task-sharing; peer-led services; financing; workforce; policy; disaster and climate; implementation across resource settings.

**Evidence.** A service map and redesign proposal is co-evaluated for access, continuity, rights, equity, feasibility, outcomes, and unintended consequences.

### Stage IV — Rights, systems, and independent scholarship (240 hours)

#### `PSY-410` Ethics, law, forensic questions, and human rights — 40 hours

**Scope.** Ethical reasoning; professional duties; confidentiality; capacity; consent; advance planning; supported decision-making; safeguarding; compulsory assessment and treatment; criminal and civil interfaces; fitness and disability; expert evidence; correctional psychiatry; dual loyalty; coercion alternatives; international human-rights principles. All legal rules are jurisdiction- and date-labelled.

**Evidence.** A comparative legal-ethical analysis separates universal principles, contested claims, and local law and proposes the least-restrictive defensible action.

#### `PSY-420` Culture, identity, spirituality, and structural competence — 40 hours

**Scope.** Cultural humility; explanatory models; interpreters; racism and colonial history; Indigenous and First Nations mental health as locally governed learning; migration and asylum; gender and sexuality; disability; poverty, housing and work; religion and spirituality; community knowledge; structural formulation; advocacy and epistemic justice.

**Evidence.** A community-reviewed formulation audit identifies where a standard case changes when power, language, history, and access are made visible.

#### `PSY-430` Quality, safety, implementation, and service design — 40 hours

**Scope.** Human factors; just culture; incident analysis; handovers; medication safety; diagnostic safety; restraint reduction; measurement systems; QI methods; implementation and de-implementation; economics; stakeholder governance; equity; learning health systems; adverse-event disclosure.

**Evidence.** A preregistered service-improvement simulation includes theory of change, process and outcome measures, balancing measures, equity stratification, governance, and stop rules.

#### `PSY-440` Digital psychiatry, telehealth, AI, and data ethics — 30 hours

**Scope.** Telepsychiatry; accessibility; digital therapeutics; electronic records; patient portals; sensors and digital phenotyping; social media; clinical decision support; generative and agentic AI; automation bias; model validity and drift; privacy, security and consent; data provenance; documentation; environmental and labour costs; procurement and incident response.

**Evidence.** A red-team evaluation of a fictional product identifies unsupported claims, subgroup harms, privacy and workflow failures, human-oversight requirements, and conditions for refusal or withdrawal.

#### `PSY-450` Leadership, teaching, supervision, and professional sustainability — 40 hours

**Scope.** Team roles; conflict; inclusive meetings; resource stewardship; management and financing; teaching design; feedback; assessment validity; supervision; remediation; whistleblowing; boundaries; moral injury; burnout; impairment; bullying and harassment; help-seeking; career transition and continuing development.

**Evidence.** Learners lead a simulated case conference, teach one difficult concept, respond to a professionalism dilemma, and revise a personal development plan from multisource feedback.

#### `PSY-460` Integration and capstone — 50 hours

**Scope.** Independent synthesis of a consequential problem in psychiatric knowledge, education, service, or policy. Acceptable forms include a systematic evidence map, reproducible analysis, rights-based service design, open teaching unit, implementation proposal, or research protocol. A personal clinical case report is not accepted.

**Evidence.** Proposal, source and rights audit, preregistered questions, public artifact, limitations statement, oral defence, stakeholder feedback, correction log, and an accessible nontechnical explanation. The defence panel should include academic/clinical, educational-method, and lived-experience perspectives; panel participation does not convert the course into clinical credentialing.

## 9. Repeating learning grammar

Every presentation and treatment lesson follows the same sequence so that diagnosis cannot crowd out the rest of care:

1. person's question, language, goals, strengths, and setting;
2. phenomenology, chronology, function, development, culture, and context;
3. urgency, safeguarding, medical and neurologic alternatives;
4. differential, dimensional description, classification, and uncertainty;
5. multi-level formulation and explicit maintenance/protective hypotheses;
6. shared option comparison: benefits, harms, burdens, access, preferences, and no-treatment alternative;
7. acute, longitudinal, recovery, family, community, and physical-health plan;
8. outcomes that matter, monitoring, review, stopping, and repair after harm;
9. evidence quality, conflicts, rights, and unanswered questions;
10. transfer to a new age, identity, comorbidity, service, or resource setting.

Cases must vary across age, disability, language, race and ethnicity, Indigeneity where locally governed, gender, sexuality, family form, religion, socioeconomic position, housing, migration history, geography, and service access. Variation must change the reasoning rather than decorate it.

## 10. Mastery and assessment

### 10.1 Academic mastery states

| State | Meaning | Minimum academic evidence |
|---|---|---|
| `N` Not encountered | no interpretable evidence | none |
| `A` Acquiring | can recognize and follow a model | guided checks and feedback |
| `P` Practising | can handle familiar fictional cases with prompts | two successful practice occasions |
| `M` Mastered | can explain and apply independently in a new bounded simulation | unseen gate, all critical actions correct, plus transfer task |
| `R` Retained | re-demonstrated after spacing in a different context | two delayed checks, one at least 28 days later |
| `T` Transfer | integrates multiple modules under uncertainty | rubric-scored integrative performance or capstone |

These states describe course evidence only. They are not clinical entrustment decisions.

### 10.2 Assessment program

The assessment program samples knowledge, reasoning, communication, uncertainty, rights, and feedback uptake through:

- low-stakes retrieval and confidence calibration;
- key-feature and script-concordance-style fictional cases;
- oral formulations and differential updates;
- standardized-person interviews and difficult-conversation simulations;
- observed psychotherapy micro-skills, never represented as patient treatment;
- evidence appraisals, medication conferences, and measurement critiques;
- simulated emergency, handover, disclosure, and team exercises;
- service maps, quality-improvement work, teaching, and capstone defence;
- multisource feedback, including appropriately supported lived-experience educators;
- a longitudinal portfolio with corrections, reflections, rights decisions, and tool-use disclosure.

No single examination can compensate for a missed critical action. Safety- and rights-critical gates include recognizing medical emergencies, seeking help, preserving privacy, avoiding discriminatory or coercive shortcuts, responding to safeguarding concerns, and not presenting a simulation result as clinical authority.

### 10.3 Progress decisions

Academic progression requires multiple observations, different cases, more than one assessor, retained performance, and response to feedback. A competence committee model is recommended for host institutions, but EmbeddedKnowledge does not constitute such a committee. Appeals, accessibility adjustments, remediation, and reassessment must be documented and separate from punitive conduct processes.

## 11. Supervised-practice crosswalk

Residency frameworks require real clinical experience that an open book cannot provide. A recognized clinical host may map this academic curriculum to its own standards, but must independently supply and verify:

- supervised responsibility across inpatient, outpatient, emergency, community, consultation-liaison, child and adolescent, older-adult, addiction, forensic, medical/neurologic, and other locally required settings;
- longitudinal care and follow-up rather than isolated encounters;
- observed assessment, formulation, treatment planning, handover, teamwork, and documentation with real patients;
- protected individual and group supervision, including psychotherapy supervision;
- progressive responsibility based on demonstrated performance and patient acuity;
- workplace-based assessment from multiple observers and settings, service-user feedback where ethical, a clinical portfolio, remediation, and an appeal route;
- local training in prescribing, procedures, mental-health law, safeguarding, emergency response, privacy, infection control, and occupational safety;
- identity, medical registration, background, health, duty-hour, employment, indemnity, and site-accreditation requirements.

Simulation can prepare for rare or dangerous situations. It cannot replace required patient contact, supervised psychotherapy cases, procedure experience, or a regulator's decision. EmbeddedKnowledge will not record a learner as clinically entrusted merely because a host links to this syllabus.

## 12. Areas of depth (120–240 hours)

One area of depth is required for the course's academic completion notation. It signals extended study, **not** subspecialist status.

| ID | Area | Required product |
|---|---|---|
| `PSY-PW-RES` Research and academic psychiatry | reproducible study, systematic review protocol, or open-methods replication |
| `PSY-PW-PSY` Psychotherapy science and education | comparative formulation portfolio, supervised simulation, and process/outcome evidence review |
| `PSY-PW-NEU` Neuropsychiatry and interventional science | mechanism-to-clinic evidence map with procedure, safety, consent, and hype audit |
| `PSY-PW-CAP` Child, adolescent, and family mental health | developmentally adapted teaching or service-design capstone with youth/family governance plan |
| `PSY-PW-OAP` Older-adult and neurocognitive care | integrated cognitive, medical, social, carer, rights, and service project |
| `PSY-PW-ADD` Addiction and harm reduction | community-informed prevention, care, or recovery project |
| `PSY-PW-CLP` Consultation-liaison and reproductive mental health | integrated-care pathway or evidence synthesis across specialties |
| `PSY-PW-PUB` Public, community, rural, and global mental health | population-needs analysis and context-feasible service proposal |
| `PSY-PW-FOR` Forensic ethics, law, and policy | date- and jurisdiction-labelled comparative analysis with dual-loyalty safeguards |
| `PSY-PW-DIG` Digital mental health and implementation | independent validation or procurement framework for a digital intervention |
| `PSY-PW-EDU` Education, leadership, and quality | assessable open teaching unit or governed service-improvement portfolio |

An individually designed pathway requires a question, outcomes, supervision plan, source and rights audit, 120-hour minimum, and capstone approved before work begins. Production instrumentation for all pathways is future work.

## 13. Source, classification, and rights policy

### 13.1 Evidence hierarchy

Production lessons should triangulate:

1. current official safety, regulatory, public-health, and classification sources;
2. high-quality guidelines and systematic reviews with explicit population, date, jurisdiction, and conflicts;
3. pivotal and corrective primary research;
4. qualitative, community, implementation, and lived-experience evidence;
5. clearly licensed educational material used within its level and scope.

Guidelines do not become truth by authority, and randomized trials do not answer every question. Material claims must distinguish consensus, evidence-supported inference, plausible hypothesis, contested judgment, and local rule.

### 13.2 Classification and instruments

ICD-11 is the global classification reference. Its text and codes retain WHO's licence and may not be rewritten into a derivative classification. DSM literacy is required for jurisdictions and literature that use it, but proprietary criteria, tables, and wording must not be reproduced. Screening tools, rating scales, interview schedules, cognitive tests, and psychotherapy worksheets require instrument-level rights checks; being common or free to read is not permission to republish.

### 13.3 Reuse

EmbeddedKnowledge curriculum and original lessons are CC BY 4.0. A source carrying NonCommercial, ShareAlike, or NoDerivatives terms may be cited as evidence but is not copied, closely adapted, or embedded into this differently licensed corpus. U.S. federal material still requires item-level checks for contractor, image, logo, instrument, and jurisdictional exceptions. Every source and asset records author, title, version/date, URL, access date, licence, reuse decision, and changes.

OpenStax books are excluded from model-assisted authoring under their stated AI-use restriction. Patient stories, forums, charts, and clinical media are never scraped for cases.

## 14. Cases, privacy, accessibility, and learner welfare

### 14.1 Case standard

Production cases are fictional composites written from public knowledge or are purpose-commissioned with documented consent and rights. They contain no identifiable patient data and no disguised one-to-one clinical narrative. A case has a provenance note, uncertainty, non-stigmatizing language, relevant strengths, and review for stereotype and diagnostic leakage.

Lived-experience narratives require informed consent for the exact use, a negotiated CC BY 4.0 licence, attribution preference, compensation, accessibility, withdrawal terms before publication, support, and protection against pressure to disclose. Consent to care, research, or a public interview is not consent to create an open derivative lesson.

### 14.2 Accessibility and potentially distressing material

All material must work with keyboard and assistive technology, reflow without horizontal scrolling, avoid colour-only meaning, provide transcripts/captions and text equivalents, and use plain-language summaries for dense sections. Learners may use accessible standardized-person alternatives where an activity depends on speech, vision, hearing, movement, or synchronous participation; the same construct must be assessed.

Modules involving suicide, self-harm, violence, abuse, trauma, eating disorders, coercion, or discrimination provide a specific preview, the pedagogical reason for inclusion, an equivalent route where feasible, pause/return controls, and non-punitive support guidance. The course does not monitor learners or provide crisis care. Public pages direct immediate concerns to current local emergency and crisis resources rather than assuming one country's service.

## 15. Digital and AI safeguards

Learners and contributors may use declared tools for brainstorming, translation, accessibility, coding, or critique only when the assignment permits it. They remain accountable for every claim, citation, calculation, right, and disclosure.

The following are prohibited in this curriculum:

- entering identifiable patient, learner, staff, or service data into an unapproved tool;
- asking a model to diagnose, predict suicide or violence, recommend detention, prescribe, or replace supervision;
- presenting synthetic empathy, a chatbot, or a score as a therapeutic relationship;
- using generated citations without checking the primary source;
- training or evaluating a tool on copyrighted or consent-restricted material without permission;
- deploying a clinical model because it performs well in a classroom simulation;
- concealing model assistance in assessed or published work.

AI-related lessons must include a no-tool route, subgroup performance, uncertainty, human factors, workflow, privacy, security, drift, red-team cases, incident reporting, and a defensible condition under which the system should not be used.

## 16. Completion, records, and boundaries

Academic completion requires:

- every required program outcome at `M` or above, with retained evidence for all safety- and rights-critical outcomes;
- completion of every stage and one area of depth;
- a portfolio containing at least six observed simulations, six evidence appraisals, four longitudinal case revisions, one quality or service-design project, a teaching performance, and a rights/licensing audit;
- an accepted capstone and oral defence with all material corrections resolved;
- final calibration showing that confidence tracks performance and that the learner seeks help when a problem exceeds the evidence or their role.

The record names the curriculum version, assessment modality, accommodations, completed pathway, and whether any evidence came from an external host. It does not convert external clinical activity into an EmbeddedKnowledge claim of competence.

The following statements are never permitted:

- “completed psychiatry residency” or “psychiatrist” on the basis of this course;
- “board eligible,” “licensed,” “credentialed,” or “competent to practise”;
- “qualified to diagnose, prescribe, provide psychotherapy, conduct capacity evaluations, detain, restrain, or perform a procedure”;
- “equivalent to” a named national training program without that authority's formal determination.

## 17. Quality governance and revision

Before the candidate can become version 1.0, the project should obtain documented review from:

- practising psychiatrists from more than one training jurisdiction and service setting;
- residents or recent graduates;
- people with lived and living experience, including perspectives critical of psychiatric systems;
- peer specialists and family/carer representatives;
- child/adolescent, older-adult, addiction, psychotherapy, consultation-liaison, community/public, forensic, neuroscience, and primary-care expertise;
- nursing, psychology, social work, occupational therapy, pharmacy, and other multidisciplinary partners;
- experts in education, assessment, statistics, human rights, culture, accessibility, privacy, and content licensing.

Conflicts, compensation, jurisdiction, and role are disclosed. Review findings and dispositions are public; private clinical detail is not. High-impact safety claims receive primary-source re-verification and subject-matter review. Classification, major guideline, medication-safety, law, rights, crisis-resource, and AI-regulation sources are audited at least annually and whenever a material update is reported.

Versioning uses semantic curriculum labels:

- patch: clarification or correction without changing outcomes or completion;
- minor: additive or reorganized content with a migration note;
- major: changed graduate capabilities, safety gate, workload, or completion contract.

## 18. Research basis and known gaps

The source audit behind this candidate is published in `research/PSYCHIATRY-SOURCE-CATALOG.md`. The design was synthesized from current official training requirements, competency frameworks, residency curricula, public-health and human-rights guidance, and an item-level open-source and licensing audit. Three role-separated research runs examined United States training, international training, and reusable knowledge sources; their conclusions were checked and reconciled by the authoring run. This is research provenance, **not** formal review quorum.

Known gaps before a production curriculum can open include:

- jurisdiction-specific law and culturally governed Indigenous/First Nations learning;
- rights-cleared diagnostic instruments, psychotherapy materials, clinical media, and specialist-depth OER;
- commissioned, compensated lived-experience cases and governance;
- a machine-readable knowledge graph, lesson boundaries, misconception map, and assessment blueprint;
- external expert review of workload, sequence, omissions, safety gates, and global applicability;
- a clinical-host crosswalk that regulators and accredited programs can inspect without confusing academic and workplace evidence.

Until those gaps are resolved through separate reviewed contributions, this remains an evolving academic curriculum and not a residency or clinical credential.

## Appendix A — Clinical-framework crosswalk

This is a directional design audit, not a claim of formal equivalence.

| Source family | Curriculum response |
|---|---|
| ACGME and ABPN (United States) | four-stage progression; medicine/neurology; inpatient, outpatient, emergency, consultation, child, older-adult, addiction, community and forensic breadth; psychotherapy; QI; observed communication and reasoning |
| RCPsych (United Kingdom) | compact higher-level capabilities; person-centred complexity; uncertainty, law, safeguarding, leadership, teaching and research; longitudinal reflective and psychotherapy learning |
| Royal College of Physicians and Surgeons of Canada | staged competence; observable tasks across cases, settings and assessors; transition-to-practice; longitudinal care; cultural safety and scholarship |
| RANZCP (Australia and Aotearoa New Zealand) | developmental entrustment logic; protected supervision; psychotherapy depth; cultural safety, Indigenous mental health, critical thinking and reflective formation |
| UEMS | cross-setting breadth; rights and recovery; weekly supervision principle; psychotherapy theory and supervised practice; curriculum-to-assessment mapping |
| WPA and WHO | lived-experience participation; public/global mental health; recovery; supported decisions; community inclusion; rights and alternatives to coercion |

## Appendix B — Minimum lesson evidence

Every future production lesson must include:

1. an explicit learner model, prerequisite check, outcomes, and boundary;
2. a case or problem that makes the knowledge consequential without soliciting personal disclosure;
3. a coherent explanatory model plus its limits and at least one alternative;
4. worked reasoning, retrieval, misconception repair, varied practice, feedback, and transfer;
5. same-scene learner-visible source notes for material claims;
6. current safety and jurisdiction labels where applicable;
7. a source, instrument, asset, case, and AI-use rights record;
8. accessible equivalents and a content preview for potentially distressing material;
9. a non-stigmatizing and stereotype audit, including lived-experience review where relevant;
10. a clear statement of what the lesson does **not** authorize.

## Appendix C — Candidate decision log

| Decision | Rationale | Revisit trigger |
|---|---|---|
| 1,440-hour common academic curriculum | permits residency-level breadth without pretending to count clinical service | workload review finds redundancy, missing practice, or infeasible pacing |
| Four stages, not PGY years | preserves a developmental spine across countries with different program lengths | an adopted host standard requires a dated local crosswalk |
| ICD-11 reference plus DSM literacy | supports global use while acknowledging jurisdictions and literature | classification or licence changes |
| Lived experience and rights are longitudinal | these change assessment, treatment, governance, and systems—not only ethics content | co-production review recommends stronger authority or safeguards |
| Psychiatry lesson corpus begins empty | stable graph and production boundaries exist; each lesson still needs evidence, rights, independent review, and adjudication | focused lesson packs pass the public contribution protocol |
| No clinical entrustment notation | an open academic provider cannot verify workplace performance or authority | a separately governed accredited-host mechanism is approved |


---

## Repository file: course/PSYCHIATRY-KNOWLEDGE-GRAPH.md

# Psychiatry knowledge graph and contribution intake

**Version:** 0.1.0 proposal

**Status:** external review required; merging this proposal opens lesson intake for the reviewed common-academic outcomes

**Canonical source:** `scripts/build-psychiatry-graph.mjs`

**Generated artifact:** `site/data/psychiatry-graph.json`

**Content licence:** CC BY 4.0

## Purpose and boundary

This graph turns the 1,440-hour common academic core in `PSYCHIATRY-SYLLABUS.md` into stable, contribution-sized learning outcomes. It is a curriculum map for open education, not a diagnostic system, treatment guide, residency, clinical placement, licence, board-eligibility route, or authorization to practise psychiatry. Real-patient care, prescribing, psychotherapy delivery, compulsory-care decisions, workplace supervision, and protected professional titles remain outside EmbeddedKnowledge.

Every learner-facing clinical example must therefore use fictional, de-identified, or fully consented material; distinguish education from advice; preserve uncertainty and alternatives; and state when local law, policy, culture, access, or service context controls the answer.

## Shape

The generated graph contains:

- 4 developmental stages;
- 26 syllabus modules;
- 154 atomic, observable outcomes;
- 1,440 hours of common academic study and simulation;
- 180 hierarchy links, 178 prerequisite links, and 27 bidirectional cross-concept pairs (54 directed cross-links are represented as 27 graph relationships);
- three non-credential routes: open scholar, physician learner, and educator/system learner.

The optional 120–240-hour area of depth is deliberately excluded. It should receive a separate graph only after pathway hosts, supervision, assessment, and jurisdictional boundaries exist.

## Stable identity

Topic IDs use `topic-psy-<module>-<slug>`, for example `topic-psy-101-field-and-boundaries`. Human-facing topic codes use `<module>.<sequence>`, for example `PSY-101.01`.

Once this graph is merged, an outcome ID is permanent. Editorial improvements change its title, summary, evidence, or version; they do not silently recycle its ID. A materially different learning destination receives a new ID, and any retirement is recorded with an explicit replacement or no-replacement decision. Lesson IDs use `PSY-<AREA>-<NNN>` and map only to Psychiatry graph outcomes.

## Relationship semantics

- `contains` expresses stage → module and module → atomic-outcome membership.
- `prerequisite` points from the earlier outcome to the outcome that depends on it. The prerequisite layer must remain acyclic.
- `cross-link` records a meaningful conceptual bridge that is not a prerequisite. The source declarations are symmetric; the public graph stores each pair once.

Prerequisites are deliberately sparse. They signal a learning dependency, not merely a useful association. Cross-links make recurring strands—rights, culture, lived experience, uncertainty, safety, evidence, and systems—visible without forcing a false linear sequence.

## Evidence and status

The graph synthesizes the Psychiatry syllabus, its dated source catalog, and primary specialist-training frameworks from ACGME, the Royal College of Psychiatrists, the Royal College of Physicians and Surgeons of Canada, RANZCP, UEMS, WPA, WHO, and WHO/OHCHR. Source tags support audit; they do not imply endorsement or equivalence to any training program.

All 154 outcomes carry `evidenceConfidence: pending-external-review` in this proposal. No expert, trainee, or lived-experience approval is claimed by the artifact or by automated validation.

## Review requested before merge

This proposal should not be marked ready until its scope, omissions, ordering, workload, language, and rights posture have been reviewed in public by a deliberately plural group that includes, where available:

- practising psychiatrists from more than one jurisdiction and service setting;
- current psychiatry residents or trainees;
- people with lived experience, survivor perspectives, peer workers, carers, and family advocates, without treating any one voice as representative of all;
- nursing, psychology, social work, occupational therapy, pharmacy, primary care, emergency medicine, and public-health perspectives;
- specialists in education, measurement, statistics, disability access, culture, human rights, privacy, security, and open licensing.

Reviewers should identify missing or overclaimed outcomes, unsafe sequencing, culturally narrow assumptions, coercive or stigmatizing language, hidden prerequisites, infeasible workload, and areas where a route should stop rather than simulate clinical authority. Disagreement should be documented, not averaged away.

## Intake rule after merge

Merging this graph opens focused lesson proposals for its common-academic outcomes under the existing Lesson Format v1, evidence, rights, provenance, independent review, and fresh-finalization contracts. It does not publish a lesson, count the specimen, or waive quorum. An outcome is available only when it has no merged lesson and no active lesson proposal in `site/data/psychiatry-open-prs.json`.

The first intended contribution target is `topic-psy-101-field-and-boundaries`. A local welcome-lesson prototype may inform a later candidate, but it must not enter `lessons/` or count toward coverage until this graph and intake contract are merged and the authoring skill is run fresh against the live target state.

## Deterministic maintenance

Run:

```bash
npm run graph:build
npm run graph:validate
npm run site:build
npm run validate
```

Edit the builder, not the generated JSON. Validation checks identity, endpoints, graph shape, declared relationships, course hours, prerequisite acyclicity, routes, source tags, outcome form, and the pending-review marker. Automated validation establishes internal consistency only; it cannot substitute for the external review requested above.


---

## Repository file: research/PSYCHIATRY-SOURCE-CATALOG.md

# Psychiatry curriculum source catalog

**Audit date:** 22 July 2026<br>
**Supports:** `course/PSYCHIATRY-SYLLABUS.md` version 0.1 research candidate<br>
**Scope:** postgraduate psychiatry training architecture, public program curricula, human-rights and lived-experience frameworks, open educational sources, and reuse constraints<br>
**Status:** checked synthesis for curriculum design; not a clinical guideline, prescribing reference, or formal equivalence determination

## 1. Research question

What should a modern, internationally legible academic psychiatry curriculum teach; how do current specialist programs sequence, supervise, and assess that learning; and which sources may lawfully support an openly licensed, model-assisted lesson corpus?

The answer required three different evidence problems:

1. **Training requirements:** what regulators and specialty colleges currently require.
2. **Enacted curricula:** how real programs turn requirements into sequences, teaching, supervision, and assessment.
3. **Knowledge and rights:** which sources are authoritative, which are reusable, and where specialist content must be commissioned.

## 2. Method and provenance

Three role-separated research runs were commissioned on 22 July 2026:

| Track | Question | Model family | Output used |
|---|---|---|---|
| United States training | national requirements, certification, Milestones, and a purposive sample of public residency curricula | OpenAI GPT-5.6 Sol | source map, required experiences, assessment patterns, program variation, modern gaps |
| International training | UK, Canada, Australia/Aotearoa New Zealand, UEMS, WPA, and WHO frameworks | OpenAI GPT-5.6 Sol | comparative progression, supervision, psychotherapy, rights, culture, and lived-experience requirements |
| Open knowledge and rights | official, public-domain, and OER sources plus copyright, licence, robots, and AI-use restrictions | OpenAI GPT-5.6 Sol | reusable/citation-only decisions, access cautions, and content gaps |

The authoring run reconciled the reports, checked the primary-source links used in the syllabus design, separated requirements from illustrative program practice, and excluded unverified secondary claims. These runs were isolated by research question but used the same provider/model family. They are **research assistance, not independent review quorum**. No raw model report is published as evidence; only checked conclusions and source links appear here.

### 2.1 Selection rules

Priority order:

1. current official regulator, certifying-board, specialty-college, government, intergovernmental, or university-program source;
2. a dated and versioned primary PDF where available;
3. a current official landing page when requirements are maintained online;
4. openly licensed educational sources only after item-level rights inspection.

Program examples were purposively sampled to expose different sequencing and emphases, not to rank programs or estimate prevalence. A feature on one residency page is evidence that the design exists, not that all residents receive it.

### 2.2 Claim labels

- **Requirement:** stated by a responsible training or certifying body.
- **Program practice:** described by an official program page.
- **Design inference:** this project's reasoned response to agreements, differences, or omissions.
- **Rights decision:** the project's conservative reuse determination; it is not legal advice.

### 2.3 Important limits

- Rules, webpages, releases, licences, and curricula change. Each production lesson must recheck time-sensitive claims.
- Public access is not open licensing. Robots permission is not copyright permission, and copyright permission is not ethical permission.
- English-language frameworks and high-income-country specialist systems are overrepresented.
- National frameworks do not fully express local, Indigenous, survivor-led, community, or multidisciplinary knowledge.
- A curriculum document cannot establish what happens consistently in clinical workplaces.

## 3. United States national framework

| Source | Version/status checked | What it establishes | Curriculum use | Rights posture |
|---|---|---|---|---|
| [ACGME Program Requirements for Graduate Medical Education in Psychiatry](https://www.acgme.org/globalassets/pfassets/programrequirements/2026-prs/400_psychiatry_2026.pdf) | requirements effective 1 July 2026, with 2026 interim revisions/FAQs | 48-month program; medicine and neurology grounding; inpatient and continuous outpatient experience; child, older-adult, addiction, consultation-liaison, emergency, community and forensic breadth; graded responsibility; supervision; didactics; QI; evaluation | requirement and architecture reference | cite and paraphrase; do not treat the document as CC BY substrate |
| [ACGME Psychiatry Milestones 2.0](https://www.acgme.org/globalassets/PDFs/Milestones/PsychiatryMilestones2.0.pdf) | current ACGME-linked Milestones; second revision, March 2020 | developmental subcompetencies in assessment, formulation, treatment, psychotherapy, somatic therapies, consultation, neuroscience, evidence, systems, safety, professionalism and communication | developmental design reference; not a copied rubric | educational reuse is allowed in its intended context; preserve attribution and do not imply ACGME endorsement |
| [ACGME Psychiatry Milestones Supplemental Guide](https://prep.acgme.org/globalassets/pdfs/milestones/psychiatrysupplementalguide.pdf) | official supplemental guide | examples and assessment suggestions for interpreting Milestones | assessment inspiration only | cite/paraphrase conservatively |
| [ABPN Psychiatry certification requirements](https://www.abpn.com/become-certified/taking-a-specialty-exam/psychiatry/) | current September 2026 examination page | qualifying training sequence, continuous outpatient requirement, clinical-skills evaluations, licensure and continuity conditions | certification context and boundary | citation only |
| [ABPN Psychiatry Certification Content Specifications](https://abpn.org/wp-content/uploads/Psychiatry-Certification-Content-Specifications.pdf) | posted 29 September 2025 | board blueprint crosses disorders/topics with mechanisms and competencies; treatment and neuroscience are substantial domains | coverage audit, not exam-prep copying | citation only; no item or table reproduction |
| [ABPN Clinical Skills Evaluation requirements](https://abpn.org/wp-content/uploads/CSE-Psychiatry.pdf) | revised August 2024 | three observed evaluations of relationship, interview/MSE, and presentation; real-patient and telehealth conditions | evidence that direct observation matters; the EK course uses simulation only | citation only |
| [AADPRT Curriculum Committee](https://www.aadprt.org/training-directors/curriculum) | current page checked 22 July 2026 | peer review and Model Curriculum designation; most resources are member-only | ecosystem context, not an open foundation | do not ingest or reproduce member-gated material |

### 3.1 Design conclusions

The durable U.S. spine is physician/neurologic grounding, acute assessment, longitudinal biopsychosocial treatment including psychotherapy, then increasing responsibility for systems and scholarship. The national floor is specific about clinical settings but permits local sequencing. It supports a four-stage academic spiral, but it does **not** authorize an open provider to count real clinical experience.

The syllabus therefore includes medicine and neurology, a continuous longitudinal-care model, multiple psychotherapy traditions, consultation, emergency and community systems, assessment by observation, QI, and a clear clinical-host boundary.

## 4. Official United States program sample

These pages illustrate enacted choices. Their update dates vary; undated pages are treated as current descriptions only as of the audit date.

| Program source | Distinctive public practice | Design implication |
|---|---|---|
| [Yale clinical curriculum](https://medicine.yale.edu/psychiatry/education/residency/curriculum/clinical-curriculum/) | medicine, neurology and psychiatry in PGY-1; inpatient, geriatrics, consultation-liaison and emergency in PGY-2; continuous outpatient PGY-3; individualized PGY-4 | a recognizable sequence remains useful, but stage purpose matters more than fixed labels |
| [Yale didactics](https://medicine.yale.edu/psychiatry/education/residency/curriculum/didactics/) | year-specific case-based learning in neuroscience, psychopharmacology, psychotherapy, forensics, structural competency, leadership and equity | spiral didactics should meet learners near their current clinical questions |
| [Columbia core curriculum](https://www.columbiapsychiatry.org/education-and-training/psychiatry-residency/core-curriculum) | sequenced central and service-specific teaching, process groups, safety conferences, journal clubs and an independent project | professional formation, safety, scholarship and service learning should remain visible |
| [Columbia PGY-1](https://www.columbiapsychiatry.org/education-and-training/psychiatry-residency/residency-training-program/pgy-1-year), [PGY-3](https://www.columbiapsychiatry.org/education-and-training/psychiatry-residency/core-curriculum/pgy-3-year), and [PGY-4](https://www.columbiapsychiatry.org/education-and-training/psychiatry-residency/residency-training-program/pgy-4-year) | early subspecialty exposure, a psychotherapy-rich longitudinal outpatient year, and extensive senior choice | child/addiction content can start early; senior depth should not displace continued core practice |
| [UCSF clinical experiences](https://psychiatry.ucsf.edu/rtp/clinical) | longitudinal clinic from PGY-1; later junior-attending and systems activity | continuity can start before a dedicated outpatient stage |
| [UCSF didactic curriculum](https://psychiatry.ucsf.edu/rtp/curriculum) | more than 500 scheduled hours; adult-learning design; structural competency, interventional and psychedelic psychiatry, integrated care, reproductive and LGBTQ mental health | modern and contested areas belong in an evidence-and-hype framework, not as promotional electives |
| [Stanford: How We Learn](https://med.stanford.edu/psychiatry/residents/general/learn.html) | early and modality-specific psychotherapy, neuroscience course and substantial scholarly-concentration time | psychotherapy and scholarship require longitudinal protected practice |
| [University of Wisconsin program overview](https://www.psychiatry.wisc.edu/education-training/residency/program-overview/) | public-system, rural, integrated-care, ECT, forensic, community and QI experiences | geography and service context should be represented explicitly |
| [MGH–McLean adult residency](https://www.massgeneral.org/psychiatry/education-and-training/adult-psychiatry-residency) and [public/community psychiatry](https://www.massgeneral.org/psychiatry/treatments-and-services/division-of-public-and-community-psychiatry) | neuroscience identity plus public/community work including homelessness, integrated and correctional settings | neuroscience and public psychiatry should not be rival tracks |

### 4.1 Variation that the syllabus preserves

- Longitudinal outpatient experience may begin early or concentrate later.
- Psychotherapy depth, observed sessions, complete cases, and protected scholarship vary.
- Community and forensic exposure are required nationally but not assigned a uniform duration.
- Board blueprints emphasize knowledge distribution, while program requirements emphasize settings, supervision and continuity.
- National requirements do not yet give sufficient explicit coverage to AI governance, digital phenotyping, implementation science, co-production, diagnostic harm, or a robust instrument-rights practice.

These are not claims that every program has a gap. They explain why a synthetic syllabus makes the topics explicit.

## 5. United Kingdom

| Source | Version/status | What it establishes | Design use | Rights posture |
|---|---|---|---|---|
| [RCPsych curricula documents and resources](https://www.rcpsych.ac.uk/training/curricula-and-guidance/curricula-implementation/curricula-documents-and-resources/) | current implementation hub checked 22 July 2026 | nine higher learning outcomes and current implementation resources | compact capability architecture above granular modules | citation and independent synthesis |
| [RCPsych Core Psychiatry Curriculum](https://www.rcpsych.ac.uk/docs/default-source/training/curricula-and-guidance/2022-curricula/core-psychiatry-curriculum-final-17-august-2022.pdf?sfvrsn=36b5ba25_10) | version 1.0, August 2022; 36 months WTE | professional relationships; communication and complexity; law; population health; teamwork; safety/QI; safeguarding; education; research; lifespan and holistic formulation | capability and assessment crosswalk | copyright retained; paraphrase/cite |
| [RCPsych General Psychiatry Curriculum](https://www.rcpsych.ac.uk/docs/default-source/training/curricula-and-guidance/2022-curricula/general-psychiatry-curriculum-final-16-june-22.pdf?sfvrsn=b0983a16_4) | 2022 higher-specialty curriculum | adult care across inpatient, community, crisis/home and other contexts; physical health, leadership, research and QI | advanced systems-stage coverage | copyright retained; paraphrase/cite |
| [RCPsych psychotherapy capabilities guide](https://www.rcpsych.ac.uk/docs/default-source/training/curricula-and-guidance/psychotherapy-capabilities---a-guide-for-psychiatry-resident-doctors-in-core-training.pdf?sfvrsn=4c855c57_3) | published May 2026 | longitudinal reflective groups, supervised long and short cases, protected supervision and spiral development | psychotherapy must be practised and supervised in a host, while EK limits itself to academic/simulation evidence | copyright retained; no table or text adaptation |

The UK framework contributed concise higher-level capabilities, explicit complexity and uncertainty, safeguarding, behavioural progression, and psychotherapy as longitudinal professional formation.

## 6. Canada

| Source | Version/status | What it establishes | Design use | Rights posture |
|---|---|---|---|---|
| [Royal College Psychiatry information by discipline](https://www.royalcollege.ca/en/standards-and-accreditation/information-by-discipline?afc-diploma=&special-program=&specialty=royal-college%3Aibd%2Fspecialty%2Fpsychiatry&subspecialty=) | current official document hub | Competence by Design requirements, competencies, EPAs, experiences and accreditation | primary discovery source | copyright retained |
| [Psychiatry Training Experiences](https://www.royalcollege.ca/content/dam/documents/ibd/psychiatry/psychiatry-training-experiences-e.pdf) | version 2.0, applies to entrants from 1 July 2026 | planned five-year program and four developmental stages; breadth, longitudinal care, psychotherapy, QI and transition-to-practice | four-stage spine and breadth audit | limited educational reproduction terms; this CC BY curriculum paraphrases only |
| [Psychiatry EPA Guide](https://www.royalcollege.ca/content/dam/documents/ibd/psychiatry/epa-guide-psychiatry-v2-e.pdf) | version 2.0, July 2026; June 2026 typographical corrections | observable tasks with required diversity of cases, settings, observations and observers | multiple samples and observers; EK does not claim workplace entrustment | copyright retained; no EPA copying |
| [Psychiatry Competencies](https://www.royalcollege.ca/content/dam/documents/ibd/psychiatry/psychiatry-competencies-e.pdf) | 2020 | detailed medical expert and CanMEDS-related psychiatric capabilities | coverage audit | copyright retained; paraphrase/cite |

Canada contributed the clearest four-stage continuum, explicit observation sampling, medical/neurologic foundations, cultural safety, longitudinal severe-illness care, adverse-event responsibility, and a true transition-to-practice stage. The syllabus borrows the design logic, not the copyrighted wording or tables.

## 7. Australia and Aotearoa New Zealand

| Source | Version/status | What it establishes | Design use | Rights posture |
|---|---|---|---|---|
| [RANZCP Fellowship Program overview](https://www.ranzcp.org/training-exams-and-assessments/fellowship-program/program-overview) | updated 29 June 2026 | minimum 60-month, three-stage program with rotations, assessments and formal education | stage and workload comparator | citation/paraphrase |
| [RANZCP Stage 1](https://www.ranzcp.org/training-exams-and-assessments/fellowship-program/stage-1), [Stage 2](https://www.ranzcp.org/training-exams-and-assessments/fellowship-program/stage-2/assessments-and-requirements), and [Stage 3](https://www.ranzcp.org/training-exams-and-assessments/fellowship-program/stage-3) | pages updated June 2026 where stated | adult acute foundations; required child, consultation, addiction, older-adult, ECT, law, risk and cultural learning; senior electives and psychotherapy | breadth and progression audit | citation/paraphrase |
| [RANZCP Entrustable Professional Activities](https://www.ranzcp.org/training-exams-and-assessments/exams-assessments/rotation-assessments/entrustable-professional-activities) | current page checked 22 July 2026 | high-value observable professional tasks supported by multiple workplace assessments | triangulated performance evidence, with entrustment reserved for hosts | no reproduction of copyrighted tools |
| [RANZCP Fellowship competencies](https://www.ranzcp.org/training-exams-and-assessments/fellowship-program/fellowship-competencies-overview) | current page | biopsychosociocultural competence and developmental expectations | culture integrated with clinical work | citation/paraphrase |
| [RANZCP required online learning](https://www.ranzcp.org/training-exams-and-assessments/exams-assessments/self-paced-online-learning-requirements) | current requirements including 2026 changes | required Aboriginal and Torres Strait Islander mental-health learning and critical-thinking modules | local governance and assessed cultural safety; cannot be replaced by generic global content | citation/paraphrase; local Indigenous material requires appropriate authority |

RANZCP contributed protected supervision, task-level assessment, developmental descriptors, serious longitudinal psychotherapy, cultural safety, Indigenous mental-health requirements, reflective formation, and learner/program safety.

## 8. European harmonization

| Source | Version/status | What it establishes | Design use | Rights posture |
|---|---|---|---|---|
| [UEMS European Training Requirements for Psychiatry](https://www.uemspsychiatry.org/s/ETR2022-compiled.pdf) | September 2022 | minimum five years, progressive responsibility, cross-setting breadth, supervision, assessment and curriculum review | breadth and supervision comparator | UEMS site states rights reserved; cite/paraphrase only |
| [UEMS European Framework for Competencies in Psychiatry](https://www.uemspsychiatry.org/s/EUFRAMEWORKCOMPETENCYPAPER-lj6n.pdf) | 2022 publication | expert, communicator, collaborator, leader, advocate, scholar and professional roles | capability audit | rights reserved; cite/paraphrase |
| [UEMS Psychotherapy Learning Experiences](https://www.uemspsychiatry.org/s/UEMS-Psychiatry-Section-Guidelines-on-Psychotherapy-Learning-Experiences-in-Psychiatry-Training.pdf) | approved October 2019 | theory, broad supervised practice, outcome evaluation and protected/funded learning | distinguishes book knowledge from supervised therapy practice | rights reserved; no adaptation |
| [UEMS training reports hub](https://www.uemspsychiatry.org/section-training-reports) | current official hub | source discovery and later updates | audit route | rights reserved unless item says otherwise |

UEMS contributed a cross-setting minimum, recovery and human rights, a regular individual-supervision floor, working-time psychotherapy, and complete curriculum-to-assessment mapping.

## 9. Global, rights, and lived-experience sources

| Source | Version/status | What it contributes | Rights decision |
|---|---|---|---|
| [WPA: High Quality Post-Graduate Training in Psychiatry](https://www.wpanet.org/wp-content/uploads/2025/09/3.GA23.10.3-High-Quality-Psych-Training-Position-Statement-7.2.-14.6-1.pdf) | approved by WPA General Assembly 30 September 2023; document header notes earlier review history | breadth, protected learning, public mental health, research, accreditation, feedback, learner welfare, user/carer participation, supported decisions and coercion alternatives | cite/paraphrase; verify any later replacement |
| [WHO QualityRights guidance and training tools](https://www.who.int/publications/i/item/who-qualityrights-guidance-and-training-tools/) | published 12 November 2019 | recovery, legal capacity, right to decide, freedom from coercion/abuse, peer support and service transformation | normally CC BY-NC-SA 3.0 IGO; citation-only evidence in this CC BY corpus |
| [WHO/OHCHR Mental health, human rights and legislation](https://www.who.int/publications/i/item/9789240080737) | 9 October 2023 | dignity, consent, non-discrimination, community inclusion, deinstitutionalization and reducing/eliminating coercion | CC BY-NC-SA 3.0 IGO; do not adapt into CC BY prose |
| [WHO/Europe roadmap for lived-experience expertise](https://www.who.int/europe/publications/i/item/WHO-EURO-2025-12307-52079-79927) | 30 June 2025 | co-created actions for integrating lived and living experience into policy, services and workforce | inspect exact publication terms; cite as governance evidence |
| [WHO pre-service education guide for mental, neurological and substance-use care](https://www.who.int/publications/i/item/9789240104129/) | 14 April 2025 | competency-based feeder model across health professions | citation-only architecture reference pending exact reuse audit |
| [WHO Community Mental Health Services Guidance](https://www.who.int/publications/i/item/9789240025707) | 9 June 2021 | peer, outreach, crisis, supported-living and community service models | CC BY-NC-SA 3.0 IGO; citation only |
| [WHO Mental Health Atlas](https://www.who.int/teams/mental-health-and-substance-use/data-research/mental-health-atlas) | 2024 atlas published in 2025 | current global systems, workforce, financing and access context | WHO publication terms; cite facts, do not reuse graphics without item-level clearance |

The design inference is strong: lived-experience authority, supported decisions, recovery, human rights, community inclusion, public mental health, and alternatives to coercion must change required outcomes and assessment. They cannot remain an optional ethics lecture.

Commissioned first-person material must use explicit consent for open publication, negotiated attribution, compensation, accessibility, withdrawal terms before publication, support, and a clear CC BY licence. Scraping public stories is not an acceptable substitute.

## 10. Classification and clinical guidance

| Source | Current status | Educational value | Reuse decision |
|---|---|---|---|
| [ICD-11 browser](https://icd.who.int/browse11) and [2026 release notice](https://www.who.int/news/item/16-02-2026-icd-11-2026-release) | 2026 release announced 16 February 2026 | global classification, terminology, coding and comparison | [ICD-11 licence](https://icd.who.int/docs/icd-api/license/) is CC BY-ND 3.0 IGO; use unchanged codes/terms within terms, cite, and do not create a derivative classification |
| [WHO mhGAP Guideline](https://www.who.int/publications/i/item/9789240084278) | 2023 | evidence recommendations across common conditions and service contexts | CC BY-NC-SA 3.0 IGO; citation and independent synthesis only |
| [mhGAP Intervention Guide v2.0](https://www.who.int/publications/i/item/9789241549790) | 2016 edition; WHO page dated 24 June 2019 | non-specialist assessment and integrated management in global contexts | CC BY-NC-SA 3.0 IGO; do not adapt algorithms/tables into CC BY lessons |
| [APA practice-guideline portal](https://psychiatryonline.org/books/guidelines) | current portal | specialist guideline comparison in U.S. contexts | free clinical access is not CC BY reuse; cite, do not reproduce |
| [VA/DoD mental-health clinical-practice guidelines](https://www.healthquality.va.gov/guidelines/mh/) | collection includes current disorder- and safety-specific guidelines | recommendations, evidence tables, safety and shared-decision context | government-authored portions may be public domain in the U.S.; page-level audit third-party tools, figures and international status before reuse |
| [NICE guidance](https://www.nice.org.uk/guidance) | current guidance portal | comparison of pathways in UK contexts | UK-only open-content licence and separate international/AI terms; human citation review only; do not ingest into authoring agents without permission |

DSM-5-TR is proprietary. The curriculum teaches DSM literacy where necessary but does not reproduce criteria, tables, or wording. Diagnostic and rating instruments require their own rights registry; a citation or free download is not republication permission.

## 11. Reusable and open educational backbone

### 11.1 Preferred sources

| Source | Rights status | Suitable use | Required caution |
|---|---|---|---|
| [NIMH health topics](https://www.nimh.nih.gov/health/topics) and [NIMH site policy](https://www.nimh.nih.gov/site-info/policies) | NIMH-authored text is U.S. public domain; attribution requested | independently edited foundational explanations, current public research context and patient communication | NIMH images are not reusable; avoid implying endorsement and recheck updates |
| [NIMH RDoC educational resources](https://www.nimh.nih.gov/research/research-funded-by-nimh/rdoc/rdoc-educational-and-training-resources) | item-level federal/public-domain and third-party status must be checked | dimensional and transdiagnostic neuroscience examples | RDoC is a research framework, not a diagnostic system or settled ontology |
| [Open RN: Nursing—Mental Health and Community Concepts, 2e](https://www.ncbi.nlm.nih.gov/books/NBK616982/?report=reader) | CC BY 4.0 except where otherwise noted; second edition, 1 November 2025 | adaptable foundational teaching on communication, mental-status observation, community concepts and cases | level is nursing education, not specialist psychiatry; inspect every figure, table and embedded item |
| [FDA website](https://www.fda.gov/) and [FDA website policies](https://www.fda.gov/about-fda/about-website/website-policies) | FDA-authored material generally U.S. public domain unless marked otherwise | safety communications, regulatory evidence and medication warnings | labels, trademarks, contractor/third-party content and non-U.S. status require checks |
| [openFDA](https://open.fda.gov/) and [terms](https://open.fda.gov/terms/) | data/code generally CC0, with marked exceptions | reproducible regulatory-data teaching and adverse-event data literacy | spontaneous reports cannot establish incidence or causality |
| [VA Pharmacy Academic Detailing](https://www.pbm.va.gov/academicdetailingservicehome.asp) | VA states its academic-detailing materials are public domain | medication-safety, deprescribing and shared-decision teaching | inspect embedded third-party instruments, images and local recommendations |
| [SAMHSA](https://www.samhsa.gov/) and [SAMHSA FAQs](https://www.samhsa.gov/about/faqs) | most site information is public domain, but not all; publication distribution terms vary | addiction, recovery, trauma-aware systems, crisis services and peer support | item-level notices, logos, contractor content and fee restrictions |
| [SAMHSA peer-worker competencies](https://www.samhsa.gov/substance-use/recovery/peer-support-workers/core-competencies) | subject to SAMHSA item-level policy | recovery orientation and peer-role education | do not appropriate peer expertise into clinician-only authorship |
| [CDC reuse policy](https://www.cdc.gov/other/agencymaterials.html) | most CDC-authored material is U.S. public domain with conditions and exceptions | population health, prevention and surveillance | attribute, avoid endorsement, inspect contractor/third-party items, preserve statistical context |

### 11.2 Citation-only or incompatible educational sources

| Source | Why it is not general substrate |
|---|---|
| [MSU Introduction to Neuroscience](https://open.umn.edu/opentextbooks/textbooks/introduction-to-neuroscience-2022) | ShareAlike terms are not imported into the project's CC BY-only curriculum; cite and independently synthesize |
| [Fundamentals of Psychological Disorders](https://open.umn.edu/opentextbooks/textbooks/fundamentals-of-psychological-disorders) | CC BY-NC-SA and undergraduate scope; citation/background only |
| [PubMed Central](https://pmc.ncbi.nlm.nih.gov/about/copyright/) | no blanket open licence; each article's copyright and licence statement controls |
| [Cochrane](https://www.cochrane.org/about-us/our-open-access-strategy) | free access or an abstract is not derivative-use permission; only explicitly compatible items may be adapted |
| [LibreTexts terms](https://libretexts.org/terms-conditions) | licences vary by page and embedded component; never treat the platform as one reusable corpus |
| OpenStax books | excluded from model-assisted authoring because the publisher's book pages state an AI-ingestion/training restriction; no lesson should use them as agent input without permission |

### 11.3 Operational access signals

Robots files were checked only to avoid disrespectful automated access. WHO explicitly blocks multiple named AI crawlers; NICE legal terms separately restrict AI uses; FDA requests a crawl delay and provides openFDA as a preferred interface; PMC provides APIs and systematic-download rules. A missing or permissive robots file does not grant copyright, data-protection, contractual, or ethical permission. The project should use selected official pages and APIs, not bulk scraping.

## 12. Rights-critical content gaps

No comprehensive, current, residency-grade, permissively reusable source was found for:

1. full diagnostic criteria: DSM is proprietary and ICD-11 is NoDerivatives;
2. integrated modern clinical psychopharmacology at specialist depth;
3. a complete rights-cleared set of rating scales, cognitive tests and structured interviews;
4. complete psychotherapy manuals and worksheets across major modalities;
5. specialist-depth child, perinatal, older-adult, consultation-liaison, forensic, eating-disorder, neuropsychiatry and neuromodulation teaching;
6. jurisdiction-independent mental-health law;
7. consent-cleared clinical interviews and standardized-person media;
8. openly licensed lived-experience and survivor scholarship suitable for derivative lessons;
9. culturally governed psychiatric teaching outside dominant high-income-country frameworks;
10. a postgraduate question bank with transparent claim and rights provenance.

These are commissioning and partnership needs, not invitations to stretch introductory OER or copy restricted material.

## 13. Synthesis decisions

| Evidence pattern | Curriculum decision |
|---|---|
| Most specialist systems use staged development but differ in duration and sequence | use four purpose-defined academic stages rather than claiming equivalence to PGY/CT/ST years |
| Direct observation, multiple assessors and multiple contexts recur | assess simulations across cases and observers; reserve real-patient entrustment for accredited hosts |
| Medicine, neurology, emergencies, longitudinal outpatient care and subspecialty breadth recur | keep them in the common core, not optional enrichment |
| Psychotherapy expectations vary but the strongest systems require sustained supervised cases | teach psychotherapy science and simulation longitudinally; state that reading/role-play is not patient treatment |
| Culture, Indigenous mental health, rights and lived experience are strongest when assessed and governed | make them longitudinal; require local governance, paid participation and rights safeguards |
| Modern digital/AI, measurement, implementation and service-design needs are inconsistent | make them explicit core modules with refusal/withdrawal criteria and privacy safeguards |
| National law and certification are irreducibly local | label jurisdiction/date and prohibit equivalence claims |
| No single source is both comprehensive and CC BY-compatible | use a layered evidence architecture and commission missing original material |

## 14. Minimum source architecture for future lessons

Each claim-bearing production lesson should include:

- one current official or primary source for safety, classification, regulatory, epidemiologic, or legal claims where relevant;
- one current evidence synthesis or guideline plus pivotal/corrective research for intervention claims;
- evidence of harms, burden, withdrawal, access, equity, and patient-important outcomes;
- qualitative or lived-experience evidence where the claim concerns meaning, acceptability, coercion, recovery, or service experience;
- a version, date, jurisdiction, population, uncertainty, conflict, licence and reuse decision;
- a current medication-label or safety check for time-sensitive pharmacology;
- an instrument-level rights check before reproducing any scale, item, scoring rule or worksheet;
- explicit separation of source fact, author synthesis, contested judgment and local rule.

No single textbook, guideline, diagnostic manual, model output, or expert opinion may serve as the lesson's entire evidentiary spine.

## 15. Audit schedule

| Source class | Minimum recheck |
|---|---|
| medication warnings, crisis resources, law, regulation, AI policy | at use and at least every 6 months while a lesson is active |
| diagnostic classifications, national training standards, major guidelines | annually and on reported revision |
| epidemiology, service data and technology claims | annually or at each new release |
| foundational mechanisms and education evidence | every 24 months, sooner after a major corrective review |
| licences, terms, robots and AI-use restrictions | at ingestion, at publication, and before material reuse |

Corrections should preserve the original claim, correction date, source, affected lessons, and migration action. Silent replacement is not acceptable for safety-critical content.

## 16. Next research work

Before Psychiatry lesson intake opens:

1. obtain multi-jurisdiction specialist, resident, multidisciplinary, education, rights, accessibility and lived-experience review of the syllabus;
2. commission culturally governed and lived-experience material under explicit CC BY terms;
3. create an instrument-rights registry and medication-source update protocol;
4. build a machine-readable outcome graph, prerequisite map, misconception map and assessment blueprint;
5. define a clinical-host interface that cannot be mistaken for EmbeddedKnowledge clinical credentialing;
6. pilot representative lessons in interviewing, formulation, psychopharmacology, psychotherapy, emergencies, culture/rights and digital psychiatry before scaling the corpus;
7. audit non-English and lower-resource training frameworks to correct the current source imbalance.


---

## Repository file: research/CONTENT-AUTHORING-EVIDENCE-2026-07-19.md

# Evidence synthesis: writing university-grade learning content

**Research date:** 2026-07-19
**Question:** What does research imply for writing web-first, exportable academic lessons that maximize clarity, durable learning, and transfer for novice university learners, especially in STEM and Premed?
**Product informed:** [`CONTENT-STANDARD.md`](../CONTENT-STANDARD.md)

## Executive conclusion

The evidence does not support a single “textbook style” or a magic word count. It supports a system of mutually reinforcing design choices:

1. define the learner, outcome, prerequisite boundary, and evidence of mastery;
2. build a coherent explanatory model around a question or phenomenon;
3. make causal and logical relations explicit for novices;
4. coordinate relevant representations rather than accumulate media;
5. model expert decisions with worked examples, then fade support;
6. make learners retrieve, explain, compare, predict, and apply;
7. repair consequential misconceptions with evidence and an alternative model;
8. distribute and mix later practice when the learner is ready to discriminate;
9. give information-rich feedback and a recovery route; and
10. preserve accessibility, epistemic scope, and a sustained-reading path.

Readable exposition is necessary but not sufficient. A lesson can feel clear because it is familiar or easy to reread while producing weak delayed recall or transfer. Conversely, effort is not automatically productive: novices need guidance, and difficulty becomes useful only when the learner has enough knowledge and support to perform the intended cognitive operation.

## Method and limitations

This is a rapid, decision-oriented evidence synthesis, not a registered systematic review. Searches prioritized:

- meta-analyses and systematic reviews;
- evidence-rated practice guides and consensus frameworks;
- primary experiments for mechanisms or especially relevant boundary conditions; and
- authoritative accessibility standards.

The search covered expository-text comprehension, prior knowledge, worked examples, self-explanation, retrieval, spacing, interleaving, comparison, conceptual change, feedback, multimedia design, digital reading, concept maps, active learning, accessibility, and unsupported learning-style matching.

Important limitations:

- Much learning research uses short laboratory materials, K–12 learners, undergraduate convenience samples, or immediate tests. It does not directly estimate the effect of a full Premed corpus.
- Meta-analytic effect sizes below come from different outcomes, control conditions, durations, and populations. They must not be ranked as if they shared one scale and design.
- Educational interventions are heterogeneous. “Feedback,” “active learning,” “multimedia,” and “interleaving” are families of treatments, not single ingredients.
- Publication bias, study quality, implementation fidelity, prior knowledge, and assessment alignment can change results.
- Accessibility standards establish an ethical and technical floor; they are not in themselves estimates of learning gains.
- The final rules remain hypotheses about this product until tested with representative learners using delayed, transfer-sensitive outcomes and accessibility evaluation.

Evidence labels in this report are practical judgments:

- **Robust default:** convergent meta-analysis, evidence-rated synthesis, or repeated experimental support with a clear application.
- **Conditional:** useful average effect with important moderators, a narrower evidence base, or a design choice that depends strongly on learner/task.
- **Do not mandate:** insufficient evidence for a universal rule or affirmative evidence against the common claim.

## 1. Start from target ideas, learner knowledge, and purpose

**Finding — robust default.** Materials should be evaluated outcome by outcome, not by whether the chapter “covers” a topic. AAAS Project 2061's instructional analysis asks whether material establishes purpose, attends to prerequisites and student ideas, engages relevant phenomena, develops and uses scientific ideas, promotes reasoning, assesses progress, and supports all students. The National Academies' science framework similarly prioritizes depth and the integration of core ideas, practices, and crosscutting concepts over isolated fact accumulation.

**Implication.** Before prose, an author needs a target model, declared learner, prerequisite route, likely learner ideas, relevant phenomena, and observable mastery evidence. The lesson sequence must build toward that target; adjacent true facts are not automatically relevant.

**Sources.** [National Academies Framework for K–12 Science Education](https://nap.nationalacademies.org/resource/13165/interactive/); [National Academies excerpt of the Project 2061 instructional analysis](https://www.nationalacademies.org/read/9607/chapter/6); [AAAS effective learning and teaching principles](https://www.project2061.org/publications/rsl/online/SFAA/CHAP13.HTM).

## 2. Coherence should be calibrated to the learner

**Finding — robust default for a novice course.** Comprehension depends on an interaction between text and reader. High-cohesion science text makes relations explicit through conceptual overlap, causal/temporal organization, connectives, headers, and topic sentences. In a college biology study, higher cohesion improved text-based comprehension, while prior knowledge and reading skill materially affected outcomes. Earlier work found that low-knowledge readers tended to benefit from coherent text, while additional explicitness can be redundant or reduce generative inference for high-knowledge readers.

**Implication.** Premed's default route should be written for the declared novice: explicit causal links, stable terminology, intermediate reasoning, and purposeful headings. Expert routes may compress or hide scaffolds, but the project should not create gaps merely to make prose feel sophisticated.

**Caution.** “Make it coherent” does not mean “make it longer” or “state every inference.” Added explanation can increase density, and skilled readers differ in how well they exploit it. Diagnostic/adaptive presentation is preferable to one density for all expertise levels.

**Sources.** [Ozuru, Dempsey, and McNamara, 2009, *Learning and Instruction*](https://doi.org/10.1016/j.learninstruc.2008.04.003); [McNamara et al., 1996, *Cognition and Instruction*](https://doi.org/10.1207/s1532690xci1401_1).

## 3. Plain language helps; readability formulas do not define comprehension

**Finding — conditional but important.** A randomized trial found that plain-language versions improved adults' understanding of health recommendations. W3C cognitive-accessibility guidance recommends clear common words, explanation of jargon and abbreviations, short blocks, literal language, and explicit implied content. However, traditional readability formulas largely count word and sentence features and omit prior knowledge, cohesion, conceptual density, syntax, purpose, and layout. Evidence and government guidance warn against treating a grade-level score as an overall comprehension measure.

**Implication.** Prefer precise common words, define necessary technical language, remove ornamental complexity, and test comprehension. Keep scientific terms the student must learn. Use a readability score only to flag passages for human inspection; never optimize to it by deleting logical connectives or substituting inaccurate familiar words.

**Sources.** [Sayfi et al., 2024 randomized trial](https://pubmed.ncbi.nlm.nih.gov/38008266/); [W3C: Use Clear Words](https://www.w3.org/WAI/WCAG2/supplemental/patterns/o3p01-clear-words/); [AHRQ: Use Caution With Readability Formulas](https://www.ahrq.gov/talkingquality/resources/writing/tip6.html).

## 4. Novices benefit from worked examples; assistance should fade

**Finding — robust default.** A 2023 meta-analysis of 55 studies and 181 effects found a medium average benefit of worked examples for mathematics performance (`g = 0.48`). Earlier synthesis found that faded solution steps and self-explanation support were particularly useful features. Worked examples reduce the need for a novice to search blindly while also holding problem elements in working memory.

**Boundary condition.** A 2025 meta-analysis of 60 experimental studies found a robust expertise reversal pattern: low-prior-knowledge learners benefited from high assistance (`d = 0.505`), while high-prior-knowledge learners did better with lower assistance (`d = -0.428`). The practical asymmetry matters: giving novices adequate guidance appears more important than perfectly removing every scaffold for experts.

**Implication.** Initial instruction should show task representation, model selection, plan, execution, checks, and interpretation. Then move through completion problems to independent and mixed problems. A worked example that displays only arithmetic or an answer does not provide the expert model the evidence concerns.

**Sources.** [Barbieri et al., 2023 worked-examples meta-analysis](https://eric.ed.gov/?id=EJ1364058); [Crissman, 2006 worked-examples meta-analysis](https://digitalcommons.unl.edu/dissertations/AAI3208114/); [Tetzlaff et al., 2025 expertise-reversal meta-analysis](https://doi.org/10.1016/j.learninstruc.2025.102142).

## 5. Self-explanation is useful when prompts target principles and gaps

**Finding — conditional.** Self-explanation can prevent superficial example study by asking learners to connect a step to a principle, prior knowledge, a goal, or the next step. A mathematics meta-analysis supports prompted self-explanation but emphasizes design conditions; simply asking “explain” can produce paraphrase rather than inference. Instructional explanations do not automatically add value to every example, and poorly targeted prompts add load.

**Implication.** Put a small number of prompts at decisive points: “Why is this model applicable?”, “What changes between these lines?”, “Which assumption permits this step?”, or “How would the result change if…?” Supply feedback or a model explanation after a genuine attempt.

**Sources.** [Rittle-Johnson, Loehr, and Durkin, 2017 meta-analysis](https://doi.org/10.1007/s11858-017-0834-z); [Renkl, 2002 worked examples and instructional explanations](https://doi.org/10.1016/S0959-4752(01)00030-5); [Wittwer and Renkl, 2010 meta-analytic review](https://eric.ed.gov/?id=EJ906661).

## 6. Retrieval produces more durable learning than repeated exposure

**Finding — robust default.** In experiments with prose passages, repeated study produced better performance after five minutes, but retrieval produced substantially better retention after two days or one week. Meta-analyses have since found a broad positive testing effect across formats and conditions. Retrieval can also support inferential transfer rather than only memory for the exact answer.

**Implication.** Questions should occur before answer reveal; delayed and cumulative retrieval must be built across lessons. A lesson-end recognition quiz is not the whole intervention. Ask learners to reconstruct relations, explanations, predictions, representations, and procedures. Provide corrective reasoning and a route to restudy.

**Caution.** Practice tests differ in retrievability, format, feedback, stakes, and final-test match. Feedback is especially important for correcting errors even though some meta-analyses find retrieval benefits without it.

**Sources.** [Roediger and Karpicke, 2006 prose experiments](https://www.psychologicalscience.org/journals/psychological-science/j.1467-9280.2006.01693.x/); [Rowland, 2014 meta-analysis](https://pubmed.ncbi.nlm.nih.gov/25150680/); [Adesope, Trevisan, and Sundararajan, 2017 meta-analysis](https://doi.org/10.3102/0034654316689306); [Butler, 2010 transfer experiments](https://pubmed.ncbi.nlm.nih.gov/20804289/).

## 7. Space retrieval and use interleaving for discrimination

**Finding — robust for spacing, conditional for interleaving.** A quantitative synthesis of 317 experiments found that spaced presentations generally improve later retention and that the best interval depends on the desired retention interval. The IES practice guide rates spacing as supported by moderate evidence. Interleaving has a positive average effect but is strongly moderated by the similarity and type of material; it is especially valuable when learners must discriminate among confusable categories or strategies.

**Implication.** Revisit high-value outcomes after delays across the course. Use blocked initial acquisition when a novice is learning a procedure, then interleave once choosing the correct concept or strategy becomes the target. “Random mixture from the first problem” is not a universal rule.

**Sources.** [Cepeda et al., 2006 distributed-practice meta-analysis](https://www.escholarship.org/content/qt3rr6q10c/qt3rr6q10c.pdf); [IES Organizing Instruction and Study practice guide](https://ies.ed.gov/ncee/wwc/PracticeGuide/1); [Brunmair and Richter, 2019 interleaving meta-analysis](https://pubmed.ncbi.nlm.nih.gov/31556629/).

## 8. Comparison reveals structure and supports abstraction

**Finding — robust/conditional.** A meta-analysis of 57 experiments found that comparing cases improved learning over sequential, single-case, nonanalogous, traditional-instruction, and control conditions on average (`d = 0.50`). Benefits were larger when learners looked for similarities and when a principle followed the comparison. Evidence-rated guidance also supports explicitly connecting concrete and abstract representations. Concreteness fading is promising for transfer but is not equally studied across all advanced science domains.

**Implication.** Use multiple examples when one case could hide the governing relation. Compare analogous cases to extract common structure and contrasting cases to learn a boundary. Map a concrete phenomenon to notation, then apply the abstraction elsewhere. Do not let colorful concrete features become the rule the learner remembers.

**Sources.** [Alfieri, Nokes-Malach, and Schunn, 2013 case-comparison meta-analysis](https://doi.org/10.1080/00461520.2013.775712); [IES practice-guide recommendation on concrete and abstract representations](https://ies.ed.gov/ncee/WWC/Docs/PracticeGuide/20072004.pdf); [Fyfe et al., 2014 concreteness-fading systematic review](https://eric.ed.gov/?id=EJ1036777).

## 9. Misconceptions need refutation plus a replacement model

**Finding — robust default when a misconception is relevant.** A random-effects meta-analysis of 44 comparisons and 3,869 learners found a moderate benefit for refutation text (`g = 0.41`) over other learning conditions. A refutation structure identifies the mistaken conception, contradicts it, explains why it fails, and provides a more adequate account. Biology-specific synthesis also found strong effects but noted that many studies were small and nonrandomized.

**Implication.** Select consequential, evidenced learner ideas; elicit a prediction; show the evidence or consequence the old model cannot explain; construct the alternative; and recheck it in another case. Merely naming “the misconception” or displaying a correction is insufficient.

**Sources.** [Schroeder and Kucera, 2022 refutation-text meta-analysis](https://eric.ed.gov/?id=EJ1334754); [Aleknavičiūtė, Lehtinen, and Södervik, 2023 biology review and meta-analysis](https://doi.org/10.1016/j.edurev.2023.100556); [Kendeou, Walsh, Smith, and O'Brien, 2017 textual/graphical refutation experiments](https://doi.org/10.1016/j.cedpsych.2017.03.007).

## 10. Relevant visuals help when words and pictures are integrated

**Finding — robust default with content-specific choices.** A 2022 overview of 29 reviews covering 1,189 studies and 78,177 participants found positive meta-analytic support for multiple multimedia design principles. Particularly durable findings include signaling important structure and spatially integrating corresponding words and visual elements. A meta-analysis of 58 comparisons found an overall benefit for integrated over split-attention designs (`g = 0.63`).

**Implication.** A visual should reveal a relation, pattern, spatial structure, process, scale, or evidence that prose alone handles poorly. Put labels and explanations near what they describe, cue the relevant feature, and explicitly connect representations. Preserve an equivalent text/table route.

**Boundary conditions.** The traditional redundancy rule is not universal, especially where captions are required for accessibility, the learner controls media, or the text supplies a needed alternative. More than two representations can help in STEM, but a 2024 meta-analysis found a small performance advantage and emphasized appropriate support. Representation count is not a quality metric.

**Sources.** [Noetel et al., 2022 meta-meta-analysis](https://eric.ed.gov/?id=EJ1338120); [Schroeder and Cenkci, 2018 spatial-contiguity meta-analysis](https://eric.ed.gov/?id=EJ1186641); [Schneider et al., 2018 signaling meta-analysis](https://doi.org/10.1016/j.edurev.2017.11.001); [Rexigel et al., 2024 multiple-representations meta-analysis](https://doi.org/10.1007/s10648-024-09958-y).

## 11. Remove irrelevant “seductive” detail; earn interest through the idea

**Finding — robust average, conditional in application.** A meta-analysis of 39 effects found small-to-medium harms to retention and medium harms to transfer from interesting but instructionally irrelevant details. Experiments suggest that such details can capture reading time and disrupt the coherent model. More recent work finds motivational trade-offs under some conditions, so the conclusion is not “make everything visually austere.”

**Implication.** Use a case, story, visual, or historical detail when it establishes purpose, supplies evidence, instantiates a relation, supports transfer, or develops scientific practice. Remove it when its only job is to be memorable. Attractive color and emotional design can be modestly useful, but they must signal and support the learning structure rather than compete with it.

**Sources.** [Rey, 2012 seductive-details meta-analysis](https://doi.org/10.1016/j.edurev.2012.05.003); [Sanchez and Wiley, 2007 processing experiment](https://doi.org/10.1016/j.cedpsych.2006.07.002); [Wong and Adesope, 2021 emotional-design meta-analysis](https://eric.ed.gov/?id=EJ1295978).

## 12. Feedback works through information, not praise alone

**Finding — robust average, heterogeneous implementation.** A meta-analysis of 435 studies, 994 effects, and more than 61,000 learners found a medium average feedback effect (`d = 0.48`) with substantial heterogeneity. Information content matters: task, process, and self-regulation feedback are distinct from comments about the person.

**Implication.** After a genuine attempt, state correctness, explain the governing principle or error, and prescribe a useful next action. Answer-only correctness can be enough for a narrow fact but is insufficient for conceptual and procedural errors. Praise can support affect but does not replace diagnostic information.

**Sources.** [Wisniewski, Zierer, and Hattie, 2020 feedback meta-analysis](https://pmc.ncbi.nlm.nih.gov/articles/PMC6987456/); [Medical-education feedback RCT meta-analysis](https://pmc.ncbi.nlm.nih.gov/articles/PMC8651958/).

## 13. Active learning means relevant cognitive work

**Finding — robust at course level, not a license for activity for its own sake.** A meta-analysis of 225 undergraduate STEM studies found higher performance and lower failure under active learning than traditional lecturing. This broad category includes problem solving, formative assessment, discussion, and other designs, so it does not identify a single optimal interaction.

**Implication.** The web lesson should require learners to retrieve, predict, explain, compare, construct, calculate, interpret, or choose a model. Clicking, dragging, or revealing without relevant thinking does not instantiate the evidence. Initial generative activity must be adequately guided for novices.

**Sources.** [Freeman et al., 2014 STEM active-learning meta-analysis](https://doi.org/10.1073/pnas.1319030111); [IES instruction-and-study practice guide](https://ies.ed.gov/ncee/wwc/PracticeGuide/1).

## 14. Concept maps can support science learning when guided

**Finding — conditional.** A 2024 meta-analysis of 55 studies and 5,364 K–12 learners found a moderate overall benefit for concept mapping in science (`g = 0.776`) with significant heterogeneity; stronger guidance was among the favorable conditions. The evidence concerns activities and instruction using concept maps, not proof that displaying a large finished network produces the same learning.

**Implication.** Use the knowledge graph for orientation, prerequisites, local relationships, retrieval, and learner construction. Show a small relevant subgraph, prompt the learner to predict or explain links, and connect it to the current model. Do not treat the complete 404-node graph as a substitute for instruction.

**Source.** [Anastasiou, Wirngo, and Bagos, 2024 concept-map meta-analysis](https://doi.org/10.1007/s10648-024-09877-y).

## 15. Digital reading needs a sustained-reading path

**Finding — conditional product guidance.** Meta-analyses report a small average comprehension advantage for paper over screens, especially for expository text and time pressure; Delgado et al. reported `g = -0.21` for digital relative to paper. A 2025 network meta-analysis found paper advantages were larger when digital reading required scrolling. Device, genre, navigation, interactivity, time, and reader behavior moderate results.

**Implication.** Retain printable/exportable source, remove artificial time pressure, preserve stable location and backtracking, and offer a continuous reading mode alongside guided frames. Page-like framing is a presentation strategy, not permission to fragment a causal explanation or force every semantic unit into one viewport.

**Sources.** [Delgado et al., 2018 paper-versus-screen meta-analysis](https://doi.org/10.1016/j.edurev.2018.09.003); [Clinton, 2019 systematic review and meta-analysis](https://doi.org/10.1111/1467-9817.12269); [Clinton-Lisell and Litzinger, 2025 network meta-analysis](https://doi.org/10.1007/s10639-025-13843-8).

## 16. Accessibility alternatives should preserve the learning operation

**Finding — authoritative standard.** WCAG 2.2 requires text alternatives, adaptable relationships and sequence, keyboard access, navigability, and understandable content. W3C guidance for complex images recommends short identification plus a long description that exposes values, relationships, trends, labels, and structure, ideally available to everyone.

**Implication.** A figure alternative must allow the learner to achieve the same outcome; “diagram of the Krebs cycle” does not. Equations need semantic math and prose interpretation, charts need underlying tables and trend explanations, and interactions need keyboard/non-drag equivalents. Accessibility is a content property, not a post-render checkbox.

**Sources.** [WCAG 2.2 overview](https://www.w3.org/WAI/WCAG22/Understanding/); [W3C complex-images tutorial](https://www.w3.org/WAI/tutorials/images/complex/); [W3C page-structure tutorial](https://www.w3.org/WAI/tutorials/page-structure/).

## 17. Do not match supposed learning styles

**Finding — do not mandate.** The review of the learning-styles “meshing” hypothesis found no adequate evidentiary basis for assigning instruction to self-reported visual, auditory, or kinesthetic styles. Preferences can matter to experience and accessibility, and the subject matter can demand a representation, but that is different from stable style matching.

**Implication.** Choose media because the content and operation need them; provide alternatives for access and useful learner control. Do not label a learner or create separate scientific explanations around an unsupported style taxonomy.

**Source.** [Pashler et al., 2008, *Psychological Science in the Public Interest*](https://www.psychologicalscience.org/journals/pspi/j.1539-6053.2009.01038.x/).

## 18. Evidence-to-rule matrix

| Design decision | Evidence status | EmbeddedKnowledge default |
|---|---|---|
| Declare outcome, learner, prerequisites, and purpose | Robust default | Required authoring brief and alignment gate |
| High-cohesion explanatory prose for novices | Robust default | Explicit relations, stable terms, intermediate reasoning |
| Fixed readability grade or universal word count | Do not mandate | Use human review and learner testing instead |
| Worked examples before independent complex problems | Robust default | Complete model, then faded support |
| Self-explanation prompts | Conditional | Prompt decisive principles/steps and provide feedback |
| Retrieval before answer reveal | Robust default | Required, low-stakes, cumulative, delayed |
| Spacing across lessons | Robust default | Map later retrieval in course graph/sequence |
| Interleaving | Conditional | Use for discriminating known categories/strategies |
| Compare analogous and contrasting cases | Robust/conditional | Use where a single case hides structure or boundary |
| Concrete-to-abstract mapping | Conditional | Explicit mapping and later application |
| Refutation text | Robust when misconception is real | Elicit, contradict with evidence, replace, recheck |
| Relevant graphics plus words | Robust default | Signal and spatially integrate; preserve equivalent route |
| More representations | Conditional | Add only with an instructional job and integration support |
| Decorative stories/images/facts | Usually harmful | Exclude unless they serve purpose, evidence, model, or transfer |
| Information-rich feedback | Robust average | Correctness + principle/process + next action |
| Active learner generation | Robust average | Relevant thought, not interaction count |
| Complete knowledge graph as instruction | Unsupported inference | Use guided local maps and learner explanation/construction |
| Guided frames plus continuous reading/export | Conditional product decision | Provide both; never truncate meaning to fit a frame |
| Accessibility-equivalent content and operation | Required standard | Content-level acceptance gate |
| Learning-style matching | Unsupported | Prohibited as an authoring rationale |

## 19. What should be tested next

The research cannot select the final EmbeddedKnowledge lesson design without product evidence. The first real lessons should be evaluated with representative learners using:

1. prerequisite and prior-knowledge measures;
2. immediate comprehension and explanation;
3. delayed retrieval after at least several days;
4. near and bounded-far transfer;
5. misconception-sensitive items and confidence;
6. reading-mode versus guided-mode behavior without forcing either mode;
7. location, backtracking, abandonment, and recovery-route use;
8. keyboard, screen-reader, reflow, low-vision, and reduced-motion task completion;
9. qualitative think-aloud evidence about hidden inference gaps; and
10. item and content revision based on errors, not completion metrics alone.

The project should preregister or at least version the question, outcome, analysis, and candidate content before using learner data to claim effectiveness. Completion rate, preference, reading speed, and self-reported clarity are useful experience measures but cannot replace delayed learning and transfer outcomes.


---

## Repository file: site/agent/contribution.json

{
  "schemaVersion": 1,
  "protocol": "EmbeddedKnowledge Agent Contribution Protocol",
  "protocolVersion": "3.0.0-founding",
  "status": "repository-public-intake-open",
  "canonicalBaseUrl": "https://embeddedknowledge.io/",
  "canonicalWriteMechanism": "pull-request",
  "repository": "https://github.com/VSBDev/EmbeddedKnowledge",
  "contentLicense": "CC-BY-4.0",
  "courses": {
    "premed": {
      "intake": "open",
      "knowledgeGraph": "/data/premed-graph.json",
      "lessonIndex": "/data/premed-lessons.json",
      "openLessonPullRequests": "/data/premed-open-prs.json",
      "progress": "/data/premed-progress.json",
      "terminology": "/data/premed-terminology.json"
    },
    "psychiatry": {
      "intake": "open",
      "knowledgeGraph": "/data/psychiatry-graph.json",
      "lessonIndex": "/data/psychiatry-lessons.json",
      "openLessonPullRequests": "/data/psychiatry-open-prs.json",
      "progress": "/data/psychiatry-progress.json",
      "terminology": "/data/psychiatry-terminology.json",
      "boundary": "Academic study and simulation only; no clinical credential or authority."
    }
  },
  "agentEntryPoints": {
    "conciseContext": "/llms.txt",
    "fullContext": "/llms-full.txt",
    "humanProtocol": "/contribute/",
    "agentSkills": "/agent/skills.json",
    "lessonAuthorSkill": "/skills/author-embeddedknowledge-lesson/SKILL.md",
    "learningContentStandard": "/content-standard.txt",
    "sourceReusePolicy": "/rights-policy.txt",
    "lessonFormatGuide": "/contribute/format/",
    "lessonFormatSpecimen": "/premed/lessons/specimen/",
    "progress": "/data/premed-progress.json",
    "knowledgeGraph": "/data/premed-graph.json",
    "lessonIndex": "/data/premed-lessons.json",
    "openLessonPullRequests": "/data/premed-open-prs.json",
    "quorumPolicy": "/agent/quorum-policy.json"
  },
  "schemas": {
    "lesson": "/schemas/lesson.schema.json",
    "assessment": "/schemas/assessment.schema.json",
    "references": "/schemas/references.schema.json",
    "claims": "/schemas/claims.schema.json",
    "glossary": "/schemas/glossary.schema.json",
    "diagram": "/schemas/diagram.schema.json",
    "review": "/schemas/review.schema.json",
    "adjudication": "/schemas/adjudication.schema.json",
    "lessonIndex": "/schemas/lesson-index.schema.json",
    "openLessonPullRequests": "/schemas/open-lesson-prs.schema.json",
    "agentSkills": "/schemas/agent-skills.schema.json"
  },
  "agentSkills": {
    "standard": "https://agentskills.io/specification",
    "canonicalDirectory": ".agents/skills",
    "manifest": "/agent/skills.json",
    "roleIsolationRequired": true
  },
  "webmcp": {
    "status": "progressive-enhancement",
    "mode": "read-only",
    "requiresSecureContext": true,
    "writeOperations": "Pull requests only",
    "tools": [
      "embeddedknowledge.get_project_status",
      "embeddedknowledge.get_premed_progress",
      "embeddedknowledge.get_outcome",
      "embeddedknowledge.list_uncovered_outcomes",
      "embeddedknowledge.get_lesson_state",
      "embeddedknowledge.list_lesson_states",
      "embeddedknowledge.get_contribution_contract"
    ]
  },
  "standardLessonQuorum": {
    "mode": "founding-stage-one-pass-finalization",
    "eligibleReviewInputs": 2,
    "eligibleVerdicts": ["approve", "request-changes"],
    "minimumApprovingReviews": 0,
    "academic": 1,
    "learningDesign": 1,
    "accessibilityRightsReview": 0,
    "minimumDistinctAgentRuns": 2,
    "minimumDistinctAgentModelFamilies": 2,
    "additionalFreshFinalizationRun": 1,
    "finalizerMayReviseContentOnce": true,
    "finalizerAccessibilityRightsAuditRequired": true,
    "allFindingsRequireDisposition": true,
    "singleAccountablePrincipalMayOperateRuns": true
  },
  "lessonFormat": {
    "id": "embeddedknowledge-lesson-v1",
    "lessonSchemaVersion": 3,
    "canonicalSceneSourcePattern": "content/*.md",
    "sceneKinds": [
      "orientation", "diagnostic", "concept", "definition", "derivation", "worked-example",
      "investigation", "retrieval-check", "misconception", "practice", "transfer", "synthesis",
      "assessment", "references"
    ],
    "directiveWhitelist": [
      "definition", "theorem", "derivation", "worked-example", "check", "misconception",
      "investigation", "figure", "diagram", "equation", "chemistry", "source-note", "callout"
    ],
    "rawHtmlAllowed": false,
    "executableContentAllowed": false,
    "sceneClaimCoverage": ["claims-mapped", "no-material-claims"],
    "learnerVisibleClaimSourcesRequired": true,
    "publishedClaimsMustBeReviewed": true,
    "referencesSchemaVersion": 2,
    "sourceUseAndAgentAccessDeclarationsRequired": true,
    "permittedExpressiveReuseBases": ["cc-by-4.0", "cc0-1.0", "public-domain", "written-permission"],
    "specimenSource": "examples/lesson-pack/",
    "specimenCountsTowardCoverage": false
  },
  "lessonPackFiles": [
    "lesson.json", "content/*.md", "assessment.json", "references.json", "claims.json",
    "glossary.json", "diagrams/*.diagram.json", "assets/**/*", "ATTRIBUTION.md",
    "reviews/*.json", "adjudication.json"
  ],
  "submissionState": "Premed and Psychiatry lesson contributions are open through focused pull requests. Psychiatry remains academic study and simulation only. Branch protection, the two-review plus fresh-finalizer workflow, structured GitHub provenance, deterministic validation, and the MIT software and CC BY 4.0 content licences are active."
}


---

## Repository file: site/agent/skills.json

{
  "schemaVersion": 1,
  "standard": "https://agentskills.io/specification",
  "canonicalDirectory": ".agents/skills",
  "repositoryRequired": true,
  "discovery": {
    "generic": "Read this manifest, then load the matching SKILL.md explicitly.",
    "geminiCli": "Workspace discovery supports the .agents/skills alias.",
    "claudeCode": "Install the bundle or load the canonical SKILL.md explicitly.",
    "otherClients": "Use the downloadable bundle when custom skills are supported; otherwise use the raw instructions URL."
  },
  "skills": [
    {
      "name": "author-embeddedknowledge-lesson",
      "role": "author",
      "description": "Author one clear, human-readable, evidence-based EmbeddedKnowledge Lesson Format v1 pack for an uncovered Premed or Psychiatry outcome. Use when asked to draft, build, revise, or prepare a lesson contribution, its learner-facing scenes, assessment, claims, references, glossary, diagrams, accessible assets, attribution, or author provenance before independent review.",
      "invocation": "$author-embeddedknowledge-lesson",
      "source": ".agents/skills/author-embeddedknowledge-lesson/SKILL.md",
      "instructions": "/skills/author-embeddedknowledge-lesson/SKILL.md",
      "bundle": "/skills/author-embeddedknowledge-lesson/author-embeddedknowledge-lesson.zip",
      "instructionsSha256": "sha256:9a3cdd1f2968da92dbb8a5bf5c867fdd6f8494c6c909c6d981d1e5ad6ee30290",
      "bundleSha256": "sha256:3be10907c70360dcd8f7194de848acbad6bc8c46050d0a54cee390f4d77f30df"
    },
    {
      "name": "review-embeddedknowledge-academic",
      "role": "academic-review",
      "description": "Independently review one frozen EmbeddedKnowledge lesson candidate for academic accuracy, evidence, calculations, model boundaries, uncertainty, outcome scope, and assessment correctness. Use when assigned the academic quorum role and asked to produce a structured academic review artifact against an exact candidate commit.",
      "invocation": "$review-embeddedknowledge-academic",
      "source": ".agents/skills/review-embeddedknowledge-academic/SKILL.md",
      "instructions": "/skills/review-embeddedknowledge-academic/SKILL.md",
      "bundle": "/skills/review-embeddedknowledge-academic/review-embeddedknowledge-academic.zip",
      "instructionsSha256": "sha256:5af0bfabbaacb699f8d07b6d12e1cd1399bf114db1fda450b7b5016f56c3eb91",
      "bundleSha256": "sha256:9ada67bd5ceef4a0316a2fe1d42144b5c79a193116d7a2e3e7e359e0efb0374a"
    },
    {
      "name": "review-embeddedknowledge-learning-design",
      "role": "learning-design-review",
      "description": "Independently review one frozen EmbeddedKnowledge lesson candidate for first-read clarity, natural learner-facing prose, learner fit, explanatory coherence, cognitive support, worked examples, retrieval, practice, feedback, misconception repair, transfer, assessment alignment, mastery, and remediation. Use when assigned the learning-design quorum role and asked for a structured review against an exact candidate commit.",
      "invocation": "$review-embeddedknowledge-learning-design",
      "source": ".agents/skills/review-embeddedknowledge-learning-design/SKILL.md",
      "instructions": "/skills/review-embeddedknowledge-learning-design/SKILL.md",
      "bundle": "/skills/review-embeddedknowledge-learning-design/review-embeddedknowledge-learning-design.zip",
      "instructionsSha256": "sha256:5e75974677d92e46bbe0da25bd591879b61a02ceef08e2c53d55a4e912bbb9b4",
      "bundleSha256": "sha256:47b3e713d16c2a8e4d9a821bd946bf07ae7fdd7dcb884b4892d3df43d9adaf53"
    },
    {
      "name": "review-embeddedknowledge-accessibility-rights",
      "role": "accessibility-rights-review",
      "description": "Independently review one frozen EmbeddedKnowledge lesson candidate for accessibility, equivalent learning paths, semantic structure, reflow, interaction access, media provenance, licensing, privacy, security, attribution, and agent disclosure. Use when assigned the accessibility-and-rights quorum role and asked for a structured review against an exact candidate commit.",
      "invocation": "$review-embeddedknowledge-accessibility-rights",
      "source": ".agents/skills/review-embeddedknowledge-accessibility-rights/SKILL.md",
      "instructions": "/skills/review-embeddedknowledge-accessibility-rights/SKILL.md",
      "bundle": "/skills/review-embeddedknowledge-accessibility-rights/review-embeddedknowledge-accessibility-rights.zip",
      "instructionsSha256": "sha256:91b39569bf56bf8b4142230eab3a0b899f77d12aea02728eae13f53a16d03383",
      "bundleSha256": "sha256:0a2708a362cc8f8e03095a648f9398add4b673d779b30a5cecf0fb4ff51f4fd7"
    },
    {
      "name": "adjudicate-embeddedknowledge-lesson",
      "role": "adjudicator",
      "description": "Finalize and adjudicate one EmbeddedKnowledge lesson after its required frozen-candidate reviews are complete. For a standard lesson, use one fresh run to read the original candidate and both advisory reviews, dispose every finding, make the single final content revision, audit accessibility and rights, and produce the merge, revise, or reject artifact. For other risk tiers, apply their approval-quorum rule without changing content.",
      "invocation": "$adjudicate-embeddedknowledge-lesson",
      "source": ".agents/skills/adjudicate-embeddedknowledge-lesson/SKILL.md",
      "instructions": "/skills/adjudicate-embeddedknowledge-lesson/SKILL.md",
      "bundle": "/skills/adjudicate-embeddedknowledge-lesson/adjudicate-embeddedknowledge-lesson.zip",
      "instructionsSha256": "sha256:aa9d660385225c2ee2cdb7610b2f24ef8aebd04afbc41f77c8aa6f2cf93818c2",
      "bundleSha256": "sha256:519f15bfc66dfc42e018a7731fe47e3e10ab8d073d5d41b887ee4905eda4807d"
    }
  ]
}


---

## Repository file: site/agent/quorum-policy.json

{
  "schemaVersion": 3,
  "policyId": "lesson-quorum-v3",
  "status": "adopted-founding-stage",
  "effectiveDate": "2026-07-20",
  "countingUnit": "isolated-agent-review-run",
  "foundingStage": {
    "active": true,
    "purpose": "Permit one disclosed maintainer to operate an auditable cross-model review quorum while the contributor community is being established.",
    "accountablePrincipalMayOperateAuthorReviewAndAdjudicationAgents": true,
    "githubCommentReviewAcceptedForSelfOperatedAgents": true,
    "revisitAfterPublishedLessons": 25
  },
  "agentIndependence": {
    "distinctRunIdsRequired": true,
    "reviewRunsMustDifferFromAuthorRuns": true,
    "adjudicationRunMustDifferFromAuthorAndReviewRuns": true,
    "allCountedReviewsRequireAgentProvenance": true,
    "adjudicationRequiresAgentProvenance": true,
    "requiredDisclosure": [
      "system",
      "provider",
      "model",
      "version",
      "runId",
      "instructionsDigest"
    ]
  },
  "tiers": {
    "minor-correction": {
      "policyId": "minor-correction-v2",
      "reviewMode": "approval-quorum",
      "eligibleReviewVerdicts": [
        "approve"
      ],
      "minimumReviewInputs": 2,
      "minimumApprovingReviews": 2,
      "roleMinimums": {
        "academic": 1,
        "learning-design": 0,
        "accessibility-rights": 1
      },
      "minimumDistinctPrincipals": 1,
      "minimumDistinctAgentRuns": 2,
      "minimumDistinctAgentModelFamilies": 2,
      "finalAdjudicators": 1
    },
    "standard": {
      "policyId": "standard-lesson-v3",
      "reviewMode": "advisory-finalization",
      "eligibleReviewVerdicts": [
        "approve",
        "request-changes"
      ],
      "minimumReviewInputs": 2,
      "minimumApprovingReviews": 0,
      "roleMinimums": {
        "academic": 1,
        "learning-design": 1,
        "accessibility-rights": 0
      },
      "minimumDistinctPrincipals": 1,
      "minimumDistinctAgentRuns": 2,
      "minimumDistinctAgentModelFamilies": 2,
      "finalAdjudicators": 1,
      "finalCommitRequired": true,
      "adjudicatorMayFinalizeContentOnce": true,
      "adjudicatorAccessibilityRightsAuditRequired": true,
      "allReviewFindingsRequireDisposition": true
    },
    "high-impact": {
      "policyId": "high-impact-lesson-v2",
      "reviewMode": "approval-quorum",
      "eligibleReviewVerdicts": [
        "approve"
      ],
      "minimumReviewInputs": 5,
      "minimumApprovingReviews": 5,
      "roleMinimums": {
        "academic": 3,
        "learning-design": 1,
        "accessibility-rights": 1
      },
      "minimumDistinctPrincipals": 1,
      "minimumDistinctAgentRuns": 5,
      "minimumDistinctAgentModelFamilies": 3,
      "finalAdjudicators": 1
    },
    "rephrasing": {
      "policyId": "rephrasing-v1",
      "reviewMode": "maintainer-attested",
      "purpose": "Repair the readability of an already published lesson \u2014 typography, grammar, ambiguous phrasing \u2014 without convening the reviewer quorum, because the change cannot alter what the lesson teaches.",
      "eligibleReviewVerdicts": [],
      "minimumReviewInputs": 0,
      "minimumApprovingReviews": 0,
      "roleMinimums": {
        "academic": 0,
        "learning-design": 0,
        "accessibility-rights": 0
      },
      "minimumDistinctPrincipals": 1,
      "minimumDistinctAgentRuns": 0,
      "minimumDistinctAgentModelFamilies": 0,
      "finalAdjudicators": 1,
      "finalCommitRequired": false,
      "adjudicatorMayFinalizeContentOnce": false,
      "adjudicatorAccessibilityRightsAuditRequired": false,
      "allReviewFindingsRequireDisposition": false,
      "machineGate": "rephrasing-gate-v1",
      "gateEnforcedBy": "scripts/validate-rephrasing-pr.mjs",
      "requires": [
        "The lesson is already published, so the content being repaired has already cleared a full quorum.",
        "The version advances by exactly one patch step.",
        "Only scene prose changes: mathematics, every number, scene directives, claim and source mappings, section structure, assessment answers, objectives, glossary meanings, and references are identical.",
        "The accountable maintainer merges it and is answerable for the judgement that the wording is clearer."
      ],
      "forbids": "Any change to teaching content, claims, evidence, assessment, structure, or status. A change that cannot pass the machine gate uses the standard tier."
    }
  },
  "legacyAdjudications": [
    {
      "lessonId": "PREM-WEL-001",
      "lessonVersion": "0.4.2",
      "riskTier": "standard",
      "policyId": "standard-lesson-v2",
      "reviewMode": "approval-quorum",
      "eligibleReviewVerdicts": [
        "approve"
      ],
      "minimumReviewInputs": 3,
      "minimumApprovingReviews": 3,
      "roleMinimums": {
        "academic": 1,
        "learning-design": 1,
        "accessibility-rights": 1
      },
      "minimumDistinctPrincipals": 1,
      "minimumDistinctAgentRuns": 3,
      "minimumDistinctAgentModelFamilies": 3,
      "finalAdjudicators": 1,
      "finalCommitRequired": false,
      "adjudicatorMayFinalizeContentOnce": false,
      "adjudicatorAccessibilityRightsAuditRequired": false,
      "allReviewFindingsRequireDisposition": false
    }
  ],
  "mergeRequirements": {
    "allReviewsTargetSameCandidateCommit": true,
    "noUnresolvedBlockingFindings": true,
    "quorumSnapshotMustBeSatisfied": true,
    "finalAdjudicationFileMustBeCommittedWithLesson": true,
    "adjudicationRunMustBeFresh": true,
    "standardLessonFinalization": "Two role-isolated reviews advise one fresh adjudicator. The adjudicator may make one final content revision, must dispose every finding, and owns the final accessibility-and-rights audit and merge decision.",
    "postCandidateMutableFields": [
      "lesson.json.status",
      "lesson.json.sourceConfidence",
      "claims.json.claims[].reviewStatus"
    ],
    "staleReviewsAreDismissedAfterLessonChanges": true
  }
}