ForgePlan

A manufacturing control environment. Gantt, capacity, and recovery, with the consequence chain shown.

1. Dashboard, schedule, orders, capacity, materials, routings, bottlenecks, analytics, and simulator views over one schedule result.
2. Guided demo: a calculated walkthrough of an eight-hour CNC failure.
3. Factory Recovery: an outage scenario with lexicographic ranking of recovery options and a consequence chain.
4. Policy comparison: eight dispatch policies compared, with a benchmark guard.
5. CSV export of schedule results.

• Next.js 16, React 19, TypeScript, Zod, Recharts, react-hook-form. 15 routes.
• Deterministic finite-capacity scheduler with invariant auditing. Greedy heuristic, no optimality proof claimed.
• Fixture: 1 facility, 7 work centers, 14 machines, 30 products, 48 production orders, 4 routing structures (README).
• Persistence in localStorage (forgeplan-state, schema v2, Zod-validated, legacy key migrated). Invalid state restores the fixture with a notice.
• CSP frame-ancestors 'none' with X-Frame-Options DENY. This app cannot be iframed. Link-out only.

• All views consume the same ScheduleResult: Gantt, capacity, exceptions, case study, and KPIs cannot drift apart.
• No claim of global optimality, production deployment, or measured customer savings (README).
• Invariant auditing runs over the schedule, so a broken schedule fails loudly instead of rendering nicely.

• Install outcome is UNVERIFIED. A prior npm ci failure cited missing @emnapi packages; the lockfile now contains @emnapi 1.10.0 nested entries, but no clean install was run in recon. Do not claim the lockfile is fixed until npm ci is run.
• Persistence is origin-wide localStorage, not tab-isolated. Two tabs share one demo state.
• Four stacked PRs (M6 design system and accessibility sweep, M7 digital twin, M8 disruption laboratory, M9 schedule intelligence) are unmerged, with M6 as the base of the chain.

1. CNC-02 is removed for 8 working hours. The flagship experiment table reports a baseline and a recommended failure-EDD recovery with on-time percent and tardiness hours (doc-claimed, README).
2. Run Factory Recovery from the landing page. Compare ranked recovery options.
3. Follow the consequence chain through orders and capacity, then reset to the fixture.
4. Compare dispatch policies in the policy comparison view.

• 9 test files and 5 end-to-end spec files in the repo. Verification doc claims 174 unit tests and 16 of 16 end-to-end tests at an earlier commit, with a working-tree-changes flag. Neither was re-run in VW-100 recon. [OBSERVED files; DOC-CLAIMED results, UNVERIFIED]
• Benchmarks are recorded as measured, not projected, with hardware and Node version stated. Nothing claims a policy is optimal. [OBSERVED, BENCHMARKS.md]
• Synthetic-data disclaimer is present, and the landing page states no sign-in and the fixture size. [OBSERVED]

OBSERVED means read from repository evidence at the stated commit. DOC-CLAIMED means a repository document states it but it was not re-run during reconnaissance. UNVERIFIED means no supporting evidence was obtained. See the engineering inventory.

• No verified public demo URL. README states: prepared for Vercel, but no live URL has been published from this workspace.
• No CI workflow exists. No license file exists; usage terms are pending an owner decision.
• Greedy heuristic with no optimality proof. Non-preemptive operations fitting one shift. Not full MRP.
• UI quality, responsiveness, and dead controls were not observed (no live URL, no browser run in recon).
• Screenshot files referenced in docs were not verified (binary fetch limitation, not evidence of absence).

Origin-wide localStorage persistence with reset to the fixture. No expiry, no per-visitor isolation, no time-box. A session-scoped, tab-isolated demo mode is a change request, not shipped work. 60-minute exploration depth is UNVERIFIED.

Current portfolio status: Demo unavailable in this harness: no deployment verified, demoUrl null, allowlist empty.