A service is deployed only with railway up from CI. The service also has a GitHub repository connected, with automatic deploys disabled, Wait for CI off, and branch main. It uses Railpack, has a pre-deploy command, and Skipped Builds is off.

1. When a variable change is saved and its staged change is deployed (dashboard Deploy, or CLI variable set without --skip-deploys), does the new deployment build from the stored source of the latest railway up deployment, or from the connected branch's latest commit? Is that guaranteed while a repo is connected but auto-deploy is disabled?
2. Is that build always a full rebuild, and does the pre-deploy command always run?
3. Does that deployment use the service's current saved variables, or the previous deployment's snapshot?
4. When a workspace compute hard limit is raised or removed, are stopped services redeployed (rebuild plus pre-deploy, current variables) or restarted from their existing image and snapshot? Does changing the limit while services are running redeploy them?
5. Is there a supported setting that makes a deployment fail instead of building from a source other than the latest upload?

Platform-semantics question only. No account action is requested, and no identifiers, logs, code or secrets are included.