Summary

Every deployment for this service fails before any build logs are emitted. The failure happens during Railpack plan creation, not during install/build. Railway's AI diagnose gives contradictory fixes on each run; we've tried all of them and none work. A separate service on the same account deploys successfully with engines.node: "22.20.0" and engines.npm: "11.7.0".

Push to main triggers deployment

Deployment enters build phase

No build logs appear (empty log panel)

Deployment fails/crashes

AI Diagnose suggests a different config change each time

What we've already tried (none helped)

engines.node: >=24.13.1, >=22.20.0, 22.0.0, "22", "22.20.0"

Same failure

engines.npm: present (11.7.0) and removed

Same failure

Added railpack.json with "provider": "node"

Same failure

Relevant commits on main:

The actual error from Railpack's plan-creation phase, stack trace, panic, or exit reason from your backend/build orchestrator. The customer-facing log panel shows nothing; we need whatever your system logged internally when the plan failed.

Confirmation of provider detection: what provider(s) did Railpack consider, and which did it select? Our repo has signals for both Node and static/Vite:

package.json with build and start scripts (Express server)

index.html at repo root (Vite)

vite.config.ts at repo root

railpack.json explicitly sets "provider": "node"

Why this service fails but another on the same account succeeds with identical engine pins (see comparison below).

Start command: NODE_ENV=production tsx src/server/main.ts (Express + vite-express)

Build command: tsc -b && vite build

Vite output dir: dist/client (non-default; configured in vite.config.ts)

Runtime: Express serves the SPA; production uses prebuilt client assets via vite-express

We do not want Railpack's Vite SPA/Caddy path. If SPA detection is involved, we'd expect RAILPACK_NO_SPA=1 or honoring the explicit start script, but we can't tell because plan creation never completes.