Project: bump-platform-server (ad744a78-dcb6-442a-b908-abf40241d5b4)

Environment: production (cd4e4abf-605b-43f1-b4ba-88ec43680c78)

Affected service: platform-server (79846b58-9968-4fda-b646-9abf2d7240fd)

Every deploy attempt to this one service has failed since 2026-08-21, and is still failing today (2026-08-23). The build log shows only:

using build driver railpack-v0.37.0

railpack prepare exited with an error <- [err] line, no further detail

scheduling build on Metal builder "builder-zvijhr"

scheduling build on Metal builder "builder-zvijhr" (sometimes also "builder-whvjyg" on an earlier retry)

...then the deployment is marked FAILED a few seconds later. No compiler output, no dependency-install log — the build never reaches the actual build steps.

What we've ruled out:

Not the code — deploying the exact unmodified commit already live in platform-server-prod fails identically.

Not git/worktree structure — a plain directory with no .git at all fails identically.

Not the upload path — reproduced via railway up from a worktree, railway up from a plain directory, and the Railway MCP server's own deploy tool. All three fail the same way.

Not usage/billing limits — well under any limit, no soft/hard limit set.

Not a broad platform incident — status.railway.com showed fully operational throughout; our sibling service platform-server-prod and an unrelated project in the same account both deployed successfully during this window.

Not a config issue — builder is RAILPACK, rootDirectory is correct, no custom build/start command, no Dockerfile override — matches platform-server-prod's working config exactly. Re-asserting the builder setting was a no-op.

Railway Agent's own deployment-diagnosis tool independently concluded this is a build-infrastructure issue at the BUILD_IMAGE stage on builder-zvijhr specifically, and recommended filing this report.

Consistently affected builder: builder-zvijhr (every failed attempt over 2 days)

Sample failed deployment IDs (11 attempts total, identical signature): 74e1adf4-e285-4f07-a4eb-77b157c5ee2f, 6b5e69a8-0e03-4d5c-ad50-1b0e75102c3a, f5862b0c-0620-4736-a76c-690cf7728314, 7082b1ed-ab88-4a8f-b03b-a0d4f563ad65, 7ad15060-d87a-4508-9171-0823c1f10449, 80e1fdb0-cc94-4233-8dbb-e943d4eff943, 9c5793a1-f5fa-4f67-930f-6e1c949b66ee, 37c04bf9-5401-4ac6-bcf7-70b426f88166, 37dbf033-f330-4d79-8b89-31b0ca30fd29

Could someone take a look at builder-zvijhr, or whatever build-plan/cache state is attached to this specific service? We have a small, low-risk fix queued up that's been ready to ship for two days and can't get past this build step.