The failed builds were caused by the service's root directory setting, not by the builder. It was set to "/ ", a slash followed by a space, so every new build looked for your source in a folder whose name ends with a space, which doesn't exist. That one setting produced both errors you saw, on Dockerfile and Nixpacks builds alike. Redeploys of earlier deployments kept working during that window because they reused the previous configuration with a plain "/". The setting has since been cleared, and along with the package name correction in your Dockerfile, every build since has succeeded, including the deployment that is live now. Those builds ran on the same build machine as the failed ones, so no reassignment is needed.