v1.0.1 — committed outputs, rewritten README, metadata aligned to the archive

A packaging and documentation release. No result, number, figure, or gate changes from v1.0.0 — the curriculum, the claim ledger, and every audit are identical.

Archive: 10.5281/zenodo.21865644 still archives the v1.0.0 manuscript and remains the correct citation. This release revises none of the science, so it does not warrant a new deposit.

Read the artifacts without installing anything

Generated outputs now ship with the repository. You can open the manuscript, every figure, the SkillTree export, and each audit record straight from GitHub:

What	Where
Archived manuscript, exactly as deposited	Active_Skillference_v1.0.0_DOI-10.5281-zenodo.21865644.pdf
Current build of the manuscript	output/pdf/
Figures and their registry	output/figures/
SkillTree export, claim ledger, audits	output/data/
Web, DOCX, EPUB, slide builds	output/

Committing them does not soften the regenerability claim: the kernels rebuild every artifact from source, and validate_rendered_freshness.py fails if a committed artifact drifts from what the current source produces. Only renderer logs and LaTeX intermediates stay out, since they churn on every build and carry nothing to read.

The two PDFs are deliberately different

The root PDF is pinned byte-identical to the Zenodo deposit (MD5 7e43004d…), fetched back from the archive rather than re-derived, so the filename's claim about that DOI stays true no matter how the source moves on. output/pdf/ holds the current build, which differs whenever the manuscript has changed since the deposit. A file asserting a DOI whose archive holds something else is exactly the failure this project exists to prevent.

Also in this release

• README rewritten for a public reader: badges, a one-command start up front, a table linking every committed artifact, an explicit statement of what CI does and does not cover, and a contributing section naming the invariants and pre-PR commands.
• Metadata follows the archive. The deposit's title and abstract were updated to the paper's, so CITATION.cff, codemeta.json, and .zenodo.json now carry the record's wording. This also stops a future reservation from re-submitting the older text.
• Version 1.0.1 across the citation, CodeMeta, Zenodo, and Python metadata and on the manuscript cover.
• The untracked-surface contract was inverted. It previously asserted that deliverables were ignored, which became the opposite of the policy; it now guards only renderer detritus and credential-bearing local state, with the test flipped so a deliverable reappearing there fails.

Verification at this tag

954 tests · ≥90% src/ coverage gate · ruff and mypy clean · all 11 offline validators exit 0 · release-readiness inventory valid.

CI runs the offline gate set only. Docker, a seeded SkillTree, raw FORM login, and the browser console check need a daemon and a real browser, so they stay local and are never simulated: a green badge means the offline gates passed and nothing more. Full live dependency parity has been proven locally against skills-service 4.6.0.

Two boundaries unchanged: SkillTree's NSA-origin provenance is cited for platform context and is not an endorsement, and this release makes no claim of learner efficacy — the learner-pilot contract is protocol-only.

v1.0.0 — Active Skillference

A high-integrity micro-learning platform for Active Inference and the Free Energy Principle, built on the NSA-origin open-source SkillTree framework.

Archived at 10.5281/zenodo.21865644 · Read the manuscript

What this is

The Active Inference / FEP body of knowledge rendered as a validated prerequisite graph spanning eight strata, from the mathematics of surprise through to applications, where every quantitative claim a learner sees is produced by a tested computational kernel rather than hand-typed. The validated graph exports directly into SkillTree's Project → Subject → Skill model with prerequisite learning paths and quiz-gated completion, and seeds a live gamified instance over REST.

Start here

git clone https://github.com/ActiveInferenceInstitute/Active_Skillference.git
cd Active_Skillference
./run.sh              # dependencies, build, live SkillTree, learner dashboard
./run.sh --offline    # no Docker: build and serve generated artifacts only
./run.sh --check      # report readiness and exit

The live path needs a running Docker daemon because SkillTree owns quizzes, completion, points, and ranking. The offline path needs neither Docker nor a network and still serves the curriculum explorer, generated lesson records, dependency graph, and artifact browser.

In this release

• 630 skills across 111 subjects, 1199 prerequisite edges, 630 quizzes, 2520 assessment questions, maximum prerequisite depth 75.
• 79 registered claims from tested kernels. Recent additions are invariants rather than bare numbers: the cross-entropy decomposition, maximum entropy at the uniform distribution, mutual information vanishing on a factorized joint, Jensen's gap shrinking with concentration, free-energy slack closing at the posterior, the expected-free-energy decomposition residual, and both precision limits of the policy posterior. Tests check the ledger against the mathematics.
• 26 registered figures, 21 deterministic renders plus five captured browser recordings of the running learner dashboard.
• 188 cited sources resolved against both the bibliography and a source-role matrix.
• A local learner dashboard over the official SkillTree runtime, with a WAI-ARIA tab pattern, a skip link, status words rather than colors, and an accessible prerequisite-flow table beside the dependency graph.

How it is verified

Fail-closed gates, not aspirations: the provenance linter rejects hand-typed result numbers, the export refuses on any unresolved claim token, the curriculum DAG must stay acyclic and connected with roots only in the foundational strata, a semantic-prerequisite gate blocks assessing a concept a learner could not yet have met, every citation must resolve in two places, and the manuscript and dashboard artifact audits must agree.

The offline gate set runs in CI. The live proofs — Docker, a seeded SkillTree, raw FORM login, and a browser console check — stay local and are deliberately not simulated, so a green badge never implies a check that did not run. Full live dependency parity (1199/1199 edges) has been proven locally against skills-service 4.6.0.

Evidence and its limits

The five dashboard screenshots in the manuscript are captured recordings, marked as such in the figure registry, version-controlled rather than generated, excluded from the deterministic render parity check, and shipped with a provenance record naming the dashboard address, viewport, live counts, proof mode, timestamp, and a digest per image. They cannot be rebuilt from project data, and the repository says so rather than implying otherwise.

Two boundaries stated plainly: SkillTree's NSA-origin provenance is cited for platform context and is not an endorsement, and this release makes no claim of learner efficacy. The learner-pilot contract is protocol-only.

License and citation

MIT, copyright Daniel Ari Friedman. Vendored dependencies and their notices are in THIRD_PARTY_NOTICES.md. Cite with CITATION.cff or the DOI above; the same identifier appears in the citation file, CodeMeta, Zenodo metadata, the manuscript configuration, and on the cover page, cross-checked by a test against the recorded deposit.