6.5 KiB
6.5 KiB
Bounded acceptance-gap audit
Baseline: develop 69dd6922f6, official Stage run1352. Audit is bounded to existing specs, runtime scripts, corporate service/controller, UI/services and focused/full-suite evidence; not a claim of enterprise acceptance.
| Criterion | Code/tests | Deployment/behavior | Verdict / next dependency |
|---|---|---|---|
| Real persisted source provenance | corporate research/source tables, validated HTTPS metadata, manual ingest stores fetched bytes and SHA256 | Baseline both Stage images independently read as 69dd6922, restart0; source inventory pilot exists, not autonomous business research |
partial: discovery/synthesis and provider-backed intelligence missing |
| Monthly research -> persisted roadmap/goals | service roadmap requires persisted research with sources; manual CLI manifest | Existing integration tests; no demonstrated automatic strategy synthesis | partial |
| Daily NEW article plan separate from refresh | UI promises daily new; roadmap counts only; fanout creates source inventories | No dated article planning or article executor | missing |
| Needs-based refresh/catalog/wiki/engineering/devops | dormant deterministic structural department audits and GET-only plan_day | Not clinical enrichment, code execution, or publication; plan is not persisted task fanout | partial |
| Atomic claim/execution/evidence/independent QA | service lease fencing, idempotency, TEST, distinct review identity, exact hash verification | Existing real disposable DB tests and source-inventory pilot | partial: only manual bounded inventory/audits, no business action adapters |
| Visible instruction/action/result/evidence/QA | Task detail shows description and event summaries only; evidence/reviews/effects omitted; TEST absent in frontend status type/filter | API detail includes these relations; UI cannot inspect exact receipts | missing next slice: read-only persisted execution dossier + TEST status |
| PMO outcomes/report | persisted backend report snapshots, explicit truncation; manual GET-only plan | No integrated daily plan/report consumer or report UI | partial |
| Market pagination/filter/search | pagination/category UI + backend; no search | Not authenticated/mobile accepted in this audit yet | partial |
| Roadmap pagination/filter/search | pagination UI, month backend only; no search | Not authenticated/mobile accepted yet | partial |
| Tasks pagination/filter/search | server pagination, normalized title/code, dept/status/priority | Missing TEST option; execution dossier absent | partial |
| SEO under Hermes/mobile | existing tracking page and API, independent telemetry limits | Requires distinct authenticated mobile review, not proven by compile | partial |
| No fake fallback | APIs reject unavailable collections; UI explicit errors; metrics unavailable != zero | Preserve; no demo business data added | code present, scoped verification required |
| Test gates | baseline full Jest: 2 failures incl wiki expectation and absent isolated DB URLs | Repair full gate with fresh private PostgreSQL harness, no live DB reuse | partial until repaired run |
| Public author attribution | credentialless genuine authors restored; marker denied all auth paths | Baseline API repair confirmed by parent; public attribution UI still unresolved | partial; inspect real byline/metadata, never invent credentials/clinical expertise |
| Autonomous operation | runtime explicitly MANUAL_DORMANT, schedules false | Must remain cron-disconnected until whole acceptance | deliberately not enabled |
2026-10-30 next slice (local, not deployed)
- Persisted monthly planning/API prior Stage acceptance is recorded by release owner (engineering fixtures, not clinical roadmap approval).
scripts/corporate_plan_inventory.pynow supports explicitprepare roadmap month task,execute child, separate-keyqa child evidence, andreport childauthenticated readback. It copies exact instruction/PLANNED source lineage into a linked RESEARCH prerequisite, verifies selected existing source bytes and hashes, and never submits evidence for the original BLOG_NEW/needs action.- Optional claim
taskIdselects only the exact eligible child, preserves future-plan fences and historical no-selector idempotency hashes; no fallback claim. - Real disposable HTTP/PostgreSQL scenario exercises creation/replay, exact execution, self-review denial, independent rejection, corrective execution, independent pass and PMO readback. Original task remains BLOCKED after rejection (TODO on happy path), with no article evidence/review. This is inventory, not synthesis/publication/clinical QA.
- Independent review found rejection recovery blocked by TODO-only parent guard; fixed for existing linked child snapshot only, with RED/GREEN HTTP regression. New prepare still requires TODO. PMO records observed TODO/BLOCKED, not an invented completion.
- Local gates: backend 105 suites/950 tests pass (2 existing suites/3 tests skipped), Python 27 tests pass, backend build pass; admin full tests/build pass. Exact logs/source hashes:
/opt/data/content-maintenance/corporate-action-evidence-20261030/manifest.json. - Remaining: action-specific writers/provider-backed real research, approved business source store/roadmap, Stage acceptance of this slice, integrated monthly/report UI, complete mobile acceptance, controlled activation approval. No schema/approval/publisher/cron/Hermes config/Production changes; MANUAL_DORMANT remains.
Slice selection
First unblock truthful inspection of persisted execution: without exact tool/result/hash and independent QA visible, a higher-level planner would create more uninspectable tasks. Implement read-only dossier, real TEST status/filter, empty and bounded-history disclosures. No schema/business writes, new scheduler, approval modules, Hermes config/restarts, Production or main changes.
Remaining sequence
- Accept this dossier and corrected isolated DB gate on exact-SHA Stage.
- Research-backed dated daily-new article planning, separate needs-based refresh/catalog/wiki/engineering task lineage, API idempotent persistence (review schema separately if needed).
- Implement narrowly allowed action adapters with immutable instruction/action/result/evidence and independent QA; no model-output-is-DONE shortcuts.
- Persist PMO outcomes and downstream daily plan; add market/roadmap search/filter/report UI, finish authenticated mobile acceptance and real public author checks.
- Only after complete independent acceptance consider explicit activation of the currently dormant/manual runtime; no cron changes in this program slice.