Impact Gate
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote
Impact Gate answers a question every team with a public API hits: "we changed the API - which consumers break, and how do we fix them?" It connects API changes to the services, functions and consumers they affect, then proves breakage instead of guessing it. Where static indexing leaves gaps, a post-index AI layer - served by Meta's Muse Spark through the governed model registry - investigates missing call relationships, with source evidence gating every proposal. In this demo, a real commit on ig-lab-catalog-api renames priceCents to unitPriceCents. Impact Gate runs the TypeScript consumer's own test against immutable base and head fixtures: the base passes, the changed API fails, and the regression is recorded with a SHA-256 evidence digest. From that evidence it produces a source-anchored migration (product.priceCents -> product.unitPriceCents in src/order-service.ts) and verifies it: the original consumer fails the head fixture, the patched consumer passes. Every execution is durable - state, revision and evidence are recorded, and resume/retry/cancel are revision-checked so stale executions are never accepted. Verification is fixture-based: no arbitrary repository shell is executed, and fixture oracle mode is labeled in every artifact.