Service Variables not injected into container at runtime

I have a service ("sistema-stock") deployed from a GitHub repo via Dockerfile (Root Directory: /backend). The container builds successfully but crashes on start because DATABASE_URL and other Service Variables are not present in the running container's environment.

I confirmed this by adding a temporary debug line to the container's start command that prints env | cut -d= -f1 | sort before the app starts. The output shows all of Railway's own auto-injected variables (RAILWAY_ENVIRONMENT_NAME=production, RAILWAY_SERVICE_NAME=sistema-stock, RAILWAY_PROJECT_NAME=resilient-cat, plus ~30 other RAILWAY_* variables) — so the service identity/environment is correct. However, none of the Service Variables I configured in the dashboard are present — not DATABASE_URL, not JWT_SECRET, and not even a brand-new test variable (TESTVAR=hello123) I added specifically to rule out any issue with the existing ones.

Things I've already tried:

Deleting and recreating the variables from scratch using "+ New Variable"

Using the Raw Editor

Referencing the linked Postgres plugin's DATABASE_URL via ${{Postgres.DATABASE_URL}} (also didn't resolve/inject)

Manually redeploying multiple times after each change

Only 2 variables appear to "work" (PORT and NODE_ENV), but that's misleading — those two also have hardcoded ENV defaults baked into the Dockerfile itself, so I can't confirm whether Railway is actually passing them through or whether the image's own defaults are just masking the same underlying problem.

Project: resilient-cat

Service: sistema-stock

Environment: production

Can you check on your end why Service Variables aren't being injected into this service's container at runtime?