Production edge routing broken for 3+ hours — stuck wildcard 404 cache entry, confirmed via reproducible evidence
Severity: Active production downtime, Edge Network
Every dynamic/auth-gated route on our production domain (www.firstprincipledlife.com) returns a 404 with X-Matched-Path: /404, while static marketing pages continue to work normally.
Evidence this is a stuck edge cache, not a deployment issue:
- A guaranteed-never-before-requested URL (a random nonce path) returns
X-Vercel-Cache: HITwith anAgeheader around 185,000+ seconds (~51 hours). A URL that’s never been requested cannot legitimately be a cache hit under normal per-URL caching. - Every affected route shows an identical Age value within 1-2 seconds of each other, indicating one single, shared, stale 404 response is being served for the entire unmatched-path space.
- We confirmed this is not a build or deployment problem: three separate deployments (including one empty commit specifically to trigger a fresh build) all produced identical build output with correct route manifests (
ƒ /account,ƒ /dashboard/builder, etc.), and runtime error logs show zero invocations for these requests, meaning they’re being resolved at the edge before ever reaching the application. - We promoted multiple confirmed-healthy deployments to production, including the pre-incident deployment that previously held the live alias successfully, and the identical stale 404 behavior persists regardless of which deployment is aliased.
Current production deployment: dpl_74GdSpjAkwJmZLQFaoFQgxcnsNuy (commit 175158c), confirmed healthy build, correctly holds the custom domain alias, still affected.
We’re on the Hobby plan and don’t have access to a self-service cache purge. This needs a manual edge cache purge on Vercel’s side. Happy to provide any additional logs or deployment details needed.