Your CLI uploads were failing at the code snapshot step. That step has a hard 15 minute ceiling covering the entire archive transfer, and when it is hit nothing reaches the builder, so no build logs are written and the deploy sits at Initializing before reporting a failed code snapshot. Redeploying a prior build skips the step, which is why those finish in seconds.

Archive size is the main lever you control. A .railwayignore that keeps node_modules, .git, build output and large local assets out of the upload is the direct fix. Removing .railwayignore and .dockerignore made the upload bigger, so that test pushed in the wrong direction, and --detach only stops the CLI from tailing, so it changes nothing here.

Your most recent CLI deploy did go through and built in about a minute, so this is working for you again. Connecting the service to a GitHub repo would take the upload out of the path entirely if you want to avoid it for good.

Reply here if a deploy stalls at Initializing again after adding the .railwayignore.