SMLRA/rogue-protocol ? reverse-engineered prompt

Build me a notebook that takes these ship log rows and predicts whether each line is a real fault or just noise.

I already have train, validation, and test CSV files, and I need the notebook to train on the labeled data, test honestly on the validation set, and then create a submission CSV for the hidden test file with one verdict per log id. Start with a simple model if that’s easiest, but please make it better than a basic baseline if you can, and let me adjust the cutoff so I can trade off missed faults versus false alarms. Use the text message column plus the other log fields in a sensible way, and show me the validation F1 so I know what is working.

Keep it simple enough that I can run the notebook top to bottom and get a valid submission file at the end. Also include a quick check that the output file has the right format before I upload it. If you need to look up current docs online for anything, go ahead.