Project: yogibo-dashboard (Project ID: prj_Y2OJOZEpLasOlUZXwfPgl7ADYcV7). Production domain: yogibo-dashboard-eight.vercel.app. Team: masatonishimura121-lab's projects (Hobby plan). Summary: Since merging a PR (commit 1e77665) through a later PR (commit dbf0ea3717d82fb70b86b3b4181116051366d9ce), and now including an empty commit (5d4fc7390f322f75778f3673107c468b40af3377) pushed specifically to force a fresh build, every subsequent Production deployment reports Ready/success, but the built static asset assets/index-*.js served to visitors is byte-for-byte identical to the build from an earlier commit (c73a247e2d247012019658f3036146043df1c659), which was the last deployment that actually reflected new source changes. What we verified: (1) Source is correct - confirmed via git show and via directly downloading the GitHub codeload tarball for the exact commit SHA, both show the intended code changes (a new function called weeklyStdPrmSales, 5 occurrences in src/App.jsx). (2) Build logs show a genuine fresh build each time - "Skipping build cache, deployment was triggered without cache.", correct commit cloned, "Cloning completed", vite v5.4.21 building for production, 41 modules transformed - all normal, no errors, about 15-19s duration. (3) Not a CDN/edge-cache artifact - checked the build's own unique, never-before-requested deployment URL directly via fetch with cache:no-store. It serves the same old asset hash and byte length as every prior deployment, and does NOT contain the new code that is definitely present in the committed source. (4) Project settings are default/unmodified - Framework preset Vite, no Build/Output/Install Command overrides, Root Directory ./, no Ignored Build Step script, no .vercel/output or dist committed to the repo, package.json build script is plain vite build. (5) Tried Promote to Production (reported "This Deployment is already in Production") and Redeploy (new deployment ID, same stale output) - no change. (6) Just tried a full CDN cache purge (CDN + ISR + Image Cache, "All content") from the CDN > Caches dashboard - confirmed "purged just now" - and the production domain STILL serves the exact same stale asset afterward. This rules out edge/CDN caching entirely, since a full purge plus a brand-new deployment URL both still serve the old bytes. What this rules out: expired/invalid GitHub integration token (would cause a clone failure, not a silent stale build); browser or CDN edge caching (reproduced on a brand-new never-before-requested deployment URL, and survives a full CDN purge); misconfigured project settings (all defaults, verified in Build and Deployment settings); committed build artifacts shadowing the real build (no dist or .vercel/output in the repo); stale GitHub tarball/codeload delivery (directly downloaded the tarball for the latest commit and confirmed it has the correct, up-to-date source). What we suspect: a corrupted or stuck internal build output/artifact store for this specific project that is producing or reusing stale output even though the build step itself appears to execute normally against the correct, fresh source tree, and even survives a full CDN purge. Has anyone seen this exact symptom before, and is there a way to force a full reset of a project's build artifact store (short of deleting and recreating the whole project)? Happy to provide any additional diagnostics. Deployment IDs for reference (all Ready/Production, all serving the stale output): 6435779372, 6439171376. Last known-good deployment was for commit c73a247. Thanks for any pointers.