I'm investigating an unexpected builder selection when using railway up

against an existing GitHub-connected Railway service.

The service is connected to a GitHub repository and has historically deployed

successfully using:

builder: DOCKERFILE

dockerfilePath: /backend/Dockerfile

The repository contains a tracked /backend/Dockerfile and

/backend/railway.json.

However, when I checked out a clean branch and ran railway up against the

existing production service:

railway up --service <existing-service> --environment production

the resulting deployment metadata showed:

builder: RAILPACK

dockerfilePath: null

The deployment then failed during:

using build driver railpack-v0.36.4

railpack prepare exited with an error

The existing production deployment was not replaced and remained healthy.

The important difference I observed is:

Successful GitHub deployment:

configFile: /backend/railway.json

builder: DOCKERFILE

dockerfilePath: /backend/Dockerfile

CLI deployment:

configFile: /railway.json

builder: RAILPACK

dockerfilePath: null

The CLI was executed from the repository's /backend directory.

I'm trying to understand the expected Railway behavior rather than report

this as a platform bug.

Questions:

1. Does railway up inherit the builder configuration of an existing

   GitHub-connected service?

2. Or does railway up independently determine the builder from the local

   upload/build context?

3. When running railway up from /backend, is that directory treated as

   the build context/root?

4. Should --path-as-root be used when deploying from a subdirectory?

5. Is there an officially supported CLI option for explicitly selecting

   /backend/Dockerfile for railway up?

6. What is the configuration precedence between service configuration,

   GitHub source configuration, railway.json, Dockerfile detection and

   Railpack detection?

7. Is there a more detailed error available for the Railpack

   prepare exited with an error failure?

I would particularly appreciate clarification on whether this is expected

behavior for railway up or whether the CLI should inherit the existing

service's Dockerfile configuration.

Thank you.