Forge Platform

Documentation operating cadence (Hydration v2)

How the governed documentation intelligence layer runs day to day. Every cadence maps to concrete script invocations from the forge-platform root; none of them auto-promote — human reviewer decision manifests gate all…

Updated

Per build (every site rebuild / deploy)

Step Command
Emit source map automatic — every generator writes website/docs_source_map.json
Docs-quality artifacts python3 scripts/site_docs_artifacts.py --site-root <site>/website --site-name <site> (also run by deploy-websites.sh)
Scorecard + delta python3 scripts/docs_scorecard.py (prints build-to-build delta; regressions, stale claims, and link rot append to the backlog)
Quality gates python3 scripts/check_quality_gates.py (G1–G8; check_foundation.py runs the registry-scope gates on every foundation check)

Per change signal (continuous ingest)

Step Command
Canonical repo changes python3 scripts/adapters/adapter_markdown_repo.py
Design doc status transitions python3 scripts/adapters/adapter_design_doc.py
External signals (T3/T4) python3 scripts/adapters/adapter_external_note.py --source-id … --observed … --detail …
Search misses (log import) python3 scripts/adapters/adapter_search_log.py --log … --source-id …
Plan responses python3 scripts/hydration_planner.py (11-decision enum; conflicts block with a report)

Unregistered sources quarantine automatically; quarantined events never plan.

Weekly content council (~30 min)

  1. python3 scripts/hydration_cycle.py — full PDCA pass; read the cycle summary.
  2. Review open plans in docs-governance/plans/ — approve, reject, or hold each.
  3. Review conflict reports in docs-governance/conflicts/ — pick a winning source, update claims, unblock.
  4. Triage docs-governance/rehydration_backlog.jsonl open items; close or schedule.
  5. Record decisions as reviewer decision manifests where promotion follows.

Per release cycle

  1. Rerun claim extraction over changed strategic pages: python3 scripts/extract_claims.py …
  2. python3 scripts/detect_drift.py — seed/page/claim drift after the release.
  3. Promote approved packs: python3 scripts/promote_hydration_pack.py --dry-run → review output → run without --dry-run.
  4. Rebuild + deploy affected sites (./deploy-websites.sh from the workspace root prints the scorecard delta in its summary).

Quarterly journey and prune audit

  1. python3 scripts/agent_retrieval_eval.py — retrieval misses per persona.
  2. Review docs-governance/plans/archive-proposals/ (generated by hydration_cycle.py for pages unreviewed > 365 days) — archive, merge, or rehydrate via manifests.
  3. Refresh persona_journey_map.yaml stages/journeys against real usage.
  4. Re-run python3 scripts/check_page_contract.py --pilot --strict and expand the contract-enforced page set.

Trust boundary (unchanged)

Scripts observe, plan, and report. Humans decide truth: claim status transitions, promotions, archives, and conflict resolutions all require a recorded reviewer decision.