Watched repositories

Public repositories we audit and re-audit, each with the evidence label that produced its number. Audited the same way as anything you paste on the home page: a clean clone, the documented path, the same gates.

Shape

Repositories published by Vercel. They are here because auditing the platform we are built on is the honest test of the instrument: if it only ever finds fault elsewhere, it is not measuring anything. The FAILs stay on the board. These are not rankings and nothing here sorts. We tried ordering them by time to first value and our own data killed it: three runs of one commit timed 87s, 49s and 50s — a wider spread than the score we had rejected for moving — because the planner proposes a different journey each time. (Those three pre-date the engine that fixed how we measure the span.) The times below are readings, not places. Compare within one shape: a CLI reaches first value differently from an SDK.

6 repositories shown, 3 with a timed path

RepositoryFirst valueGatesScoreEvidenceRun

vercel/edge-runtimeRuntimeA runtime, so its quickstart is a build rather than a server.the documented path did not reach a verified outcome4/1/234FAILObserved0f5504b

vercel/msLibraryA single-purpose utility; the shortest documented path we audit.5.7s6/0/173PASS WITH DEBTObserved4ff48ce

vercel/react-tweetLibraryA monorepo whose root README is a symlink — the case that found a bug in our own extractor.22s4/0/3UNVERIFIEDnot scored074901d

vercel/serveCLIA CLI, so time to first value is a command rather than an import.we could not finish the run4/1/2UNVERIFIEDnot scored37b3a74

vercel/swrLibraryThe stability reference: three runs at one commit scored 75, 73 and 71.12s6/0/169PASSObserved01cb488

vercel/workflowSDKDurable workflows. Its build exits 1 from a clean clone.we could not finish the run4/0/3UNVERIFIEDnot scored7740388

Gates read passed / fired / unverified over the seven this engine decides. A re-run needs a signed-in account and counts against your daily audits; an unchanged commit returns the cached report instead of running again. How the score is produced.