Oracle round-2 review caught two real bugs: 1. Production baseUrl bypassed the nginx /api proxy - api.ts defaulted to https://demo.flow-master.ai in prod - Browser would hit cross-origin and CORS-fail - Now: baseUrl='' everywhere; nginx.conf already reverse-proxies /api/* - vite.config.ts proxy still handles dev 2. Refresh button didn't refresh - setMode('live') early-returned when already in live mode - Now: setMode() and refreshLive() share runLiveFetch(); refreshLive ignores the same-mode guard and always re-runs - 4 new vitest regressions in state/store.test.ts cover the contract - Smoke now asserts /api/ea2/work-items is called twice after Refresh Also: - buildScenarios.ts parallelized: cap N=6 candidates, Promise.all per- candidate fetches → live mode now ~3s instead of 30s - CommandBar + LeftRail preview toasts now name the exact endpoint (/api/runtime/transactions/{id}/actions) in the visible text - Landing 'Go live' button rebound to refreshLive() when already live; copy changed to 'Live · refresh' - README: scenario table now renders (added separator row); deploy section points at the real ops PR + the actual overlay path (overlays/demo, not overlays/mc.flow-master.ai); CORS doc clarifies same-origin requirement Constraint: browsers reject cross-origin → same-origin /api/* required Rejected: dev/prod baseUrl divergence | created production bug Confidence: high Scope-risk: narrow Not-tested: production image actually built + served by ops PR (gated by trusted updater + DNS)