Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,8 @@ it is doing and what to do if it stops; a run that refuses with a reason is work
intended. The live path needs a running Docker daemon because SkillTree owns quizzes,
completion, points, and ranking. The offline path needs neither Docker nor a network.

Help, readiness, and artifact-only exits do not initialize passwords. Live launchers share the password owner; bootstrap may skip seeding only with `--skip-seed-if-live-content-ready` after semantic content verification, not matching counts. Offline artifacts and controlled regressions do not prove a live official SkillTree display.

## Read it without installing anything

Generated outputs are committed, so the artifacts are browsable straight from the
Expand Down
4 changes: 4 additions & 0 deletions docs/gates.md
Original file line number Diff line number Diff line change
Expand Up @@ -90,6 +90,10 @@ between them fails instead of silently changing the resolution.
The live half of the gate set (Docker, a real SkillTree, browser proofs, captured
screenshots) stays local and explicit. No arm here simulates it.

Strict live proof requires semantic content parity and actual official SDK project/subject/quiz views with visible same-origin frames, ready/accessibility evidence, and matching project/instance/export digest. Capture consumes a successful v2 `--proof-path` and matching PNG associations. Pending/fallback/unchecked views and controlled loopback fixtures are not live proof.

The 2026-11-13 full non-browser census failed: its native source binding changed, seven tests failed, and coverage was below the unchanged 90% floor. The separate seven-node continuation and appended coverage database are narrower observations, not a fresh immutable whole-source certificate. The complete per-module census still records under-floor modules with no approved exceptions. Script smoke tests now run real generators in a disposable copy of public source inputs instead of rewriting the audited checkout's tracked output; final qualification must still verify the native unchanged-source receipt. Source-bound machine reports remain authoritative. Keep science K4, the pinned GNN origin, real stack/browser, and roadmap-owner holds separate; generated `tasks.yaml`/TODO/readiness projections are not hand-updated.

## Measured numbers

Per-arm wall times and the test/coverage counts are generated, not written by hand:
Expand Down
15 changes: 6 additions & 9 deletions docs/learner_dashboard.md
Original file line number Diff line number Diff line change
Expand Up @@ -95,14 +95,9 @@ uv run python scripts/validate_full_app_stack.py --keep-running
Use this when reporting that the app is live. It starts or reuses the Docker Compose
SkillTree stack, regenerates project artifacts, bootstraps and seeds local SkillTree auth,
reconciles missing live prerequisite edges, starts the dashboard, requires `/healthz`,
requires live admin count parity, requires live dependency parity, checks
figure-registry/dashboard chart-contract parity, proves the raw SkillTree `FORM` login
through `chrome-devtools-axi`, and runs the Chrome console gate. Docker daemon
availability is a hard precondition: if Docker or Colima is not running, this command
fails rather than silently downgrading to artifact-only mode. `--app-ready` is the faster
local usability proof and is explicitly not a full dependency-parity proof.
`--keep-running` leaves the dashboard process open for manual inspection at
`http://127.0.0.1:8765`.
requires semantic admin content parity and complete live dependency parity, checks figure-registry/dashboard chart contracts, and proves real FORM login plus the official SDK's project, subject, and quiz views. Visible matching-origin frames, ready callback/accessibility snapshots, and matching project/instance/export digest are required; pending, fallback, unchecked, or count-only states are not green. `chrome-devtools-axi` and a real Docker/SkillTree stack are prerequisites, not substituted boundaries.

Successful v2 browser/full-stack proof supplies `--proof-path` to screenshot capture. Use explicit task-owned screenshot output with matching PNG associations. `--keep-running` retains only a successfully admitted dashboard server; failure must settle it. `--app-ready` remains a narrower usability scope, not full dependency parity.

Artifact-only mode:

Expand Down Expand Up @@ -163,7 +158,7 @@ lesson.
are absolute, credential-free HTTP(S) URLs; unsafe schemes remain plain text.
- SkillTree deployment matrix: exposes only browser-safe deployment mode status for local
FORM bootstrap, ignored local state, token-backed admin seeding, and externally operated
SkillTree. Admin secrets, client secrets, and bearer tokens remain server-side.
SkillTree. Root credentials and trusted-client secrets remain server-side. The temporary learner bearer token reaches official browser-client memory, never persistent browser storage; it is not accurate to call all bearer tokens server-only.
- Official display host: imports `@skilltree/skills-client-js` and configures
`SkillsConfiguration` with `serviceUrl`, `projectId=activeInference`, and an absolute
dashboard URL for the local `/api/skilltree/token` authenticator when bootstrap is
Expand All @@ -176,6 +171,8 @@ lesson.
proof that the live SkillTree process is currently running or accepting browser
passwords.

Public artifact/status projections remove quiz answer options and correctness keys recursively; the raw validated export and semantic digest remain server-side authorities. `/api/status.project.project` supplies the nested public identity used by live validation. No public projection or fixture response is a scientific, security-review, or live-service qualification by itself.

## Accessibility

The shell targets WCAG 2.1 Level A keyboard and structure basics: a visually-hidden
Expand Down
2 changes: 2 additions & 0 deletions docs/output_inventory.md
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,8 @@

Paths in the table are relative to the project root (`Active_Skillference/`).

The complete registered figure population includes captured artifacts; it is not the deterministic rendered subset. Registry-owned counts and generated provenance must describe the right population without inventing a fixed total.

## File inventory

| File | Role | Generated by | Step |
Expand Down
2 changes: 2 additions & 0 deletions docs/pilot/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,8 @@ The contract requires consent and data minimization, prohibits direct identifier
analysis export, names the unit of analysis and missing-data policy, and keeps SkillTree as
the system of record for progress and quiz state.

Four nonempty linked populations establish schema readiness only. They do not establish recruitment, participant data, demographics, collection completion, or educational efficacy.

## Protocol sequence

1. Pre-register recruitment, inclusion/exclusion, outcomes, observation windows, stopping
Expand Down
25 changes: 6 additions & 19 deletions docs/publication_readiness.md
Original file line number Diff line number Diff line change
Expand Up @@ -29,11 +29,7 @@ roadmap tasks are complete.
- License, third-party notice, citation metadata, DOI, and public-repository boundaries
are signposted in `LICENSE`, `THIRD_PARTY_NOTICES.md`, `docs/ip_license_status.md`,
`CITATION.cff`, `.zenodo.json`, `codemeta.json`, and `manuscript/config.yaml`.
- Renderer-facing repository and release-date metadata remain blank or omitted until a real
published release exists, so rendered PDFs, CFF, and CodeMeta do not advertise an
unavailable release endpoint. Probed on 2026-08-09,
`https://github.com/ActiveInferenceInstitute/Active_Skillference` exists and is public but
holds zero refs, so the target is reachable and still unpublished.
- The 2026-08-09 observation found a reachable public target with zero refs; that is historical evidence, not today's remote state. Later v1.0.0/DOI records below remain authoritative for their own identifiers. Current metadata must agree with actual records and a newly admitted observation, never a guessed blanket unpublished state.

## Release-Allowed Claims

Expand All @@ -43,12 +39,7 @@ roadmap tasks are complete.
- SkillTree is cited and discussed as an NSA-origin open-source training platform, with explicit non-endorsement and deployment-boundary language.
- Project code/content metadata currently declares MIT, while third-party materials are
either inventoried as vendored notices or cited/linked without relicensing.
- The owner-confirmed public target is
`ActiveInferenceInstitute/Active_Skillference`, and that target now exists as an empty
public repository; the claim becomes release-allowed only after the repository carries
published content and a real release, and the release inventory is refreshed.
- Platform-facing source-repository links and release dates are not release-allowed while
that target holds no published content.
- Public repository/release claims must match the actual selected release record and newly admitted remote observation. A reachable target, an empty target, advertised refs, a failed probe, and an unrefreshed report are different states. Do not substitute historical reachability for current publication or infer current emptiness from a failed command.

## Claims To Avoid Until Future Evidence

Expand All @@ -69,11 +60,8 @@ roadmap tasks are complete.

## Current Blockers

- DOI/Zenodo release is blocked until an actual release tag and deposit exist.
- Public repository release is blocked until the confirmed target
`ActiveInferenceInstitute/Active_Skillference` holds published content and a real release
tag, and repository metadata is refreshed. The target's existence cleared the
reachability question only; the publication question is still open.
- Subsequent DOI/Zenodo reservation, streamed upload, and publication each require explicit authorization and actual receipts. Existing v1.0.0 identifiers remain preserved; this work authorizes no new deposit, tag, upload, or release.
- A current repository/readiness projection requires its owning generator and an admitted fresh probe. AS-O14's first-page/title identity and product-selection hold, plus legal/date/DOI decisions, remain owner obligations.
- Learner-outcome claims are blocked until a real study produces evidence.
- Live deployment claims are blocked until Docker SkillTree, dashboard handoff, privacy,
security, trademark, and access-control reviews are complete.
Expand Down Expand Up @@ -101,11 +89,10 @@ uv run python scripts/validate_web_math.py
uv run python scripts/validate_release_readiness_inventory.py
```

Also re-check the public-target blocker. An empty result means the repository exists but
nothing is published there, which keeps the blocker in force:
An explicitly admitted future remote check may inspect advertised refs; empty success is not the same as failed or unrefreshed evidence. No new remote observation is claimed here:

```bash
git ls-remote --heads https://github.com/ActiveInferenceInstitute/Active_Skillference.git HEAD
git ls-remote --refs https://github.com/ActiveInferenceInstitute/Active_Skillference.git
```

For a DOI-backed release or deployment publication rather than a static public-review
Expand Down
2 changes: 2 additions & 0 deletions docs/quickstart.md
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,8 @@ screen-reader affordances. Use `./run.sh --offline` when Docker is unavailable a
`./run.sh --check` to test readiness without starting anything. The step-by-step commands
below are what `run.sh` runs, for when you want to drive them individually.

Help/readiness/artifact-only paths never initialize passwords. Live launchers use the shared password owner. `--skip-seed-if-live-content-ready` admits actual semantic content readiness; the removed count-only shortcut is not sufficient proof. A successful artifact build is not a running official SkillTree stack.


Get Active Skillference running in 5 minutes: validate the curriculum DAG, run the
kernels that produce every claim, export a provenance-gated SkillTree project, and run
Expand Down
10 changes: 3 additions & 7 deletions docs/rendering_pipeline.md
Original file line number Diff line number Diff line change
Expand Up @@ -13,13 +13,9 @@ the manuscript's own configuration, preamble, and bibliography. It needs `pandoc
`pandoc-crossref`, `xelatex` and `bibtex` on `PATH` and nothing from any other
repository; the arm is `./gates.sh render` ([`gates.md`](gates.md)).

After finalizing the PDF, the renderer writes the tracked
`output/pdf/Active_Skillference_combined.receipt.json`: schema
`active-skillference-pdf/v1`, repository-relative output path, exact byte count,
SHA-256, and the checkout's Git `HEAD` source revision. Commit source changes before
rendering so that revision identifies the source used; commit the receipt alongside
the PDF. The render arm finishes with `validate_rendered_freshness.py`, which rejects
stale hydrated sections, graph counts, or publication-bundle roadmap text.
After finalizing the PDF, the renderer writes an `active-skillference-pdf/v2` receipt binding actual configuration, preamble, bibliography, ordered hydrated sections, literal TeX assets, and finalized output bytes. HEAD alone is not source identity. Legacy receipts require a new admitted render, never relabeling or restamping. Unknown prose placeholders fail while literal code examples survive.

The configured atlas is physical page two, before the table of contents; its placement has a real-render regression. Input/content custody, retained-publication identity, and parent visual inspection remain separate requirements. The freshness validator still rejects stale sections, graph counts, and roadmap text.

## The flow

Expand Down
32 changes: 12 additions & 20 deletions run.sh
Original file line number Diff line number Diff line change
Expand Up @@ -22,22 +22,6 @@ DASHBOARD_URL="http://${DASHBOARD_HOST}:${DASHBOARD_PORT}"
SKILLTREE_WAIT_ATTEMPTS="${SKILLTREE_WAIT_ATTEMPTS:-240}"
ACTIVE_SKILLFERENCE_LOCAL_STATE_DIR="${ACTIVE_SKILLFERENCE_LOCAL_STATE_DIR:-.active_skillference_local}"
ACTIVE_SKILLFERENCE_LEARNER_ID="${ACTIVE_SKILLFERENCE_LEARNER_ID:-local-learner}"
# Compose interpolates this; it must match the value the stack was created
# with, so reuse the stored one and generate it only on first run. This mirrors
# src/integration/full_app_stack.py::ensure_postgres_password.
_pg_file="$ACTIVE_SKILLFERENCE_LOCAL_STATE_DIR/postgres_password"
if [ -z "${ACTIVE_SKILLFERENCE_POSTGRES_PASSWORD:-}" ]; then
if [ -s "$_pg_file" ]; then
ACTIVE_SKILLFERENCE_POSTGRES_PASSWORD="$(tr -d '\n' < "$_pg_file")"
else
mkdir -p "$ACTIVE_SKILLFERENCE_LOCAL_STATE_DIR"
chmod 700 "$ACTIVE_SKILLFERENCE_LOCAL_STATE_DIR" 2>/dev/null || true
ACTIVE_SKILLFERENCE_POSTGRES_PASSWORD="$(LC_ALL=C tr -dc 'A-Za-z0-9' < /dev/urandom | head -c 40)"
printf '%s\n' "$ACTIVE_SKILLFERENCE_POSTGRES_PASSWORD" > "$_pg_file"
chmod 600 "$_pg_file" 2>/dev/null || true
fi
fi
export ACTIVE_SKILLFERENCE_POSTGRES_PASSWORD

MODE="full"
for arg in "$@"; do
Expand Down Expand Up @@ -121,6 +105,14 @@ if [ "$MODE" = "offline" ]; then
--host "$DASHBOARD_HOST" --port "$DASHBOARD_PORT"
fi

# Live launchers share one private password owner; help/check/offline never enter it.
if [ -z "${ACTIVE_SKILLFERENCE_POSTGRES_PASSWORD:-}" ]; then
ACTIVE_SKILLFERENCE_POSTGRES_PASSWORD="$(uv run python -c \
'import sys; from src.integration.full_app_stack import ensure_postgres_password; print(ensure_postgres_password(sys.argv[1]))' \
"$ACTIVE_SKILLFERENCE_LOCAL_STATE_DIR")"
fi
export ACTIVE_SKILLFERENCE_POSTGRES_PASSWORD

# ------------------------------------------------------------- live SkillTree
# Run a long command with a live elapsed-time heartbeat, so a slow step is
# visibly slow rather than indistinguishable from a hang.
Expand Down Expand Up @@ -172,9 +164,9 @@ curl -fsS "$SKILLTREE_URL/public/status" >/dev/null 2>&1 || fail \
"Or run without it: ./run.sh --offline"

step "Seeding the curriculum into SkillTree"
note "This is the slow step, and it is idempotent: if SkillTree already reports"
note "the exported subject and skill counts, seeding is skipped entirely and"
note "this finishes in seconds."
note "Seeding is skipped only when the actual project, subjects, skills, quizzes"
note "and answer keys match the export. Missing resources can be created;"
note "changed or extra content is refused rather than silently overwritten."
note "A first seed writes every subject, skill, and quiz over REST. Budget"
note "10-40 minutes on Apple/ARM Docker emulation; it is working, not stuck."
note "Prerequisite edges are NOT seeded here. They are far slower and the"
Expand All @@ -183,7 +175,7 @@ if ! run_with_heartbeat "seeding" uv run python scripts/bootstrap_local_skilltre
--base-url "$SKILLTREE_URL" \
--state-dir "$ACTIVE_SKILLFERENCE_LOCAL_STATE_DIR" \
--learner-id "$ACTIVE_SKILLFERENCE_LEARNER_ID" \
--skip-seed-if-live-counts-ready \
--skip-seed-if-live-content-ready \
--skip-dependencies; then
fail "Seeding did not complete." \
"The output above ends with the failing call." \
Expand Down
Loading
Loading