Validation evidence
Local checks dated September 30, 2026 (America/Los_Angeles). The final production browser replay used http://127.0.0.1:4187/Program-Atlas/. Generated build/test output and screenshots are ignored by Git. Independent public release verification subsequently passed on October 1, 2026, as recorded below.
Software evidence
| Check | Local result | Scope |
|---|---|---|
| Dependency installation | npm ci succeeds |
Lockfile-pinned Node 24 tooling; no credentials |
| Lint | npm run lint passes |
Source, browser tests and tooling |
| Strict types | npm run typecheck passes |
TypeScript strict/no-unused settings |
| Domain/state checks | npm run test: 22 passing tests in 3 files |
Calendar, graph, proposals, immutable records, persistence |
| Production build | npm run build passes |
Runtime/environment guard, strict types, Vite base and document copy |
| Dependency audit | npm audit --audit-level=high: zero vulnerabilities |
Installed dependency tree at check time |
| Production browser | npm run test:e2e: 14 passing Chromium tests |
Exact primary/recovery, stale/unavailable/invalid state, keyboard and mobile |
| Initial browser inspection | Meaningful heading/content, 19 initial buttons, zero console/page errors | Real production page; desktop screenshot inspected |
| Workflow action references | Four pinned action commits resolved through GitHub read API | Reference existence verified; workflow execution remains a release gate |
Independent expectations
Independently calculated expectations, including the original contract and later regression counterexamples, were asserted in tests.
- Friday Oct 9 is index 4; next workday Monday Oct 12 is index 5. Weekend origin Oct 10 rolls to Oct 12.
- Original readiness: T1 4 + T2 3 + A2 3 + E2 2 = 12 working days, inclusive Oct 20. P1/P2/A1/E1 total slack is 1/2/5/6.
- Delayed P1 is 7 working days; T2 begins Oct 14, ends Oct 16; A2 begins Oct 19; readiness ends Oct 23 at length 15. Converging A2 waits for all required predecessors.
- Scope: 7+3+1+2 = 13 days, Oct 21. Resequencing: max(P2=9, T2=6, A1=2)+3+2 = 14 days, Oct 22. Contingency: 5+3+3+2 = 13 days, Oct 21, fictional $4,800.
- P2 duration 4 gives tied P2/T2 branches ending index 6; both have zero slack.
- T1 duration 10 makes the original partner delay add zero completion days; scope then yields length 16, Oct 26, four days after the original baseline. Consequence text uses the actual custom schedule.
- Add P1 predecessor to T1: P1/T1/T2 are zero slack, but redundant P1→T2 is not a driving critical edge because it is not temporally tight.
Graph tests reject missing, duplicate, self, disallowed, required-removed and cyclic predecessors before scheduling. Decision tests verify snapshot immutability and no unrelated milestone mutation. Storage checks include the getter itself throwing, quota failure, preserved incompatible data, same-revision content conflict and contradictory saved schedule aggregates.
Browser walkthrough coverage
Primary: original Oct 20 → delay preview/cancel (no storage) → delay confirmation Oct 23 → contingency/resequencing previews/cancel → scope confirmation Oct 21 → reviewed escalation → immutable record → nonmutating detached download → compatible refresh. Error/recovery: missing reference and cycle rejected → valid duration edit → one-level Undo; reset preview cancel/confirm/Undo. Recovery: incompatible storage preserved until reset; storage getter denied; write quota failure; same-revision cross-tab writes invalidate preview; missed events caught by actual-value check. Accessibility: skip link, native modal with explicit boundary focus wrap, Escape/focus restoration, status/alerts, equivalent complete list, 390px and 320px layouts without body overflow. Security/document route: CSP/referrer and all seven copied Markdown links checked.
An initial keyboard replay found that native dialog Shift+Tab could leave the intended first/last-button loop. The explicit boundary wrap was repaired and the full production suite passed. Program map horizontal scrolling is confined to its labeled region, with a visible hint and complete List alternative. Visual screenshots were inspected for desktop and mobile; this is not a formal screen-reader audit or cross-browser certification.
Evidence still open
- The software/publication evidence is recorded below; the public profile provides the case study, PRD, walkthrough and demo route.
- Mo’s comprehension and product tradeoff discussion are not observed.
- No human research, actual partner review, customer outcomes, forecast accuracy or commercial results are claimed.
Proposed human evaluation
Use a cause-identification task, a recovery-choice task and an escalation-completeness rubric. Proposed measures: correct identification of upstream impact; recognition of remaining risk; owner/decision/evidence/due completeness. Keep those denominators distinct from completion speed. Software tests cannot establish these human measures; no participants have been recruited or observed.
Independent release review
The primary reviewer checked domain calculations, cause explanations, immutable records and persistence rules, then replayed cancellation, partner delay, recovery comparison, scope confirmation, decision capture and return in a separate production browser. Observed dates were October 20 → 23 → 21, with recovery alternatives October 21/22/21. The 320px page defaulted to the complete List view with no body overflow and no reported browser errors. Private-source and local-link preflight passed across 40 tracked files and seven relative document links before the final release preparation. The tracked-runtime/environment guard is also enforced during builds. Public deployment checks subsequently passed, as recorded below.
Public release verification
The initial public release at 22a91f1de1750a5b11a77b31b4b8ab325269be0c passed GitHub verification and Pages deployment. Local HEAD matched GitHub main, the worktree was clean, and all 11 deployed files matched the local production build and GitHub deployment artifact byte for byte. CSP and no-referrer metadata were present. The live page loaded its original October 20 readiness and primary controls with no reported page errors. These are point-in-time observations from 2026-10-01, not uptime, human research or commercial-outcome claims. Try Program Atlas.