run-smoke

name	run-smoke
description	Run the end-to-end smoke pipeline (dummy model on the synthetic smoke dataset).

The smoke pipeline is the M0 milestone gate. It must always pass before any larger run.

Steps:

1. Validate the smoke configs:

   python -m rp_detr.cli.validate_config configs/experiments/smoke_rp_navi.yaml
   

2. Build the smoke manifest if it does not exist:

   python scripts/build_manifest.py --dataset smoke --out data/manifests/smoke.jsonl
   

3. Run the pipeline:

   make run-smoke
   

   This invokes python -m rp_detr.cli.run_pipeline against the smoke config and writes predictions, latency traces, and a run record under experiments/runs/e2e_smoke/.
4. Compute metrics:

   python -m rp_detr.cli.compute_metrics \
       --predictions experiments/runs/e2e_smoke/predictions.jsonl \
       --manifest data/manifests/smoke.jsonl \
       --out experiments/runs/e2e_smoke/metrics.json
   

5. Validate every produced JSONL against its schema:

   make validate-schemas
   

6. Report the run id (config hash + git commit), pass/fail of each step, image-level AUROC, and total wall time.

If any step fails, stop and report — do not delete the run directory.