atomic outcomes currently have neither a merged lesson nor an active proposal.
- Total outcomes
- 388
- Claimed in open PRs
- —
- Published lessons
- —
REVIEWED LESSON CONTRIBUTIONS · OPEN
People and agents propose.
Declared review decides.
Pull requests are the write interface. During the founding stage, one disclosed maintainer may operate isolated cross-model agents for drafting, review, and adjudication. Every run remains attributable, inspectable, and committed beside the lesson.
Every contribution starts from one open outcome.
Choose one gap. Give an agent the contract.
COPY / PASTE INTO A CODING AGENT
Start a lesson contribution
Contribute without running an agent
Educators, subject experts, learning designers, accessibility reviewers, and librarians are first-class contributors. Take one of the review roles below yourself, file a correction against any published outcome or syllabus entry, or adopt the syllabus in a course and report what breaks in practice. Every path — human or agent — moves through the same public pull-request record.
Each contribution run carries exactly one responsibility.
One run. One responsibility.
Build the candidate
Select one open outcome, design the instructional argument, research its claims, construct the pack, and stop before governance.
Academic
Independently verify accuracy, evidence, calculations, scope, model boundaries, uncertainty, and assessment correctness.
Learning design
Trace learner fit, explanation, retrieval, practice, feedback, transfer, assessment, mastery, and recovery.
Access + rights review
For minor-correction and high-impact gates, independently audit equivalent learning paths, interaction access, reflow, provenance, licensing, privacy, and content security.
Finalization + adjudication
For a standard lesson, dispose every review finding, write the final version once, audit accessibility and rights, and issue the merge, revise, or reject decision.
Use after cloning the repository: invoke $author-embeddedknowledge-lesson in a compatible coding agent. If the client does not discover .agents/skills/, install the ZIP or direct the agent to the raw SKILL.md. The bundle supplies the procedure; the repository supplies the authoritative standards, schemas, data, and validators.
Role isolation is mandatory: the same accountable maintainer may operate the founding-stage agents, but each review and adjudication must be a fresh run. Skills guide the work; frozen commits, structured artifacts, provenance, GitHub submissions, validators, and quorum determine eligibility.
The corpus is readable by machines and writable only through pull requests.
Readable by machines. Writable through PRs.
Load the contract
Start with concise context and the content standard, then resolve only the graph, schema, policy, or full context needed for the task.
Public · read onlyBuild typed artifacts
Map a lesson to stable outcome IDs. Record authorship, evidence, assessment logic, rights, and material agent provenance.
- Lesson schema
- Review schema
- Adjudication schema
- Run deterministic validation
Open one focused PR
A pull request is the only operation that can enter the governance queue. WebMCP and website endpoints cannot write, approve, merge, or publish.
- Select one uncovered outcome
- Create one lesson pack
- Open a draft pull request
- Keep reviews on one candidate commit
Progressive enhancement: compatible browsers in a secure context receive seven read-only WebMCP tools, including lesson-state and open-PR discovery. Every agent can still use the stable llms.txt and JSON endpoints; the contribution protocol never depends on experimental browser support.
A contribution is a structured, portable lesson pack.
A lesson pack, not a loose document.
A contribution is portable when identity, outcomes, teaching, assessment, evidence, access, rights, and provenance remain reviewable outside any particular agent or interface.
- IdentityStable lesson ID, version, risk tier, title, language, authors, outcome IDs.
- Learning contractPrerequisites, observable objectives, duration, misconceptions, mastery.
- Teaching sequenceExplanation, worked examples, practice, retrieval, transfer, recovery.
- EvidenceClaim-linked sources, confidence, route relevance, uncertainty, verification, use type, rights basis, and agent-access state.
- AssessmentChecks, answer logic, rubric, feedback, mastery and remediation criteria.
- ProvenanceAccountable principals, agent runs, instruction digests, original synthesis, assets, licences, and modification records.
Three contracts: the learning-content standard defines how explanations, examples, practice, feedback, and transfer work; Lesson Format v1 defines how that teaching is stored and rendered; the source and reuse policy defines original synthesis, compatible reuse, agent access, and rights review. The Format specimen is generated from the repository example and never counts toward coverage.
Higher-risk lessons require more independent scrutiny.
More risk, more scrutiny.
1 academic and 1 accessibility/rights agent review, then 1 fresh adjudication run.
1 academic and 1 learning-design review, then 1 fresh finalizer writes, audits, and adjudicates the final version.
3 academic, 1 learning-design, and 1 accessibility/rights agent reviews, then 1 fresh adjudication run.
For a standard lesson, the number after “+” is the fresh finalizer: it receives the original candidate and both review records, writes the final version once, audits access and rights, and owns the merge, revise, or reject rationale.
Two reviews inform one final version.
One author. Two reviews. One finalizer. No loop.
- 01
Academic
Accuracy, scope, uncertainty, calculations, sources, and route relevance.
- 02
Learning design
Outcome, explanation, practice, assessment, mastery, and remediation align.
- +1
Finalization
A fresh run resolves both records, writes the final version, audits accessibility and rights, and records the decision.
Review independence comes from isolated runs under one accountable operator.
Independent runs. One accountable operator.
Isolate each role
The standard gate uses two review runs across two providers: academic and learning design. Review agents do not inherit the authoring conversation; the finalizer is a third fresh run.
Disclose provenance
Every run records its operator, system, provider, model, version, run ID, and a SHA-256 digest of its material instructions. Provenance is disclosed and attested by the accountable operator, not cryptographically verified.
Freeze, then finalize once
Both reviews target the original candidate. One fresh finalizer may make the recorded final edit; any later teaching-content change makes adjudication stale.
Course content is CC BY 4.0: reusable with attribution.
Reusable by default. Attributable always.
CC BY 4.0
Original educational content and curriculum data may be shared and adapted, including commercially, with attribution and an indication of changes.
MIT
Application code uses the MIT licence. The content licence does not silently license code, project marks, or personal data.
Declared, never assumed
A citation is not reuse permission. Expression, media, and datasets require an allowed basis and complete attribution; agents also honor source-specific access terms.
Make every contribution inspectable.
The repository and lesson queue are open. Every new lesson still starts from an uncovered outcome, records its sources and rights, passes the human-first gate, receives two isolated review inputs, and is finalized once before a maintainer can merge it.
- Choose one uncovered, unclaimed outcome.
- Open one focused draft lesson pull request.
- Keep both reviews pinned to the same original candidate.
- Let one fresh finalizer dispose every finding and audit access and rights.
- Merge only after deterministic publication checks pass.