How to run
1. Put quote_run.py, build_predictions.py and holdout_eval.py in the job pack folder, next to jobs.json, labelled_examples.json and rate_card.json, together with the rules/ folder and splits.json (the split of the 150 labelled jobs; available on request).
2. Add a .env file there with ANTHROPIC_API_KEY=...
3. python3 quote_run.py --model=claude-sonnet-5-5 --final   (the 849 unlabelled jobs)
4. python3 build_predictions.py   (writes predictions.json)
