This is the execution roadmap for V1.0, with forecast dates from the July 13, 2026 plan. Dates are forecasts, not commitments. ROADMAP.md remains the implementation history and prior milestone record. P1 cuts cannot remove PDF comparison for active sources, Spanish V1 metadata, signing, retention, or deployment.
| Outcome | Owner | Dependencies | Exit evidence |
|---|---|---|---|
| Active source registry human-verified | Verification lead | qualified reviewers, verification CLI | all active sources named/dated; exceptions resolved |
| Verification/evidence schema and tests corrected | Engineering | registry authority decision | required evidence reference, dated robots/terms/fetch-policy decision, shared watcher/publisher eligibility predicate, migration, and derived anti-fabrication tests pass |
| Public health distinguishes quiet/partial/failed/stale | Engineering | run model and schema | status fixtures and staging output |
| Persistent production topology provisioned | Operations + engineering | host/storage decision | runner, encrypted volume, scheduler, backup target, staging promotion receipt |
| V1 safety governance adopted | Governance lead | counsel/community panel | reviewer policy, conflict policy, release authority |
| Architecture/data contracts frozen for RC | Technical lead | ADR and schema review | accepted V1 boundary and migration plan |
Risks: reviewer availability and machine-checked sources proving wrong. Mitigate with federal/high-use sources first and a parallel repair queue.
| Outcome | Owner | Dependencies | Exit evidence |
|---|---|---|---|
| Independent high-impact review and correction chain | Engineering + governance | reviewer policy, schema migration | adversarial E2E tests and tabletop |
| Versioned status/release manifests and atomic publish | Engineering | hosting configuration | reproducible staging release and rollback |
| Active PDFs yield safe passage comparison | Engineering + verification | retained bytes, extractor/manual workflow | every active PDF passes extraction or reproducible manual-comparison rehearsal |
| Public/internal note boundary enforced | Engineering + governance | schema and copy policy | legacy-note audit and publisher negative tests pass |
| Operable service | Service owner | alerts, backup target, runbooks | fault injection and restore within RTO |
| Accessibility/security/legal review | specialist owners | release candidate surfaces | reports with blocker/critical issues closed |
| Spanish V1 metadata reviewed | Language + accessibility leads | stable message catalog | translator/independent review and stale-string fail-closed receipt |
Risks: a quiet real-world period provides few alerts; use clearly labeled synthetic exercises. Never manufacture public change records. The accessible static review bundle and optional delivery helpers are the first cuts if critical-path evidence slips; active-PDF comparison is not cuttable while PDFs remain active.
Freeze and receipt the complete operational baseline by October 16, then run eight consecutive eligible production-like cycles through December 14. Use the holiday contingency and January 4–15 RC window for clean migration, golden-corpus, security, accessibility, signing, deployment, restore, and rollback tests. A pre-baseline rehearsal or a sequence invalidated by a material control change does not count. Reconcile findings with requirements, close or explicitly block every P0, and hold the joint release review by January 29. Release only if every must-pass checklist item has evidence. Otherwise publish a dated hold decision and revised critical path.
- accessible static reviewer bundle and richer PDF/visual comparison beyond the bounded V1 path;
- additional consumer integrations;
- territories or new document classes only with governance/capacity review;
- per-source adaptive outage thresholds from observed data;
- independent nonprofit or fiscal-home governance;
- daily cadence only if user evidence shows weekly latency causes harm and crawling remains responsible.
| Gate | Forecast | Go condition | Hold / kill condition |
|---|---|---|---|
| G1 verified foundation | Sep 25, 2026 | source/evidence schema, persistent runner, and attempt denominator proven | source truth is ambiguous or deployment has no durable evidence store |
| G2b operational baseline | Oct 16, 2026 | all controls listed in the operations bootstrap-eligibility rule are installed, tested, and receipted before the Oct 19 cycle, including dated source-policy decisions, shared watcher/publisher eligibility enforcement, and replacement-or-gap disposition for all six planning-time active-but-unfetchable sources | missing/unstable control, an unresolved active-but-unfetchable source, or material semantic change; rehearsal runs cannot count |
| G3 scope lock | Nov 13, 2026 | all P0 implemented/testable; no unresolved architecture blocker | dual review/correction/public-copy boundary cannot be represented safely |
| G4 bootstrap close | Dec 14, 2026 | eight weekly receipts meet pre-V1 evidence rule | falsely green status, evidence loss, or safety incident invalidates the sequence |
| G5 RC | Jan 15, 2027 | implementation complete; specialist blocker findings closed | Sev-1/2 open or restore/signing/atomic publish fails |
| G6 V1.0 | Jan 29, 2027 | release checklist and operational outcomes pass | source verification incomplete or a safety invariant fails |
Weekly delivery review covers requirement burn-up, verification, operations, risks, and explicit scope cuts. Governance reviews any harm signal immediately. At each gate, the owner records evidence and decision; schedule pressure is not evidence.