tasks.yaml

# tasks.yaml - task + Gantt source of truth for this project.
# Managed by Operator (live read-write).
# status: todo | in_progress | blocked | done. dates: YYYY-MM-DD. progress: 0-100.
project:
  lifeArea: volunteering
  title: Active Skillference - Active Inference micro-learning platform
  status: active
  updated: "2026-07-31"
  live_metrics:
    skills: 630
    subjects: 111
    prerequisite_edges: 1199
    dependencies: 1199
    quizzes: 630
    assessments: 2520
    claims: 79
    figures: 26
    cited_sources: 188
    max_depth: 75
tasks:
  - id: asf-roadmap-oracle
    title: Roadmap and live-count drift validation
    status: done
    surface: roadmap
    owner: src.roadmap
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      claims: 79
      figures: 26
      cited_sources: 188
      max_depth: 75
    evidence:
      - uv run python scripts/validate_roadmap.py
      - uv run python scripts/validate_live_counts.py
    notes: tasks.yaml is now the machine-readable roadmap source, and TODO.md is validated against it so
      stale counts cannot remain green.
    start: ""
    end: ""
    progress: 0
    milestone: false
    deps: []
    tags: []
  - id: asf-distractor-quality-and-accessibility
    title: Off-domain distractor cleanup with regression gate, dashboard accessibility, and DAG-safety
      figure
    status: done
    surface: curriculum
    owner: src.curriculum
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
    evidence:
      - uv run python scripts/build_curriculum.py
      - uv run pytest tests/test_curriculum_quality.py tests/test_dashboard.py tests/test_figures.py
        tests/test_claim_support.py tests/test_manuscript_artifacts.py -q
    notes: Replaced off-domain placeholder quiz distractors (UI/plumbing junk) with on-domain
      alternatives and added quality._audit_implausible_distractors (verbatim + display-class guard,
      scoped to incorrect answers) so the class cannot regress; the gate caught a residual joint
      after a cross-vendor (Forge) audit flagged the first verbatim-only version as near-vacuous.
      Dashboard gained WCAG keyboard tab navigation, focus-visible styles, a skip link, and an
      accessible prerequisite-flow table. plot_dependency_strata_flow now marks the backward
      (lower-triangle) zone as must-stay-empty so the no-backward-edges invariant is self-proving.
      claim_support cross-checks declared claim tokens against the kernel registry and validator
      paths against the filesystem.
  - id: asf-curriculum-hardening
    title: Curriculum expansion and stricter quiz quality
    status: done
    surface: curriculum
    owner: src.curriculum
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
    evidence:
      - uv run python scripts/build_curriculum.py
      - uv run pytest tests/test_curriculum_catalog.py tests/test_curriculum_quality.py -q
    notes: The graph includes continuous dynamics, state-space probability, rate-distortion information,
      continuous variational inference, predictive-coding process theory, hierarchical active
      inference, reproducible simulation labs, applied evaluation dossiers, supplemental VFE/EFE
      scientific critique subjects, visual reasoning review subjects, and publication-synthesis
      handoff subjects.
  - id: asf-supplement-science-expansion
    title: Supplemental mathematical analysis and scientific curriculum expansion
    status: done
    surface: manuscript
    owner: manuscript/08_supplement_variational_free_energy.md
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      figures: 26
    evidence:
      - uv run python scripts/build_curriculum.py
      - uv run python scripts/run_kernels.py
      - uv run pytest tests/test_curriculum_catalog.py tests/test_manuscript_artifacts.py
        tests/test_figures.py -q
    notes: Detailed VFE/EFE walkthroughs moved into a numbered supplement, and scientific supplement
      subjects add falsifiable VFE/EFE interpretation practice across all strata.
  - id: asf-visual-reasoning-expansion
    title: Visual reasoning curriculum and stratum-balance figure
    status: done
    surface: figures
    owner: src.curriculum._ext_visual_reasoning
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      figures: 26
    evidence:
      - uv run python scripts/build_curriculum.py
      - uv run python scripts/run_kernels.py
      - uv run pytest tests/test_curriculum_catalog.py tests/test_figures.py
        tests/test_manuscript_artifacts.py -q
    notes: Eight visual-reasoning subjects add figure-reading, captioning, dashboard, and
      manuscript-review skills across all strata, while the stratum-balance figure exposes expansion
      concentration risk.
  - id: asf-visual-evidence-contracts
    title: Visual-evidence contract curriculum and ladder figure
    status: done
    surface: figures
    owner: src.curriculum._ext_visual_evidence_contracts
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      figures: 26
    evidence:
      - uv run python scripts/build_curriculum.py
      - uv run python scripts/run_kernels.py
      - uv run python scripts/validate_manuscript_artifacts.py
      - uv run pytest tests/test_curriculum_catalog.py tests/test_figures.py tests/test_dashboard.py
        tests/test_manuscript_artifacts.py -q
    notes: Two focused six-skill subjects teach visual evidence contracts across computation and
      active-inference explanation. The visual_evidence_contract_ladder figure, registry
      chart-contract fields, manuscript prose, dashboard detail view, and artifact validator now
      share the same question, source, encoding, static/mobile, and failure-mode contract.
  - id: asf-scholarship-health
    title: Scholarship coverage and source-health validation
    status: done
    surface: scholarship
    owner: src.scholarship
    metrics:
      cited_sources: 188
    evidence:
      - uv run python scripts/analyze_scholarship.py
      - uv run python scripts/validate_scholarship_sources.py
      - uv run python scripts/validate_scholarship_sources.py --online --strict-online
      - uv run pytest tests/test_scholarship.py -q
    notes: Scholarship rows must be cited, BibTeX-backed, lane-assigned, role-specific, and
      locator-specific. The default check is offline; --online --strict-online hard-fails on a dead
      DOI/URL (404, DNS) while tolerating publisher 403 bot-blocks, and is run as a networked
      release gate (not in offline CI).
  - id: asf-dashboard-graph-diagnostics
    title: Dashboard graph diagnostics and accessible table fallback
    status: done
    surface: dashboard
    owner: src.dashboard
    evidence:
      - uv run pytest tests/test_dashboard.py tests/test_dashboard_console.py -q
      - chrome-devtools-axi eval "document.title"
      - uv run python scripts/validate_dashboard_console.py
    notes: The dashboard exposes graph bottlenecks, prerequisite neighborhoods, subject flows, max-depth
      filtering, and read-only Learn/Quiz SkillTree handoff routes. validate_dashboard_console.py is
      an opt-in browser gate that fails on any console error except the known, cosmetic upstream
      skills-client probe of SkillTree's built-in Inception project (401, non-blocking, not fixable
      from our config); it reports skipped (not clean) without a browser or running dashboard.
  - id: asf-visualization-artifacts
    title: Diagnostic figures and artifact registry validation
    status: done
    surface: figures
    owner: src.figures
    metrics:
      figures: 26
    evidence:
      - uv run pytest tests/test_figures.py tests/test_manuscript_artifacts.py -q
      - uv run python scripts/validate_manuscript_artifacts.py
    notes: Graph bottlenecks, skill-neighborhood paths, scholarship source health, roadmap risk,
      validation-surface, and validation-intent mismatch figures are generated, captioned,
      alt-texted, and exposed to the dashboard.
  - id: asf-technical-docs-visualization-contracts
    title: Technical documentation and visualization contract hardening
    status: done
    surface: figures
    owner: tests/test_docs_inventory.py
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      figures: 26
    evidence:
      - uv run pytest tests/test_docs_inventory.py tests/test_figures.py
        tests/test_manuscript_artifacts.py tests/test_dashboard.py -q
      - uv run python scripts/run_kernels.py
      - uv run python scripts/validate_manuscript_artifacts.py
      - uv run python scripts/validate_rendered_freshness.py
      - uv run python scripts/validate_roadmap.py
      - uv run pytest --cov=src --cov-fail-under=90
    notes: Technical documentation inventories now compare docs, scripts, and figure rows against the
      live repository and planned figure registry. The figure contract adds data grain, palette
      policy, palette roots, and non-color encoding fields; the manuscript artifact audit, dashboard
      artifact cards, negative controls, and targeted figures use the same fields. Graph
      bottlenecks, validation-surface, and roadmap-risk visuals now carry more of their reading in
      direct labels, totals, hatching, and status text rather than hue alone.
  - id: asf-visual-contract-specificity-and-doc-accuracy
    title: Figure contract specificity and technical documentation accuracy
    status: done
    surface: figures
    owner: src.figures
    metrics:
      figures: 26
    evidence:
      - uv run pytest tests/test_figures.py tests/test_manuscript_artifacts.py
        tests/test_docs_inventory.py -q
      - uv run python scripts/run_kernels.py
      - uv run python scripts/validate_manuscript_artifacts.py
      - uv run python scripts/validate_rendered_freshness.py
      - uv run python scripts/validate_roadmap.py
    notes: Figure registry validation now rejects generic fallback question, data-grain, static-read,
      mobile-read, and failure-mode text. All registered figures carry figure-specific chart
      contracts; sparse source views were redesigned as ledgers/status cards, concept coverage as a
      lollipop audit, graph bottlenecks as a top-N direct-label ranking, and the long-path
      diagnostic as a checkpointed route. Technical docs and manuscript prose now describe the full
      visual QA field set, learn-route versus quiz-route fragments, and the bounded 90% src coverage
      gate without claiming blanket 100% kernel coverage.
  - id: asf-cover-publication-synthesis
    title: Generated cover asset and publication-synthesis curriculum
    status: done
    surface: manuscript
    owner: src.curriculum._ext_publication_synthesis
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      figures: 26
    evidence:
      - uv run python scripts/build_curriculum.py
      - uv run python scripts/run_kernels.py
      - uv run python scripts/validate_manuscript_artifacts.py
      - uv run pytest tests/test_curriculum_catalog.py tests/test_figures.py
        tests/test_manuscript_artifacts.py -q
    notes: A deterministic first-page cover and page-two named-skill atlas are generated from live
      graph, claim, roadmap, and skill-name data as non-numbered front-matter artifacts, while eight
      publication-synthesis subjects train reader handoff across all strata.
  - id: asf-scientific-integrity-refactor
    title: Claim-support evidence taxonomy and scientific-audit curriculum
    status: done
    surface: manuscript
    owner: src.manuscript.claim_support
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      figures: 26
      cited_sources: 188
    evidence:
      - uv run python scripts/validate_claim_support.py
      - uv run python scripts/validate_scholarship_sources.py
      - uv run pytest tests/test_claim_support.py tests/test_scholarship.py tests/test_figures.py -q
    notes: Manuscript claim-support anchors now bind strong scientific wording to computed, count,
      simulation, literature, scope, delivery, or visual evidence classes; eight scientific-audit
      subjects teach these proof obligations across all strata.
  - id: asf-skilltree-nsa-security-context
    title: SkillTree NSA provenance and national-security posture documentation
    status: done
    surface: scholarship
    owner: manuscript/07_scope_and_related_work.md
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      cited_sources: 188
    evidence:
      - uv run python scripts/analyze_scholarship.py
      - uv run python scripts/validate_claim_support.py
      - uv run pytest tests/test_scholarship.py tests/test_claim_support.py
        tests/test_curriculum_catalog.py -q
    notes: SkillTree is now treated as an NSA-origin open-source substrate with explicit
      non-endorsement, deployment-posture, trusted-client, supply-chain, and APT-aware misuse
      boundaries.
  - id: asf-american-english-readiness-redteam
    title: American English manuscript gate and initial-publication RedTeam review
    status: done
    surface: manuscript
    owner: src.manuscript.american_english
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      claims: 79
      figures: 26
      cited_sources: 188
      max_depth: 75
    evidence:
      - uv run python scripts/validate_american_english.py
      - uv run python scripts/validate_claim_support.py
      - uv run pytest tests/test_manuscript_language.py tests/test_claim_support.py
        tests/test_manuscript_artifacts.py -q
    notes: Authored manuscript prose now uses American English with a fail-closed validator. The project
      is suitable for initial static public-review publication, while live SkillTree proof and
      DOI/repository publication metadata remain tracked as separate open tasks.
  - id: asf-skilltree-live-recheck
    title: Re-verify live SkillTree service and official quiz render
    status: done
    surface: skilltree
    owner: scripts/run_learning_dashboard.sh
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      dependencies: 1199
      quizzes: 630
      assessments: 2520
    evidence:
      - SKILLTREE_URL=http://localhost:8080 ACTIVE_SKILLFERENCE_LOCAL_AUTH=1
        ./scripts/run_learning_dashboard.sh
      - uv run python scripts/bootstrap_local_skilltree.py --base-url http://localhost:8080 --no-seed
      - uv run python scripts/validate_live_skilltree_dashboard.py --dashboard-url
        http://127.0.0.1:8765 --skilltree-url http://localhost:8080
      - "chrome-devtools-axi: verified dashboard title, SkillTree online status, official display
        mount, mastery_computation_07 search, Learn/Quiz links, and real four-question SkillTree
        quiz render"
    notes: Docker SkillTree is reachable at localhost:8080, the dashboard is relaunched with local auth
      at 127.0.0.1:8765, the official SkillTree display mounts, and mastery_computation_07 renders a
      real four-question quiz in Chrome. Skills and quizzes seeded through local SkillTree
      workflows; remaining large-graph dependency rows were synchronized idempotently into the local
      Docker database because per-edge REST dependency validation was too slow at the 600-skill
      scale.
  - id: asf-skilltree-452-live-reverify
    title: Live re-verify local SkillTree stack against skills-service 4.5.2
    status: done
    scale: minor
    surface: skilltree
    owner: docker-compose.yml
    deps:
      - asf-skilltree-live-recheck
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      quizzes: 630
    evidence:
      - uv run python scripts/reset_local_skilltree.py --confirm-local-reset --timeout 1200
      - "proof.ok=true; proof.mode=app-ready; liveAdminProject.skills=630;
        artifacts present=42/42; figureContracts registered=21/dashboard=21;
        browserConsole 'dashboard console gate OK'; rawSkillTreeLogin
        adminSurfaceVisible=true, title='SkillTree Dashboard - Project Administrator'"
      - uv run python scripts/validate_docker_digests.py
      - uv run pytest tests/ --cov=src --cov-report=term-missing -q (814 passed, 90.17%)
    notes: >-
      Closed 2026-07-04. Two real bugs surfaced and fixed en route (both pre-existing, not
      introduced by the 4.5.2 bump, but blocking this task): (1)
      scripts/validate_full_app_stack.py called `docker compose up -d` without first setting
      ACTIVE_SKILLFERENCE_POSTGRES_PASSWORD the way scripts/run_skilltree.sh does, so the
      documented vendor/README.md one-liner failed on a fresh checkout; fixed by adding
      src/integration/full_app_stack.py:ensure_postgres_password() (same state-dir-scoped file
      convention as the shell script) and calling it before the compose command. (2)
      scripts/reset_local_skilltree.py's `docker compose down -v` has the same interpolation
      requirement and had the same gap; fixed by calling the same ensure_postgres_password()
      helper there too. A stale skilltree_auth.json from the 2026-06-22 4.4.2 run also caused a
      transient 401 on the first clean-volume attempt (stored admin credentials no longer matched
      the freshly-initialized Postgres DB) — resolved by using the documented
      reset_local_skilltree.py path (which backs up + removes local auth state before resetting
      the volume) instead of a bare `docker compose down -v`. Final run: `--app-ready` proof is
      green end-to-end against 4.5.2 (root bootstrap, project seed, dashboard health, figure
      contracts, browser console gate, raw FORM login all passed). Full live-dependency parity
      (1199/1199 edges via the admin REST API) was NOT attempted here — `--app-ready` explicitly
      skips it, consistent with this project's own prior finding that per-edge REST dependency
      validation is too slow at the 600-skill scale (see asf-skilltree-live-recheck's notes); that
      remains a separately-scoped, still-open concern independent of this version bump. Docker
      containers and the dashboard process were stopped and the compose volume removed after
      recording evidence, per the "disposable outputs" convention.
  - id: asf-skilltree-460-live-reverify
    title: Live re-verify local SkillTree stack against skills-service 4.6.0
    status: todo
    scale: minor
    surface: skilltree
    owner: docker-compose.yml
    deps:
      - asf-skilltree-452-live-reverify
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      quizzes: 630
    evidence:
      - uv run python scripts/reset_local_skilltree.py --confirm-local-reset --timeout 1200
      - uv run python scripts/validate_docker_digests.py
    notes: >-
      Opened 2026-08-01 after bumping the pinned runtime from skills-service 4.5.2 to 4.6.0
      (upstream release 2026-07-28). docker-compose.yml postgres:15 digest and the
      skills-service tag+digest were refreshed via the documented Docker Hub procedure
      (skills-service 4.6.0 = sha256:071220ce..., single-arch; postgres:15 = sha256:74e110c4...,
      multi-arch incl. arm64), and AuthMode was re-confirmed as FORM/PKI/SAML2 at the 4.6.0 tag
      (no PKI_OR_FORM). The digest pin and docs were updated, but no live full-app boot against
      4.6.0 was recorded because Docker was not running on the reconciliation host. Per the
      provenance-over-assertion principle, the 4.5.2 live records (asf-skilltree-452-live-reverify)
      do NOT prove anything about 4.6.0; run the full-app proof (reset_local_skilltree.py
      --confirm-local-reset --timeout 1200) against 4.6.0 and record proof.ok here before reading
      this pin as live-verified.
      2026-08-09 partial progress on a Docker-capable host, recorded without claiming the phase
      exit: validate_docker_digests.py passes against the running stack, and the strict
      full-parity proof (validate_full_app_stack.py --skip-regenerate, see
      asf-live-stack-ci-rehearsal) is green against skills-service 4.6.0 with 1199/1199 live
      dependencies, 630 live admin skills, raw FORM login, and the browser console gate. What is
      still NOT proven is the fresh-volume path this task actually asks for: whether the
      documented default root credentials bootstrap correctly on a newly created 4.6.0 volume.
      That requires the destructive reset (docker compose down -v), which would also discard the
      1199/1199 parity that took roughly an hour of per-edge writes to reach on this Apple/ARM
      host, so it is deliberately left for an explicit maintainer run rather than folded in here.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - scope_limit
      boundaries:
        - "Live full-app proof against skills-service 4.6.0 is operational verification, not a scientific or learner-outcome claim."
      phase_exit:
        - "`uv run python scripts/reset_local_skilltree.py --confirm-local-reset --timeout 1200` reports proof.ok=true against 4.6.0 and the evidence is recorded in this task."
  - id: asf-captured-delivery-evidence
    title: Captured browser evidence for the delivery surface
    status: done
    scale: medium
    surface: manuscript
    owner: src.screenshots
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      figures: 26
    evidence:
      - uv run python scripts/capture_dashboard_screenshots.py
      - uv run python scripts/validate_manuscript_artifacts.py
      - uv run python scripts/validate_claim_support.py
      - uv run pytest tests/test_screenshots.py -q
    notes: >-
      Closed 2026-08-09. The delivery result was the only Results subsection carrying no
      visual evidence, which left the strongest delivery claim unfalsifiable from the paper:
      that the official SkillTree learner client actually mounts and renders the seeded
      project rather than a local copy of course text. Five browser recordings of the running
      dashboard now carry that claim. They are registered in the same figure registry as the
      deterministic plots, so they inherit the caption, alt-text, and chart-contract floors
      and the manuscript and dashboard parity checks, but they declare evidence_kind captured
      and a tracked asset root because a recording is not a function of project data and
      cannot be rebuilt from a checkout. They are therefore version-controlled under
      manuscript/assets/screenshots, excluded from the deterministic render/registry parity
      check, staged into the generated figure directory at build time so the renderer resolves
      them, and stamped with a provenance record naming the dashboard address, viewport, live
      SkillTree counts, and full-app proof mode of the capture run. Known limitation recorded
      rather than papered over: no gate detects a screenshot that has gone stale against a
      changed interface, so a capture must be rerun whenever the dashboard display changes.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - visual
        - delivery
      boundaries:
        - Captured browser evidence records interface delivery, not learner outcomes or efficacy.
        - A recording is not reproducible byte-for-byte and is not presented as a deterministic render.
      phase_exit:
        - Every captured view is declared, cross-referenced, registered, and dashboard-visible.
        - The capture provenance record names the system state the images evidence.
  - id: asf-public-release-metadata
    title: Publication DOI and repository target decision
    status: blocked
    scale: external
    surface: metadata
    owner: manuscript/config.yaml
    evidence:
      - uv run python scripts/validate_live_counts.py
      - uv run python scripts/validate_manuscript_variables.py
      - uv run python scripts/z_generate_manuscript_variables.py
      - uv run python scripts/validate_rendered_freshness.py
      - git ls-remote --heads https://github.com/ActiveInferenceInstitute/Active_Skillference.git
        HEAD
    notes: The public repository target is owner-confirmed as ActiveInferenceInstitute/Active_Skillference
      and is set in manuscript/config.yaml. The owner also confirmed MIT for authored code/content
      and generated artifacts, Daniel Ari Friedman as the copyright holder, and the need for a
      counsel-review checklist before formal release. The private docxology/Active_Skillference
      mirror is current and local offline validators are green. Probed on 2026-08-09, the public
      target now exists and is public, but git ls-remote returns zero refs and the GitHub API
      reports size 0, so the target is reachable and still unpublished. Existence is not
      publication, and no commit, no tag, no archive deposit, and no DOI exist for it yet. This task
      stays blocked rather than todo because the remaining step is an owner-executed push, release,
      and Zenodo deposit performed outside this repository, not repository work that can be
      implemented here. DOI, release-date, and repository-URL fields stay blank rather than
      carrying fabricated metadata.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys:
        - smith2016softwarecitation
        - citationfileformat2026
        - codemeta2026schema
        - github2026citingcontent
        - zenodo2026github
        - wilkinson2016fair
        - nosek2015top
      planned_source_targets: []
      claim_support_types:
        - delivery
        - scope_limit
        - literature
      boundaries:
        - The empty public repository remains an external blocker, not a failed scholarly claim.
        - DOI and repository metadata remain blank until real GitHub and Zenodo events are observed.
      phase_exit:
        - The public target carries published content and a real release, and the publication environment can complete the actual release path.
        - Any release metadata is copied from observed repository and archive records.
  - id: asf-planning-surface-reconciliation
    title: Reconcile TODO, tasks, and ISA planning surfaces
    status: done
    surface: roadmap
    owner: ISA.md
    evidence:
      - uv run python scripts/validate_roadmap.py
      - uv run python scripts/validate_live_counts.py
      - rg -n "^- \\[ \\]|DEFERRED-VERIFY|progress:|updated:" ISA.md TODO.md tasks.yaml
    notes: ISA.md now declares tasks.yaml as the canonical forward roadmap and TODO.md as the generated
      reader view. The stale 143/145 progress field was replaced with a tasks.yaml-canonical marker,
      the update date is 2026-06-18, and the duplicate Forge/GPT-5.4 ISC entries are superseded by
      asf-cross-vendor-security-audit rather than remaining as a second open-work queue.
  - id: asf-cross-vendor-security-audit
    title: Cross-vendor security audit closure for dashboard and local auth
    status: done
    surface: validation
    owner: src/dashboard/server.py
    evidence:
      - uv run pytest tests/test_dashboard_server.py tests/test_dashboard_console.py
        tests/test_skilltree_local_auth.py tests/test_dashboard.py -q
      - uv run python scripts/validate_dashboard_console.py --require-browser --require-dashboard
      - Forge/GPT-5.4 read-only audit over src/dashboard/server.py, src/skilltree/local_auth.py,
        scripts/validate_full_app_stack.py, and related tests
    notes: "Closed 2026-06-20 — scope: an independent Forge/GPT-5.4 cross-vendor, read-only (static, no
      dynamic/runtime testing) audit at reasoning_effort=high over the
      dashboard/local-auth/full-stack source + tests. This closes the specific deferred task (run
      the independent cross-vendor pass and fix findings); it is a point-in-time static scan, not a
      guarantee of completeness — dependency/supply-chain, secrets-in-git-history, and
      dynamic-runtime classes are out of its scope. Within that scope it returned NO unresolved
      critical/high issues: both prior fixes are present, correct, and non-vacuously tested — the
      DNS-rebind / cross-site token-theft fix (Host-header hostname pinning + Sec-Fetch-Site/Origin
      gate, server.py:142-190; the credible cross-site GET vector sends Sec-Fetch-Site:cross-site
      and is rejected and tested at test_dashboard_server.py:329-344) and the fail-open
      token-exchange fix (empty/typeless/token-less bodies raise; structured-but-secretless JSON is
      not laundered, local_auth.py:1009-1050). Path traversal is closed (exact-id allowlist;
      '/'+leading-'.' rejection) and CORS is allowlist-gated. The one actionable defense-in-depth
      finding was implemented this session: DashboardConfig now fails closed on a non-loopback bind
      when local_auth_enabled (src/dashboard/config.py _is_loopback_host) with a firing
      negative-control test (tests/test_dashboard.py::test_local_auth_requires_loopback_bind).
      Remaining audit notes were low-severity and non-exploitable under the localhost-browser threat
      model."
  - id: asf-kernel-bound-aif-expansion
    title: Kernel-bound numbers for belief propagation, parameter learning, model reduction, and
      multi-step planning
    status: done
    surface: curriculum
    owner: src/kernels
    evidence:
      - uv run pytest tests/test_kernels_examples.py tests/test_kernels_inference.py
        tests/test_kernels_active_inference.py -q
      - uv run python scripts/run_kernels.py
      - uv run python scripts/z_generate_manuscript_variables.py
      - Forge/GPT-5.4 read-only math + crippled-comparator audit of the new kernels
    notes: "Closed 2026-06-20. Six Active Inference / FEP concepts that the curriculum taught with zero
      kernel-bound numbers are now backed by tested kernels emitting provenance claims: exact
      sum-product belief propagation on a chain (belief_propagation_chain, cross-checked against
      brute-force joint enumeration); Dirichlet-conjugate parameter learning (digamma identity);
      Bayesian model reduction (log Bayes factor full vs reduced); the expected-free-energy
      ambiguity term; multi-step planning (evaluate_multistep_policies) with an honest measured
      look-ahead gain — zero in a decoupled model, positive in a coupled investment model, compared
      against the STRONGEST greedy baseline (not a crippled comparator, independently re-verified by
      Forge); and continuous predictive coding (linear-Gaussian prediction-error descent converging
      to the analytic posterior with balanced precision-weighted errors). Bound into the manuscript
      supplement via {{claim:...}} tokens and new claim-bound figures (planning_lookahead,
      predictive_coding_descent). The kernel input parameters are also registered as claims so the
      model definition cannot silently drift from the manuscript."
  - id: asf-predictive-coding-and-learner-facing-binding
    title: Continuous predictive-coding kernel and learner-facing binding of kernel claims
    status: done
    surface: curriculum
    owner: src/kernels/predictive_coding.py
    evidence:
      - uv run pytest tests/test_kernels_predictive_coding.py tests/test_kernels_examples.py
        tests/test_figures.py tests/test_claim_support.py -q
      - uv run python scripts/run_kernels.py
      - uv run python scripts/z_generate_manuscript_variables.py
      - Forge/GPT-5.4 read-only math + binding-honesty audit of the predictive-coding kernel and the
        new claim bindings
    notes: "Closed 2026-06-20. Added the continuous-state predictive-coding kernel (linear-Gaussian
      prediction-error descent) the curriculum taught with zero numbers: the latent estimate settles
      at the exact analytic precision-weighted posterior (1.333333) where the sensory and prior
      prediction errors balance, with the free energy descending monotonically — cross-checked
      against the closed-form posterior, not self-consistency. Bound into the manuscript supplement
      + a new claim-bound figure (predictive_coding_descent), raising the live claim and figure
      counts. Crucially, the prior round's new kernel numbers were manuscript-only; this round binds
      them into the LEARNER-FACING curriculum: the belief-propagation marginals into
      mp_belief_propagation and the predictive-coding result into ai_generalized_coordinates, so the
      SkillTree export now shows the computed numbers, not just descriptive prose. The hand-typed
      kernel input parameters in 05_experimental_setup are now bound via {{claim:param_*}} tokens so
      the experimental-setup prose cannot drift from the kernel constants."
  - id: asf-integrity-gate-and-floor-hardening
    title: Integrity-gate hardening, assessment-floor binding, and figure-layer drift fixes
    status: done
    surface: validation
    owner: scripts/validate_manuscript_variables.py
    evidence:
      - uv run pytest tests/test_manuscript_variables.py tests/test_claim_support.py
        tests/test_figures.py tests/test_release_readiness_inventory.py -q
      - uv run python scripts/validate_manuscript_variables.py
      - uv run python scripts/run_kernels.py
      - Forge/GPT-5.4 read-only vacuity audit of the new and modified gates
    notes: "Closed 2026-06-20. Fixed real defects and vacuous gates surfaced by a cross-vendor review:
      the hard-coded-count linter glob now covers the lettered subsection files (NNx_*.md) where
      result/methodology prose lives, with a firing negative-control test (previously a hand-typed
      count in 03a bypassed the gate); the manuscript-wide endorsement-phrase blocklist is now wired
      into run_kernels as a fail-closed gate (find_banned_phrases had zero production callers) and
      resists whitespace/NBSP/zero-width evasion; the enforced assessment floor
      (MIN_ASSESSMENTS_PER_SKILL=4) is now a bound manuscript variable and a dynamically-rendered
      figure value, replacing the factually-wrong hand-typed 'three-question floor' across the
      manuscript AND five figure strings (Forge caught the figure layer the manuscript fix had
      missed); strata_count is exported and bound, removing the 'eight strata' hand-typed
      word-numerals; the dashboard fails closed on a non-loopback bind when local auth is enabled;
      the typography font-floor gate also scans labelsize=; and the release-readiness blocker-probe
      subsystem gained real subprocess coverage (76.8% -> 89%)."
  - id: asf-live-full-stack-release-rehearsal
    title: Live SkillTree and dashboard full-stack release rehearsal
    status: done
    surface: skilltree
    owner: scripts/validate_full_app_stack.py
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      dependencies: 1199
      quizzes: 630
      assessments: 2520
    evidence:
      - docker info
      - colima start
      - uv run python scripts/validate_full_app_stack.py --app-ready --skip-regenerate
      - uv run python scripts/validate_full_app_stack.py --skip-regenerate
    notes: The 2026-06-19 closeout started Colima, confirmed Docker was reachable, and ran both
      full-stack release proofs. App-ready and full-parity modes passed against live SkillTree at
      localhost:8080 and the local dashboard at 127.0.0.1:8765 with 630 live skills, 111 subjects,
      1199 expected/live/matched dependencies, 630 quizzes, 37 dashboard artifacts present, 19
      registered/dashboard figure contracts, 630 course-material records, 2023 external pointers,
      clean live-admin checks, and a browser-console gate with only the known upstream Inception
      probe allow-listed.
  - id: asf-raw-skilltree-form-login-proof
    title: Raw SkillTree FORM login proof in the full-stack validator
    status: done
    surface: skilltree
    owner: scripts/validate_skilltree_form_login.py
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      dependencies: 1199
      quizzes: 630
      assessments: 2520
    evidence:
      - uv run pytest tests/test_skilltree_form_login.py tests/test_full_app_stack.py
      - uvx ruff check scripts/validate_skilltree_form_login.py src/integration/full_app_stack.py
        tests/test_skilltree_form_login.py tests/test_full_app_stack.py
      - uv run python scripts/validate_full_app_stack.py --keep-running
    notes: The full-stack proof now distinguishes local trusted-client state readiness
      from live browser-login proof. validate_skilltree_form_login.py drives
      chrome-devtools-axi through the raw SkillTree FORM login, submits the ignored
      local admin credentials server-side, fails unless the authenticated administrator
      surface loads, and redacts admin passwords, trusted-client secrets, and bearer
      tokens from diagnostics. validate_full_app_stack.py includes this rawSkillTreeLogin
      proof whenever browser checks are enabled, before treating the local app as
      fully proven.
  - id: asf-publication-bundle-render-rehearsal
    title: Re-render full publication bundle after roadmap expansion
    status: done
    surface: manuscript
    owner: output
    evidence:
      - uv run python scripts/z_generate_manuscript_variables.py
      - uv run python scripts/validate_rendered_freshness.py
      - uv run python /Users/4d/Documents/GitHub/template/scripts/03_render_pdf.py --project
        Active_Skillference
      - rg -n "open items|roadmap_open_count" output/pdf output/web output/docx output/epub
        output/manuscript
    notes: Publication freshness now has a fail-closed gate for stale roadmap-open-count prose across
      rendered bundle text artifacts. The full bundle must be regenerated from the final
      tasks.yaml/TODO state before publication; the validator rejects old metadata-only open-count
      text.
  - id: asf-release-readiness-inventory-contract
    title: Release-readiness inventory contract
    status: done
    surface: validation
    owner: src.release_readiness
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      figures: 26
    evidence:
      - uv run pytest tests/test_release_readiness_inventory.py tests/test_docs_inventory.py
        tests/test_figures.py tests/test_manuscript_artifacts.py tests/test_roadmap.py -q
      - uv run python scripts/validate_release_readiness_inventory.py
      - uv run python scripts/validate_roadmap.py
      - uv run python scripts/validate_live_counts.py
    notes: Release readiness is now a machine-checked inventory surface. src.release_readiness and
      scripts/validate_release_readiness_inventory.py validate the dirty tracked-file summary,
      untracked-file list, ignored generated-output examples, and required evidence commands, while
      --write refreshes docs/release_readiness_inventory.md with current Docker and
      public-repository blocker probes.
  - id: asf-pasted-review-notation-release-hardening
    title: Deep-review notation, route, and release-evidence hardening
    status: done
    surface: manuscript
    owner: manuscript
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      figures: 26
    evidence:
      - uv run pytest tests/test_curriculum_catalog.py tests/test_claim_support.py
        tests/test_manuscript_artifacts.py tests/test_figures.py
        tests/test_release_readiness_inventory.py -q
      - uv run python scripts/run_kernels.py
      - uv run python scripts/validate_manuscript_artifacts.py
      - uv run python scripts/validate_release_readiness_inventory.py
      - uv run python scripts/validate_roadmap.py
    notes: "The pasted technical review is encoded in source-controlled gates and manuscript prose: VFE
      now states the joint-density/posterior identity at first use; EFE states the lower-is-better
      softmax convention and policy-precision sign; hardware determinism is narrowed to supported
      software-environment reproducibility; the supplemental bridge separates learn routes from quiz
      routes; and the reproducibility section records local release evidence through the generated
      release-readiness inventory instead of implying a clean public release."
  - id: asf-reproducibility-citation-hardening
    title: Paper-embedded reproducibility and primary-citation hardening
    status: done
    surface: manuscript
    owner: manuscript
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      figures: 26
      cited_sources: 188
    evidence:
      - uv run python scripts/validate_claim_support.py
      - uv run python scripts/analyze_scholarship.py
      - uv run python scripts/validate_scholarship_sources.py
      - uv run pytest tests/test_release_readiness_inventory.py tests/test_claim_support.py
        tests/test_scholarship.py
    notes: The June 2026 rigorous manuscript review is encoded in source-owned prose,
      scholarship rows, and validation code. The paper now frames the work as a
      systems/curriculum-infrastructure artifact, removes modular scaffold blocks, adds primary
      citations for sum-product, Bayesian model reduction, linear filtering, mastery learning, and
      PROV constraints, gives supplement parameter tables, separates structural quiz checks from
      psychometric validity, and records SHA-256 hashes for generated key artifacts and registered
      figures in the release-readiness inventory.
  - id: asf-release-branch-hygiene
    title: Isolate and commit release-ready project changes
    status: done
    surface: validation
    owner: git worktree
    evidence:
      - git status --short
      - git ls-files -o --exclude-standard
      - git diff --check
      - git diff --stat
      - uv run python scripts/validate_release_readiness_inventory.py
      - git rev-parse HEAD
      - git ls-remote --heads origin main
      - /opt/homebrew/bin/gh-axi repo view docxology/Active_Skillference
      - docs/release_readiness_inventory.md
    notes: Release-ready project changes were staged, committed, and pushed to the private
      docxology/Active_Skillference main branch on 2026-06-19. Remote main was verified at
      d03f56c516ab1d813e8bde7185751b760bdbeb9f after the follow-up TODO closeout; the
      release-readiness inventory renderer also distinguishes dirty pre-commit evidence from a clean
      post-commit checkout so it cannot keep certifying stale modified/untracked file lists after
      branch hygiene is complete.
  - id: asf-coverage-frontier-hardening
    title: Raise coverage and negative controls on integration frontiers
    status: done
    surface: validation
    owner: tests
    evidence:
      - uv run pytest --cov=src --cov-fail-under=90
      - uv run pytest tests/test_course_materials.py tests/test_dependency_reconciler.py
        tests/test_full_app_stack.py -q
      - uvx ruff check src scripts tests
      - uvx ruff format --check src scripts tests
      - uv run mypy src
    notes: Added targeted negative controls for malformed material records/topic mappings, stale or
      invalid dependency progress, extra live dependency parity, unreachable Docker daemon, early
      dashboard process exit, and stale rendered publication roadmap prose. The full 90% coverage
      gate remains the release authority.
  - id: asf-fail-closed-handoff-boundaries
    title: Fail-closed artifact, URL, REST, and live-proof handoff boundaries
    status: done
    scale: minor
    surface: validation
    owner: src/skilltree/summary.py
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      claims: 79
      figures: 26
      cited_sources: 188
      max_depth: 75
    evidence:
      - uv run pytest tests/test_dashboard.py tests/test_skilltree_client.py tests/test_skilltree_model.py
        tests/test_course_materials.py tests/test_full_app_stack.py -q
      - uvx ruff check src scripts tests
      - uvx ruff format --check src scripts tests
      - uv run mypy src
      - uv run pytest --cov=src --cov-fail-under=90
      - uv run python scripts/validate_rendered_freshness.py
      - uv run python scripts/validate_roadmap.py
    notes: >-
      A first-principles and RedTeam review removed several false-certification assumptions at
      system handoffs. Exported SkillTree JSON is now structurally validated before dashboard
      consumption and before writing the canonical artifact; malformed exports fail closed rather
      than appearing available with zero counts. REST base URLs require HTTP(S), path identifiers
      are encoded as single URL segments, course-material pointers accept only credential-free
      HTTP(S) URLs, browser rendering applies the same safe-link rule, derived alignment summaries
      must match their records, live proof summaries cannot override a failed subproof, and local
      Postgres password files reject symlinks/non-regular files and normalize owner-only mode.
      The validation-surface figure now exposes export-integrity and material-link-safety coverage.
      Remaining external deployment, learner-outcome, adaptive-recommendation, and public-release
      work stays explicitly open in the following TODO tasks rather than being implied by local
      gates.
    start: "2026-07-16"
    end: "2026-07-16"
    progress: 100
    milestone: false
    deps: []
    tags:
      - first-principles
      - redteam
      - fail-closed
      - handoff
  - id: asf-residue-and-verifier-cleanup
    title: Remove stale verifier exemptions and test-scaffolding residue
    status: done
    scale: minor
    surface: validation
    owner: src.live_counts
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      claims: 79
      figures: 26
      cited_sources: 188
      max_depth: 75
    evidence:
      - uv run pytest tests/test_roadmap.py tests/test_skilltree_exporter.py
        tests/test_dependency_reconciler.py tests/test_scripts_smoke.py
        tests/test_skilltree_local_auth.py -q
      - uvx ruff check src scripts tests
      - uvx ruff format --check src scripts tests
      - uv run mypy src
      - uv run pytest --cov=src --cov-fail-under=90
      - uv run python scripts/validate_live_counts.py
      - rg -n "\\b(Fake|fake|Stub|stub)\\b" src scripts tests || echo clean
    notes: >-
      Removed the compatibility-window exemption that allowed stale authored live counts to pass
      when nearby prose described an older export. The audit now reports every stale count literal;
      its regression test covers historical wording. Removed an unused legacy artifact identity,
      renamed test-only graph/client/result fixtures and subprocess probes to precise fixture names,
      and updated non-email credential diagnostics and stale terminology. Project policy documents
      retain explicit real-data testing rules; no fabricated collaborator remains in executable code.
    start: "2026-07-16"
    end: "2026-07-16"
    progress: 100
    milestone: false
    deps: []
    tags:
      - redteam
      - verifier
      - maintenance
  - id: asf-curriculum-expert-content-audit
    title: Expert audit of high-depth curriculum paths and quiz quality
    status: done
    surface: curriculum
    owner: src/curriculum/catalog.py
    evidence:
      - uv run python scripts/build_curriculum.py
      - uv run pytest tests/test_curriculum_catalog.py tests/test_curriculum_quality.py -q
      - uv run python scripts/export_skilltree.py
      - docs/curriculum_expert_audit.md
    notes: Recorded a deterministic expert-review sample of the max-depth SkillTree mastery chain,
      high-fanout EFE/VFE/Bayes/FEP nodes, and templated visual/publication/security extension
      subjects in docs/curriculum_expert_audit.md. No repeated vacuity or template-defect pattern
      was found beyond existing DAG, semantic-prerequisite, quiz, material and artifact gates.
  - id: asf-learner-outcome-pilot
    title: Learner outcome pilot and evidence boundary
    status: done
    surface: curriculum
    owner: experiment_plan.yaml
    evidence:
      - uv run python scripts/validate_claim_support.py
      - uv run python scripts/validate_manuscript_variables.py
      - uv run python scripts/build_curriculum.py
    notes: experiment_plan.yaml now contains a preregistered design-only learner-outcome pilot with
      consent/privacy boundaries, SkillTree-owned progress export, pre/post and delayed-retention
      measures, exclusion rules, and analysis plan. Manuscript setup and limitations explicitly
      state that no efficacy, retention, or transfer claim is made without real data and
      claim-support rows.
  - id: asf-docker-digest-lint
    title: Docker image digest pinning and supply-chain lint
    status: done
    surface: validation
    owner: docker-compose.yml
    evidence:
      - uv run python scripts/validate_docker_digests.py
      - unset VIRTUAL_ENV && uv run --active pytest tests/test_scripts_smoke.py -q
    notes: docker-compose images are pinned by registry-resolved @sha256 digest, not mutable tag, and
      validate_docker_digests.py fails the build if any image line lacks a digest so the
      supply-chain claim is enforced rather than aspirational.
  - id: asf-scholarship-role-quality
    title: Lane-token scholarship role-quality gate
    status: done
    surface: scholarship
    owner: src.scholarship
    evidence:
      - uv run python scripts/validate_scholarship_sources.py
      - unset VIRTUAL_ENV && uv run --active pytest tests/test_scholarship.py -q
    notes: Scholarship role text must name a lane-relevant content token, not merely clear a word-count
      floor, so a generic sentence cannot pass as a claim-specific role.
  - id: asf-redteam-gate-hardening
    title: RedTeam hardening of vacuous integrity gates
    status: done
    surface: validation
    owner: src.provenance
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
    evidence:
      - uv run python scripts/validate_claim_support.py
      - uv run python scripts/build_curriculum.py
      - uv run pytest tests/test_provenance.py tests/test_claim_support.py tests/test_scholarship.py
        tests/test_kernels_information.py -q
    notes: "An adversarial gate-vacuity audit (8 read-only specialists, independently refuted findings)
      fixed five gates that passed real data but would not fire on an injected same-class defect:
      provenance render now fails closed on whitespaced malformed claim tokens; computed claims
      require a backing token and have their prose scanned for hand-typed result numbers; the
      off-domain distractor gate covers an extended compute/infra/web/finance class (and caught a
      real 'browser cache' distractor); the scholarship role gate no longer counts lane-NAME tokens
      as content; the manuscript variable validator uses word boundaries. kernels.cross_entropy now
      accepts a non-normalized likelihood column. Every fix carries a negative-control test that
      fires on injection. A proposed MC-index threshold change was rejected: per-index 50% is the
      ideal for 2-of-4 selection; the joint correct-set share (16.8%) is the real gameability
      bound."
  - id: asf-semantic-prerequisite-gate
    title: Semantic prerequisite gate and DAG pedagogical fixes
    status: done
    surface: curriculum
    owner: src.curriculum.quality
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
    evidence:
      - uv run python scripts/build_curriculum.py
      - uv run pytest tests/test_curriculum_quality.py -q
    notes: "A new audit_semantic_prerequisites gate flags any skill assessed on a load-bearing concept
      (KL divergence, Markov chain, Laplace approximation, ELBO, Taylor expansion) without the
      concept's defining skill in its transitive prerequisites. Scoped to assessment text and to
      same-or-earlier-stratum definers so it fires on real inversions, not forward references. It
      surfaced three genuine inversions, now fixed: cross-entropy was taught before KL (dependency
      direction reversed to entropy->KL->cross-entropy), and the ELBO and continuous-Laplace skills
      gained their missing KL and Laplace prerequisites."
  - id: asf-curriculum-depth-message-passing-precision
    title: Curriculum depth - message passing and precision/exploration subjects
    status: done
    surface: curriculum
    owner: src.curriculum
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
    evidence:
      - uv run python scripts/build_curriculum.py
      - uv run python scripts/export_skilltree.py
      - uv run pytest tests/test_curriculum_catalog.py tests/test_curriculum_quality.py -q
    notes: "Two pedagogically-sound subjects close RedTeam content gaps. 'Message Passing on Factor
      Graphs' (computation) adds a five-skill ladder: factor graphs, the sum-product algorithm,
      belief propagation on trees, variational message passing, and loopy belief propagation.
      'Precision, Exploration and Guidance' (active inference) adds four skills unifying precision
      as confidence control, entropy-regularised exploration, the explore-exploit frontier, and
      risk-sensitive guidance functions. All nine skills pass the DAG, quality,
      semantic-prerequisite, and provenance gates."
  - id: asf-triplecheck-vacuity-and-depth
    title: Triple-check hardening (unicode bypass, fail-closed loaders) and curriculum depth
    status: done
    surface: validation
    owner: src.provenance
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
    evidence:
      - uv run python scripts/build_curriculum.py
      - uv run pytest tests/test_provenance.py tests/test_dashboard.py
        tests/test_curriculum_quality.py -q
    notes: "A second 7-specialist triple-check workflow (60 findings, 13 verified, 3 of the prior fixes
      re-confirmed correct) closed residual defects: zero-width/format characters (Unicode category
      Cf) bypassed both the malformed-claim-token check and the unbacked-number linter because
      neither \\s nor \\b match them, so provenance now strips Cf characters before any integrity
      scan (a registered token split by a zero-width space resolves cleanly; an unregistered or
      empty one still fails closed); the dashboard course-material loaders were fail-OPEN, returning
      available=True for a valid-JSON-but-shapeless artifact, and now fail closed with a descriptive
      error; the semantic-prerequisite concept patterns tolerate plural forms. Two more
      pedagogically-sound subjects were added: 'Model Reduction, Structure and Bound Tightness'
      (variational methods: Bayesian model reduction, structured vs mean-field approximations,
      sparse transition structure, the exact VFE / log-evidence bound gap) and 'Continuous-Time and
      Generalized-Coordinate AIF' (generalized coordinates of motion, discrete-continuous embedding,
      multi-scale temporal hierarchies). Every fix carries a firing negative-control."
  - id: asf-convergence-quiz-gating-auth-content
    title: Convergence pass - quiz-gating integrity, fail-closed auth, rate-distortion bridge
    status: done
    surface: validation
    owner: src.skilltree
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
    evidence:
      - uv run python scripts/export_skilltree.py
      - uv run pytest tests/test_skilltree_model.py tests/test_skilltree_local_auth.py
        tests/test_curriculum_catalog.py -q
    notes: A third convergence workflow (37 findings, 12 verified) closed the remaining real defects
      after the first two rounds had hardened the content gates. SkillTreeProject now validates quiz
      gating at the project level (a Quiz-typed skill must carry a quizId that maps to a real quiz;
      no dangling references; no orphan quizzes), which the prior per-skill validation could not
      see. The local trusted-client token exchange was fail-OPEN - an empty or token-less HTTP 200
      was treated as a verified authentication - and now requires a non-empty access_token. Seed
      verification was extended from a skill-only count to also check subjects, so a dropped subject
      (whose skills are still summed) can no longer pass silently. A rate-distortion to
      expected-free-energy bridging skill was added (the epistemic-pragmatic trade-off as a
      rate-distortion trade-off). Every fix carries a firing negative-control; the keyword-set and
      docker-comment findings were assessed as acceptable cosmetic variation, a convergence signal.
  - id: asf-manuscript-modularization
    title: Modular manuscript split and lettered-section validation
    status: done
    surface: manuscript
    owner: manuscript
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      claims: 79
      figures: 26
      cited_sources: 188
    evidence:
      - uv run pytest tests/test_claim_support.py tests/test_manuscript_artifacts.py
        tests/test_docs_inventory.py tests/test_kernels_active_inference.py
        tests/test_kernels_distributions.py tests/test_kernels_examples.py
        tests/test_kernels_inference.py tests/test_kernels_information.py -q
      - uv run ruff check src/manuscript/claim_support.py src/scholarship/coverage.py
        tests/test_claim_support.py tests/test_manuscript_artifacts.py
    notes: "Completed the authored manuscript modularization pass: methodology, results, and
      scope/related-work spines now delegate to lettered modules while preserving claim-support
      anchors, citations, figure labels, and evidence boundaries. The claim-support and scholarship
      scanners now include lettered section filenames, and the manuscript artifact gate passed
      against generated figure labels."
  - id: asf-skill-material-figure-readability-hardening
    title: Skill material linkage and figure readability hardening
    status: done
    surface: figures
    owner: src.figures
    metrics:
      skills: 630
      figures: 26
    evidence:
      - uv run pytest tests/test_course_materials.py tests/test_figures.py -q
      - uv run python scripts/validate_course_material_sources.py --online
      - uv run python scripts/run_kernels.py
      - uv run python scripts/validate_manuscript_artifacts.py
      - uv run python scripts/validate_roadmap.py
    notes: Figure 9 now combines scholarship support rows with generated course-material coverage so the
      Active Inference Institute cognitive knowledge-base linkage is visible as pinned source
      evidence. Course-material records expose source, role, match-scope, topic-match, and cognitive
      knowledge-base coverage counts. Figure typography now has a source-level minimum-font
      contract; Figure 3 uses fixed metric/chip columns for the top fan-out review queue, and Figure
      4 uses a checkpoint ledger instead of angled micro-labels.
  - id: asf-visualization-equal-pass-hardening
    title: Equal-pass figure readability and dashboard preview hardening
    status: done
    surface: figures
    owner: src.figures
    metrics:
      figures: 26
    evidence:
      - uv run pytest tests/test_figures.py tests/test_manuscript_artifacts.py
        tests/test_dashboard.py tests/test_docs_inventory.py tests/test_roadmap.py -q
      - uv run python scripts/run_kernels.py
      - uv run python scripts/validate_manuscript_artifacts.py
      - uv run python scripts/validate_rendered_freshness.py
      - uv run python scripts/validate_web_math.py
    notes: All 21 registered figures plus the generated cover and page-two skill atlas received the same
      readability pass. Shared figure constants now require meaningful chart text at or above a
      9.5px-equivalent floor, reserve 8.5px for source/caveat notes, and reject hard-coded sub-floor
      font sizes. Dense outputs were rebalanced with wider layouts, wrapped stratum chips, reduced
      atlas term/checkpoint density, and a wider checkpoint ledger rather than micro-type. Dashboard
      figure cards now present a full-size image link first, group Question / Grain / Boundary /
      Failure metadata, and move encoding/QA details behind disclosure.
  - id: asf-scholarship-frontier-gap-report
    title: Scholarship frontier gap report and claim-bound source upgrade
    status: done
    surface: scholarship
    owner: src.scholarship.gap_report
    metrics:
      cited_sources: 188
    evidence:
      - uv run python scripts/analyze_scholarship.py
      - uv run python scripts/audit_scholarship_gaps.py
      - uv run python scripts/validate_scholarship_sources.py --online --strict-online
      - uv run python scripts/validate_claim_support.py
      - uv run pytest tests/test_scholarship.py tests/test_claim_support.py
        tests/test_manuscript_artifacts.py -q
    notes: Added nine curated frontier sources for active learning, cognitive engagement, retrieval
      practice, evidence-centered assessment, FAIR/open-science reporting, secure
      software-development posture, WCAG accessibility, and the Active Inference Institute cognitive
      knowledge-base. The new scholarship_gap_report.json binds each frontier source to manuscript
      citation, BibTeX, source-role and claim-support evidence, with boundary-sensitive sources
      requiring a limitation or scope-limit row so scholarship cannot be mistaken for
      learner-efficacy, public-release or security-assurance evidence.
  - id: asf-scholarship-citation-augmentation-ii
    title: Formal provenance, learning-space and reproducibility citation augmentation
    status: done
    surface: scholarship
    owner: src.scholarship.gap_report
    metrics:
      cited_sources: 188
    evidence:
      - uv run python scripts/analyze_scholarship.py
      - uv run python scripts/audit_scholarship_gaps.py
      - uv run python scripts/validate_scholarship_sources.py --online --strict-online
      - uv run python scripts/validate_claim_support.py
      - uv run pytest tests/test_scholarship.py tests/test_claim_support.py
        tests/test_manuscript_artifacts.py tests/test_roadmap.py -q
    notes: Added twelve directly verified frontier sources for W3C PROV, knowledge-space and
      learning-space theory, constructive alignment, formative assessment and feedback, design-based
      research, scientific-computing practice, and reproducible-science boundaries. The scholarship
      gap report now requires 21 frontier sources and binds the added citations to manuscript
      claim-support rows so formal provenance, learning theory, and reproducibility sources cannot
      enter as bibliography padding or learner-efficacy claims.
  - id: asf-frontier-scholarship-2026-upgrade
    title: Frontier scholarship and boundary-source upgrade
    status: done
    surface: scholarship
    owner: src.scholarship.gap_report
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      cited_sources: 188
    evidence:
      - uv run python scripts/validate_scholarship_sources.py --online --timeout 10
      - uv run python scripts/validate_claim_support.py
      - uv run python scripts/analyze_scholarship.py
      - uv run python scripts/audit_scholarship_gaps.py
      - uv run python scripts/run_kernels.py
      - uv run python scripts/validate_manuscript_artifacts.py
      - uv run pytest tests/test_scholarship.py tests/test_claim_support.py
        tests/test_curriculum_catalog.py -q
    notes: Added eight verified frontier sources for Workflow Run RO-Crate, curriculum-aware
      cognitive diagnosis, reactive message passing, RxInfer, coupled free energy, active
      predictive coding, divide-and-conquer predictive coding, and Active Inference teacher
      development. The graph remains stable at 630 skills, 111 subjects and 1199 edges; the gap
      report now requires 29 frontier sources and binds the new citations to explicit scope or
      implementation boundaries rather than learner-outcome or production-scale claims.
  - id: asf-dashboard-offline-state-polish
    title: Dashboard offline states and handoff affordance polish
    status: done
    scale: minor
    surface: dashboard
    owner: src.dashboard.assets
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      figures: 26
    evidence:
      - uv run pytest tests/test_dashboard.py tests/test_dashboard_server.py
        tests/test_dashboard_console.py -q
      - uv run python scripts/validate_dashboard_console.py --require-browser --require-dashboard
    notes: >-
      Dashboard offline, local-auth, and missing-material states now use explicit read-only shell
      language, clearer Learn/Quiz handoff labels, companion-aware figure cards, and accessible
      status treatments without moving progress, ranking, quizzes, or learner state out of
      SkillTree. Evidence passed against the local dashboard browser console and server/dashboard
      test surfaces.
  - id: asf-release-file-review-pass
    title: Dirty worktree release-file review and inventory refresh
    status: done
    scale: minor
    surface: validation
    owner: src.release_readiness
    evidence:
      - git status --short
      - uv run python scripts/validate_release_readiness_inventory.py --write
      - uv run python scripts/validate_release_readiness_inventory.py
      - git diff --check
    notes: >-
      The current tracked and untracked release-facing paths were re-inventoried without staging or
      committing. docs/release_readiness_inventory.md was regenerated from live git state and
      validated so the release boundary stays explicit while this pass remains unpublished.
  - id: asf-doc-command-drift-audit
    title: Documentation command parity and stale-count audit
    status: done
    scale: minor
    surface: documentation
    owner: docs
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      cited_sources: 188
    evidence:
      - uv run pytest tests/test_docs_inventory.py tests/test_scripts_smoke.py
        tests/test_roadmap.py -q
      - uv run python scripts/validate_live_counts.py
      - uv run python scripts/validate_roadmap.py
    notes: >-
      README, docs, and script examples now describe the current artifact-only versus Docker/live
      split, generated diagnostic artifacts, companion validation, and dashboard behavior. Live
      counts and TODO output remain validator-sourced through validate_live_counts.py and
      validate_roadmap.py.
  - id: asf-curriculum-diagnostic-gap-map
    title: Curriculum diagnostic gap map and remediation queue
    status: done
    scale: medium
    surface: curriculum
    owner: src.curriculum
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
      max_depth: 75
    evidence:
      - uv run python scripts/build_curriculum.py
      - uv run python scripts/run_kernels.py
      - uv run pytest tests/test_curriculum_catalog.py tests/test_curriculum_quality.py
        tests/test_figures.py -q
    notes: >-
      Added the source-owned curriculum_diagnostic_gap_map.json artifact with bottlenecks, long
      prerequisite chains, assessment-balance checks, and a remediation queue across the 630-skill
      graph. The dashboard reads it as artifact diagnostics only and does not convert it into a
      learner-outcome claim.
  - id: asf-scholarship-source-health-network
    title: Scholarship source-health network and boundary report
    status: done
    scale: medium
    surface: scholarship
    owner: src.scholarship
    metrics:
      cited_sources: 188
    evidence:
      - uv run python scripts/analyze_scholarship.py
      - uv run python scripts/validate_scholarship_sources.py
      - uv run python scripts/validate_scholarship_sources.py --online --timeout 10
      - uv run pytest tests/test_scholarship.py tests/test_claim_support.py -q
    notes: >-
      Added scholarship_source_health_network.json as a readable source-health report separating
      dead links, publisher bot-blocks, role gaps, boundary sources, metadata issues, and uncited
      drift. Local and online source-health evidence passed while retaining the boundary that
      scholarship supports scope and implementation claims, not learner-efficacy or security
      assurance.
  - id: asf-figure-table-companion-accessibility
    title: Figure table companions and nonvisual reading path
    status: done
    scale: medium
    surface: figures
    owner: src.figures
    metrics:
      figures: 26
    evidence:
      - uv run python scripts/run_kernels.py
      - uv run python scripts/validate_manuscript_artifacts.py
      - uv run pytest tests/test_figures.py tests/test_manuscript_artifacts.py
        tests/test_dashboard.py -q
    notes: >-
      Added generated figure_table_companions.json rows for the highest-density figures, optional
      table_companion_id linkage in the figure registry, fail-closed manuscript validation for
      missing or malformed companions, and dashboard rendering that exposes the table path for
      screen-reader and small-screen reads without relying on color or dense image text.
  - id: asf-scholarship-phase-planning
    title: Scholarship lanes and evidence plans for upcoming phases
    status: done
    scale: major
    surface: scholarship
    owner: src.scholarship
    metrics:
      cited_sources: 188
    evidence:
      - uv run python scripts/analyze_scholarship.py
      - uv run python scripts/audit_scholarship_gaps.py
      - uv run python scripts/validate_scholarship_sources.py --online --strict-online
      - uv run python scripts/validate_claim_support.py
      - uv run pytest tests/test_scholarship.py tests/test_claim_support.py -q
    notes: >-
      Maintain a phase-wide source and evidence plan for public release metadata, SkillTree runtime
      proof, live-stack rehearsal, learner-pilot design, adaptive-path research, and archival release.
      Rebalanced perception-as-inference, mathematical-foundation, and applications/scope lanes to
      the balanced threshold; resolved all single-section claim-support advisories; added only
      verified source-role rows and citations; and kept Active Inference Institute references
      contextual, community-facing, and non-endorsing. Strict online checking uses a 15-second
      per-locator timeout to avoid false negatives from slow public scholarly hosts.
    scholarship_plan:
      lanes:
        - perception_as_inference
        - mathematical_foundations
        - pedagogy_and_learning_paths
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys:
        - dayan1995helmholtz
        - shannon1948mathematical
        - beal2003variational
        - hodson2024empirical
        - liu2023_learning_analytics_privacy
        - activeinferenceinstitute2026website
      planned_source_targets:
        - key: sclater2015learning_analytics_code
          status: planned
          locator: url:https://www.jisc.ac.uk/guides/code-of-practice-for-learning-analytics
          role: Responsible learning-analytics source for transparency, consent, privacy, retention, validity, and intervention boundaries.
        - key: eldridge2016consort_pilot
          status: planned
          locator: doi:10.1186/s40814-016-0079-6
          role: Pilot and feasibility reporting source for distinguishing feasibility from definitive outcome evaluation.
        - key: hoffmann2014tidier
          status: planned
          locator: doi:10.1136/bmj.g1687
          role: Intervention-description and replication source for specifying a future learner study.
        - key: piech2015deep_knowledge_tracing
          status: planned
          locator: url:https://proceedings.neurips.cc/paper_files/paper/2015/file/bac9162b47c56fc8a4d2a519803d51b3-Paper.pdf
          role: Learner-state modelling comparison that bounds, rather than validates, adaptive tutoring claims.
        - key: tao2024q_matrix_constraint
          status: planned
          locator: doi:10.3758/s13428-024-02404-5
          role: Q-matrix cognitive-diagnosis source for assessment-model and learner-state boundaries.
      claim_support_types:
        - literature
        - scope_limit
        - delivery
      boundaries:
        - Source count is not scholarship quality; every source must have a claim-specific role and appropriate scope language.
        - Historical, community, software, and pedagogy sources motivate design context but do not determine prerequisite edges or learner outcomes.
        - Active Inference Institute resources are cited as public ecosystem infrastructure without endorsement or institutional ownership claims.
      phase_exit:
        - The gap report has no unresolved frontier contract issues or unsupported single-section source advisories.
        - Each open roadmap task has a rendered source package, boundary statement, and validator-backed exit condition.
  - id: asf-skilltree-auth-deployment-matrix
    title: SkillTree authentication and deployment-mode matrix
    status: done
    scale: medium
    surface: skilltree
    owner: src.skilltree
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      dependencies: 1199
      quizzes: 630
    evidence:
      - docker info
      - uv run pytest tests/test_skilltree_client.py tests/test_skilltree_local_auth.py
        tests/test_dashboard_server.py -q
      - uv run python scripts/validate_full_app_stack.py --app-ready --skip-regenerate
      - "proof.ok=true; deployment matrix 4/4 ready (local FORM bootstrap, ignored local
        state, token-backed admin seeding, externally operated SkillTree);
        browserSafe=yes; rawSkillTreeLogin adminSurfaceVisible=true, title='SkillTree
        Dashboard - Project Administrator'; browserConsole 'dashboard console gate OK'"
    notes: >-
      Closed 2026-08-09 on a host with a running Docker daemon (docker info succeeds;
      skills-service 4.6.0 + postgres:15 at the pinned digests). The source-owned deployment
      matrix reports 4/4 ready with a browser-safe payload, and the raw FORM login proof
      reaches the authenticated administrator surface without exposing the admin password,
      client secret, or bearer token to browser JavaScript. Two real defects surfaced only
      once live proof could actually run, both pre-existing and both fixed here: (1) the
      dashboard CSP had no frame-src and no ws:// origin, so the official skills-client-js
      learner iframe was blocked by the default-src 'none' fallback and its progress socket
      was refused — the Official Display panel rendered as an empty box while every offline
      gate stayed green; (2) a stale chrome-devtools-axi session made every browser command
      time out, which failed the whole live proof with an opaque message, so the validator
      now probes the browser session and restarts it once before running any browser proof.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - pedagogy_and_learning_paths
      existing_source_keys:
        - skilltree2020nsa
        - skilltree2026docs
        - nist2022ssdf
        - w3c2023wcag22
      planned_source_targets: []
      claim_support_types:
        - delivery
        - scope_limit
        - literature
      boundaries:
        - Offline REST and browser-contract tests do not establish a live deployment.
        - SkillTree ownership and security posture remain distinct from this curriculum artifact.
      phase_exit:
        - docker info succeeds on the validation host.
        - App-ready FORM-login, dashboard, and artifact proofs pass without exposed secrets.
  - id: asf-live-stack-ci-rehearsal
    title: Repeatable live-stack CI rehearsal with full dependency parity
    status: done
    scale: major
    surface: validation
    owner: scripts/validate_full_app_stack.py
    deps:
      - asf-skilltree-auth-deployment-matrix
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      dependencies: 1199
      quizzes: 630
      assessments: 2520
    evidence:
      - docker info
      - uv run python scripts/validate_full_app_stack.py --app-ready --skip-regenerate
      - uv run python scripts/validate_full_app_stack.py --skip-regenerate
      - "proof.ok=true; proof.mode=full-parity; liveDependencies matched=1199/1199,
        missing=0, extra=0; liveAdminProject.skills=630; counts skills=630, subjects=111,
        quizzes=630; artifacts present=42, missing=0; figureContracts registered=21,
        dashboard=21; courseMaterials records=630, externalPointers=2026; browserConsole
        'dashboard console gate OK'; rawSkillTreeLogin adminSurfaceVisible=true"
      - uv run python scripts/reconcile_live_dependencies.py --attempts-per-edge 3
        (complete=true, matched 1199/1199)
    notes: >-
      Closed 2026-08-09. The strict rehearsal ran end to end in full-parity mode against
      skills-service 4.6.0 and emitted the redacted proof artifact at
      .active_skillference_local/full_app_stack_proof.json. Reaching 1199/1199 exposed a real
      reconciler defect: at 600-skill scale on emulated Docker a single prerequisite POST can
      exceed the 600s client timeout, and the run aborted even when the edge was already
      visible in the authoritative live graph on the very next read. Two fixes: the post-error
      authoritative read now confirms such an edge instead of failing, and attempts-per-edge
      (default 2) bounds retries so one slow write no longer discards the remaining progress
      of a 1199-edge run. Full parity was then reached and the strict proof passed. Note the
      cost profile for future runs: individual late-stratum edges took 10-20 minutes each on
      this Apple/ARM host, so full parity is a long-running proof, not a per-commit gate.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - pedagogy_and_learning_paths
      existing_source_keys:
        - sandve2013reproducible
        - wilson2014scientific_computing
        - leo2024workflow_run_rocrate
        - skilltree2026docs
      planned_source_targets: []
      claim_support_types:
        - delivery
        - scope_limit
        - literature
      boundaries:
        - CI rehearsal evidence is not equivalent to production availability or learner efficacy.
        - Dependency parity must be measured through the live SkillTree API, not inferred from export files.
      phase_exit:
        - Docker readiness, seeded state, FORM login, browser console, and dependency parity are recorded.
        - Release wording names the exact live validator and host conditions.
  - id: asf-empirical-learner-pilot-package
    title: Empirical learner pilot protocol and analysis package
    status: done
    scale: major
    surface: validation
    owner: docs
    metrics:
      skills: 630
      subjects: 111
      assessments: 2520
    evidence:
      - uv run python scripts/validate_claim_support.py
      - uv run python scripts/analyze_scholarship.py
      - uv run python scripts/validate_learner_pilot_contract.py
      - uv run pytest tests/test_pilot_contract.py tests/test_claim_support.py -q
    notes: >-
      Added a protocol-only learner-pilot contract with consent/privacy and minimization fields,
      participant/event/export schemas, measurement-readiness outcomes, missing-data handling,
      preregistration and analysis-notebook fields, and explicit conditional manuscript language.
      The validator is deterministic and does not accept learner data or efficacy claims.
    scholarship_plan:
      lanes:
        - pedagogy_and_learning_paths
        - applications_and_scope
        - delivery_and_reporting
      existing_source_keys:
        - liu2023_learning_analytics_privacy
        - designbasedresearchcollective2003
        - barab2004design_based
        - mislevy2003ecd
        - black1998assessment
        - shute2008formative_feedback
      planned_source_targets:
        - key: sclater2015learning_analytics_code
          status: planned
          locator: url:https://www.jisc.ac.uk/guides/code-of-practice-for-learning-analytics
          role: Responsible learning-analytics source for transparency, consent, privacy, retention, validity, and intervention boundaries.
        - key: eldridge2016consort_pilot
          status: planned
          locator: doi:10.1186/s40814-016-0079-6
          role: Pilot and feasibility reporting source for a study that precedes definitive outcome evaluation.
        - key: hoffmann2014tidier
          status: planned
          locator: doi:10.1136/bmj.g1687
          role: Intervention-description and replication source for specifying the planned learning intervention.
      claim_support_types:
        - literature
        - scope_limit
        - delivery
      boundaries:
        - A pilot protocol establishes feasibility and measurement readiness, not learner efficacy, retention, or transfer.
        - Consent, minimization, retention, access, missing-data, and stopping rules must be specified before learner exports are analyzed.
      phase_exit:
        - The protocol, export schema, preregistration fields, analysis notebook contract, and conditional manuscript language are complete.
        - No learner-outcome claim is added until real data and a matching claim-support record exist.
  - id: asf-adaptive-path-recommendation-prototype
    title: Adaptive path recommendation research prototype
    status: done
    scale: major
    surface: curriculum
    owner: src.curriculum
    metrics:
      skills: 630
      subjects: 111
      prerequisite_edges: 1199
      assessments: 2520
    evidence:
      - uv run python scripts/build_curriculum.py
      - uv run python scripts/export_skilltree.py
      - uv run pytest tests/test_recommendations.py tests/test_dashboard_server.py
        tests/test_dashboard.py -q
    notes: >-
      Implemented a deterministic graph-only recommendation prototype using prerequisite structure,
      assessment concepts, and explicitly supplied SkillTree-owned progress signals. The dashboard
      exposes an additive /api/recommendations surface and routes learners back to official
      SkillTree Learn/Quiz surfaces; it stores no independent learner state and makes no efficacy,
      knowledge-tracing, cognitive-diagnosis, or autonomous-tutoring claim.
    scholarship_plan:
      lanes:
        - pedagogy_and_learning_paths
        - applications_and_scope
      existing_source_keys:
        - doignon1985knowledge_spaces
        - falmagne2011learning_spaces
        - fu2025curriculum_aware_gncd
        - activeinferenceinstitute2026website
        - activeinferenceinstitute2026courses
        - activeinferenceinstitute2026cognitive
      planned_source_targets:
        - key: piech2015deep_knowledge_tracing
          status: planned
          locator: url:https://proceedings.neurips.cc/paper_files/paper/2015/file/bac9162b47c56fc8a4d2a519803d51b3-Paper.pdf
          role: Learner-state modelling comparison that defines, rather than supplies, evidence for adaptive tutoring.
        - key: tao2024q_matrix_constraint
          status: planned
          locator: doi:10.3758/s13428-024-02404-5
          role: Cognitive-diagnosis and Q-matrix source for explicit assessment-model and learner-state boundaries.
      claim_support_types:
        - literature
        - scope_limit
        - delivery
      boundaries:
        - Version one selects deterministic graph candidates from SkillTree-owned progress signals and does not infer latent learner state.
        - Knowledge tracing, cognitive diagnosis, autonomous tutoring, personalization quality, and efficacy remain future comparisons.
      phase_exit:
        - Offline fixtures prove deterministic recommendations, explanation text, and official Learn/Quiz routing.
        - The dashboard stores no independent learner state and makes no effectiveness claim.
  - id: asf-public-archive-provenance-release
    title: Public archive, DOI, and provenance release package
    status: todo
    scale: major
    surface: metadata
    owner: manuscript/config.yaml
    deps:
      - asf-public-release-metadata
    evidence:
      - git ls-remote --heads https://github.com/ActiveInferenceInstitute/Active_Skillference.git
        HEAD
      - uv run python scripts/validate_release_readiness_inventory.py --write
      - uv run python scripts/validate_release_readiness_inventory.py
      - uv run python scripts/validate_manuscript_variables.py
      - uv run python scripts/validate_rendered_freshness.py
    notes: >-
      The public target exists and is public but currently holds zero refs, so reachability is no
      longer the gate; published content is. After the target carries a real published release,
      complete the publication metadata chain: release tag, Zenodo DOI, CFF/CodeMeta/manuscript URL
      fields, artifact hashes, third-party notices, counsel-review checklist, and release inventory
      refresh. This task must not fabricate DOI or repository metadata before the upstream blocker
      is cleared.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys:
        - smith2016softwarecitation
        - citationfileformat2026
        - codemeta2026schema
        - github2026citingcontent
        - zenodo2026github
        - wilkinson2016fair
        - nosek2015top
      planned_source_targets: []
      claim_support_types:
        - delivery
        - scope_limit
        - literature
      boundaries:
        - DOI, public URLs, and release timestamps are blank until the repository and archive events occur.
        - Metadata standards support release reproducibility, not scientific validity, legal advice, or learner outcomes.
      phase_exit:
        - A real public tag, archive deposit, DOI, CFF/CodeMeta metadata, hashes, notices, and inventory agree.
        - The final release record is regenerated from observed external metadata rather than placeholders.

  - id: asf-review-mypy-gate-unbuildable
    title: 'DONE: major: mypy gate unbuildable on current interpreter (zero type-checking runs)'
    status: done
    scale: major
    surface: validation
    owner: pyproject.toml
    evidence:
      - uv run mypy src
      - uvx ruff check src scripts tests
    notes: >-
      The documented gate `uv run mypy src` cannot run. It fails on the imported
      numpy type stub (numpy/__init__.pyi:737 "Type statement is only supported in
      Python 3.12 or greater"), so mypy aborts before checking any src module: zero
      static type-checking currently happens project-wide. Root cause is a
      pyproject.toml/environment mismatch: `[tool.mypy] python_version` is "3.10"
      while the active interpreter is Python 3.12/3.13 and installed numpy ships a
      PEP-695 `type`-statement stub that only parses under >=3.12. Fix: set mypy
      python_version to the interpreter actually in use (>=3.12), or pin numpy to a
      version whose stubs parse under 3.10, then run the type gate successfully and
      enforce it on CI.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - scope_limit
      boundaries:
        - "Static type checking is a repo-health signal in this gate set, not a claim about runtime behaviour."
      phase_exit:
        - "`uv run mypy src` returns 0 with no module skipped."
  - id: asf-review-ruff-gate-red
    title: 'DONE: major: ruff gate red (355 check errors, 4 format failures)'
    status: done
    scale: major
    surface: validation
    owner: src scripts tests
    evidence:
      - uvx ruff check src scripts tests
      - uvx ruff format --check src scripts tests
    notes: >-
      The documented gate `uvx ruff check src scripts tests && uvx ruff format
      --check src scripts tests` is red on the current tree: 355 check errors (126
      RUF100 unused-noqa, 60 I001 unsorted-imports, 30 EXE001 shebang-not-executable,
      30 ISC004 implicit-string-concat, 26 PLC0414 useless-import-alias, 20 FURB167
      regex-flag-alias, 19 UP035 deprecated-import, 10 PLW1510
      subprocess-run-without-check, misc) and 4 files fail `ruff format --check`;
      242 are auto-fixable. Because AGENTS.md declares ruff a release gate, the
      "green gates" claim does not currently hold. Fix the gate (auto-fix and format,
      then justify or silence the intentional remainder) and add a CI lint job so it
      cannot silently regress.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - scope_limit
      boundaries:
        - "Lint/style gates are developer-workflow hygiene, not scientific or learner-outcome evidence."
      phase_exit:
        - "`uvx ruff check src scripts tests` and `uvx ruff format --check src scripts tests` both return 0."
  - id: asf-review-coverage-floor-illusion
    title: 'FIX-ME major: >=90% coverage aggregate masks 32% security-module coverage'
    status: todo
    scale: major
    surface: validation
    owner: tests
    evidence:
      - uv run pytest --cov=src --cov-fail-under=90
    notes: >-
      Progress through 2026-08-01: aggregate src coverage is 92.03% (up from 90.30%) and the
      previously-thin logic/validation modules are now genuinely covered: summary.py 85% ->
      99%, curriculum/alignment.py 66% -> 95%, pilot/contract.py 83% -> 95%,
      form_login_validation guard branches 32% -> 50%, github_sources retry paths
      (5xx/network/exhaustion), and recommendations ordering/cycle/remediation. The
      aggregate floor no longer greenwashes these core modules. Remaining sub-90 modules are
      all live/network/browser gated and need a Docker/browser-capable host (they are folded
      into the live-stack tasks): full_app_stack.py ~61% (Docker), form_login_validation.py
      ~50% (browser subprocess), source_health_network.py ~72% (live HTTP),
      dashboard/payload.py ~68% (live-auth block, httpserver). Add live/browser-capable
      coverage for those before claiming the floor is uniformly green.
      2026-08-09 update: aggregate is 92.08% over 923 tests. The four named modules are still
      the sub-90 set (full_app_stack.py 63.77%, form_login_validation.py 49.76%,
      source_health_network.py 72.39%, dashboard/payload.py 70.37%), and this pass confirmed
      empirically why the exception matters rather than assuming it: the dashboard CSP bug that
      blanked the official SkillTree iframe was invisible to all 900+ offline tests and was
      caught only by the live browser console gate, and the reconciler's abort-on-slow-write
      bug was caught only by a real 1199-edge run. Offline unit coverage of these modules would
      not have found either. The regression tests added here are the mechanically checkable
      part (browser-session restart, post-timeout confirmation, retry bounds, CSP directives,
      deferred parity, gzip), and the live gates remain the proof for the rest.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
      boundaries:
        - "Coverage measures exercised code paths, not correctness or safety."
      phase_exit:
        - "Every sub-90 module meets the 90% floor or has a documented browser/Docker path exception."
  - id: asf-review-token-endpoint-info-leak
    title: 'DONE: medium: token endpoint leaks SkillTreeClientError body/status and auth error text'
    status: done
    scale: medium
    surface: dashboard
    owner: src/dashboard/server.py
    evidence:
      - uv run pytest tests/test_dashboard_server.py -q
    notes: >-
      /api/skilltree/token (_send_skilltree_token) returns SkillTreeClientError.status
      plus str(exc) for the local-auth and token-exchange branches, which can surface
      raw SkillTree server response bodies and internal error state to any same-origin
      caller. Local-only scope limits impact, but the endpoint should return fixed,
      sanitized messages and log diagnostic detail server-side instead of echoing
      upstream error bodies.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
        - scope_limit
      boundaries:
        - "The dashboard is a local dev surface, so impact is low; errors must still not echo upstream bodies."
      phase_exit:
        - "/api/skilltree/token returns only fixed messages; raw SkillTree body/status stays in server logs."
  - id: asf-review-token-endpoint-no-csrf
    title: 'DONE: medium: token dispenser has no CSRF token; allows Origin-less callers'
    status: done
    scale: medium
    surface: skilltree
    owner: src/dashboard/server.py
    evidence:
      - uv run pytest tests/test_dashboard_server.py tests/test_skilltree_local_auth.py -q
    notes: >-
      _send_skilltree_token has no anti-CSRF token of its own; its caller gate
      (_token_caller_allowed) relies on Host-header pinning plus Sec-Fetch-Site/Origin
      heuristics and explicitly ALLOWS requests that carry no Origin (fetch_site in
      same-origin or none). A cross-site context that omits Origin can reach the
      endpoint. Defense-in-depth: require a per-session CSRF token or an explicit
      allow-origin-only policy before issuing learner credentials.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
      boundaries:
        - "Credential issuance is protected today by Host pinning plus Origin/Sec-Fetch-Site heuristics."
      phase_exit:
        - "The token endpoint issues learner credentials only to verified same-origin callers presenting a CSRF token or an allowed Origin."
  - id: asf-review-one_shot_secret_url-cors-star
    title: 'DONE: medium: loopback secret server uses Access-Control-Allow-Origin: * over plaintext'
    status: done
    scale: medium
    surface: skilltree
    owner: src/skilltree/form_login_validation.py
    evidence:
      - uv run pytest tests/test_skilltree_form_login.py tests/test_form_login_core.py -q
    notes: >-
      one_shot_secret_url serves the loopback admin password over plaintext HTTP from
      a transient 127.0.0.1 server whose OPTIONS/GET return
      Access-Control-Allow-Origin: *. A 24-character single-use URLsafe token is the
      only guard. Acceptable for local tooling, but tighten by binding to the expected
      SkillTree origin (or documenting the accepted local-trust trade-off) and add a
      test proving the token cannot be read twice.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
        - scope_limit
      boundaries:
        - "The transient loopback server is explicitly local-only tooling guarding a single-use token."
      phase_exit:
        - "The secret server binds its CORS origin or documents the local-trust trade-off; single-use is covered by a test."
  - id: asf-review-assign_dependency-non-atomic-reseed
    title: 'DONE: medium: assign_dependency delete-then-reseed is non-atomic'
    status: done
    scale: medium
    surface: skilltree
    owner: src/skilltree/local_auth.py
    evidence:
      - uv run pytest tests/test_skilltree_seeder.py tests/test_dependency_reconciler.py -q
    notes: >-
      SkillTreeSessionClient.assign_dependency deletes a stale reverse prerequisite
      and then POSTs the replacement. If the DELETE succeeds but the follow-up POST
      fails, live SkillTree is left missing that edge while the in-memory
      _dependency_pairs cache was already discarded, so a later run can mis-seed
      non-deterministically with no recovery path. Make the delete-then-reseed
      transactional, or re-fetch the live graph on the failure path before raising.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
      boundaries:
        - "Reseed is a maintenance operation against a live local SkillTree; partial failure must not persist silently."
      phase_exit:
        - "A failed delete-then-reseed leaves the live graph consistent or raises with the dependency cache refreshed."
  - id: asf-review-live-network-no-retries
    title: 'DONE: medium: live-network proof paths lack retries/backoff'
    status: done
    scale: medium
    surface: validation
    owner: src/integration/full_app_stack.py, src/curriculum/materials/github_sources.py
    evidence:
      - uv run python scripts/validate_full_app_stack.py --skip-regenerate
      - uv run python scripts/validate_course_material_sources.py --online
    notes: >-
      The live-network validators (curriculum GitHub-tree fetch, full-app-stack
      Docker+browser proof, validate_live_skilltree_dashboard) apply per-call timeouts
      but no retry/backoff on transient failures, which makes the "repeatable live-stack
      CI rehearsal" goal flaky and produces non-deterministic red runs. Add bounded
      retries with explicit backoff (and clear partial-failure reporting) to the
      Docker/browser/network proof paths.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
        - scope_limit
      boundaries:
        - "Live-proof runs depend on Docker/browser/network availability on the validation host."
      phase_exit:
        - "Network proof paths apply bounded retries/backoff and report partial failure explicitly."
  - id: asf-review-recommendation-scoring
    title: 'DONE: medium: recommendation score heuristic can dominate dependency order'
    status: done
    scale: medium
    surface: curriculum
    owner: src/curriculum/recommendations.py
    evidence:
      - uv run pytest tests/test_recommendations.py tests/test_dashboard.py -q
    notes: >-
      recommend_next_skills scores candidates as category*100 + concept_overlap*10 -
      depth. The token-overlap term (10/word) can exceed the depth penalty (up to +75),
      and unsuccessful-retake candidates (category 3, baseline 300) always rank above
      every satisfied-frontier skill regardless of dependency order; this ordering is
      undocumented and the module is only 84.8% covered. If ordering is meant to respect
      the prerequisite graph, weight along topological depth first; add unit tests for
      ordering and the positive-limit boundary.
    scholarship_plan:
      lanes:
        - pedagogy_and_learning_paths
        - applications_and_scope
        - delivery_and_reporting
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - scope_limit
      boundaries:
        - "Recommendations are an offline prototype and never infer learner state or efficacy."
      phase_exit:
        - "Ordering is determined by a documented, dependency-respecting rule; ordering and limit boundaries are unit-tested."
  - id: asf-review-unused-noqa-126
    title: 'POLISH minor: 126 unused noqa directives (RUF100) block ruff gate'
    status: done
    scale: minor
    surface: validation
    owner: src scripts tests
    evidence:
      - uvx ruff check src scripts tests
    notes: >-
      126 RUF100 (unused noqa) violations across the tree keep the ruff gate red;
       representative example at tests/test_web_math.py:23 (a # noqa: E402 on an import
       that no longer triggers E402). Removing the stale directives (uvx ruff check
       --fix) is low-risk and needed before the gate can be green.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
      boundaries:
        - "Removing stale noqa directives changes no runtime behaviour."
      phase_exit:
        - "RUF100 resolved; ruff check returns 0."
  - id: asf-review-useless-import-alias-26
    title: 'POLISH minor: 26 useless-import-alias (PLC0414) redundant re-exports'
    status: done
    scale: minor
    surface: validation
    owner: src
    evidence:
      - uvx ruff check src scripts tests
    notes: >-
      26 PLC0414 (useless import alias) hits come from re-export shims that alias an
      imported name to itself (e.g. src/dashboard/payload.py "as live_skilltree_summary",
      src/skilltree/local_auth.py and local_auth_state.py "as X" re-exports). Harmless
      but contribute to the red ruff gate; drop the redundant aliases or mark them
      intentional under the project lint policy.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
      boundaries:
        - "Re-export shims exist for API stability; only redundant self-aliases are removed."
      phase_exit:
        - "PLC0414 resolved or explicitly justified; ruff check returns 0."
  - id: asf-review-shebang-not-executable-30
    title: 'POLISH minor: 30 EXE001 shebang-not-executable scripts'
    status: done
    scale: minor
    surface: validation
    owner: scripts
    evidence:
      - uvx ruff check scripts
    notes: >-
      30 scripts carry a #!/usr/bin/env python3 shebang but are not chmod +x (EXE001).
      Decide and apply one convention (drop the shebangs if always invoked via uv run
      python, or set the executable bit) so the lint gate is green.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
      boundaries:
        - "Scripts are invoked via uv run python in this repo."
      phase_exit:
        - "EXE001 resolved by one chosen convention; ruff check returns 0."
  - id: asf-review-implicit-string-concat-30
    title: 'POLISH minor: 30 ISC004 implicit-string-concatenation in collection literals'
    status: done
    scale: minor
    surface: validation
    owner: src scripts tests
    evidence:
      - uvx ruff check src scripts tests
    notes: >-
      30 ISC004 hits flag adjacent string literals silently concatenated inside
      collection literals; low-risk but part of the red ruff gate. Add explicit
      concatenation or trailing-comma splits to satisfy the linter without behaviour
      change.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
      boundaries:
        - "Formatting edit is mechanical and behaviour-preserving."
      phase_exit:
        - "ISC004 resolved; ruff check returns 0."
  - id: asf-review-ruff-format-4-files
    title: 'POLISH minor: ruff format --check fails on 4 files'
    status: done
    scale: minor
    surface: validation
    owner: src scripts tests
    evidence:
      - uvx ruff format --check src scripts tests
    notes: >-
      uvx ruff format --check src scripts tests reports 4 files that would be
      reformatted (while 222 already conform), leaving the documented formatting gate
      red. Run uvx ruff format src scripts tests and re-verify; formatting is mechanical
      and behaviour-preserving.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
      boundaries:
        - "Formatting is mechanical and behaviour-preserving."
      phase_exit:
        - "ruff format --check returns 0 across src/scripts/tests."
  - id: asf-review-catalog-catalog-coverage
    title: 'POLISH minor: src/curriculum/catalog.py at 85.71% (main branch untested)'
    status: done
    scale: minor
    surface: validation
    owner: src/curriculum/catalog.py
    evidence:
      - uv run pytest tests/test_curriculum_catalog.py --cov=src --cov-report=term-missing
    notes: >-
      src/curriculum/catalog.py sits at 85.71%: the if __name__ == "__main__" CLI/error
      path is untested. Add a smoke assertion (or mark that branch pragma: no cover) so
      the module does not sit below the 90 floor.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
      boundaries:
        - "Coverage is an exercised-code-path signal."
      phase_exit:
        - "src/curriculum/catalog.py is at or above the 90% floor or its CLI branch is covered."
  - id: asf-review-plw1510-subprocess
    title: 'POLISH minor: 10 PLW1510 subprocess-run-without-check flags (mostly benign)'
    status: done
    scale: minor
    surface: validation
    owner: scripts src
    evidence:
      - uvx ruff check src scripts tests
    notes: >-
      10 subprocess.run calls lack explicit check=True. Review confirmed returncodes ARE
      handled at most call sites (scripts/reset_local_skilltree.py run_checked,
      scripts/z_generate_manuscript_variables.py uses check=True), so these are
      lint-hygiene rather than real command-failure bugs, but they keep the ruff gate
      red. Resolve by making the checks explicit or adding a targeted, justified noqa.
    scholarship_plan:
      lanes:
        - delivery_and_reporting
        - applications_and_scope
      existing_source_keys: []
      planned_source_targets: []
      claim_support_types:
        - delivery
      boundaries:
        - "Subprocess returncodes are already checked at the flagged call sites."
      phase_exit:
        - "PLW1510 resolved via explicit checks or a justified targeted noqa; ruff check returns 0."