Brief
Rebuild the weaker of your two mini-labs and defend both plus the shared eval protocol from m4-l13. Deterministic checks then LLM review.
Dataset
Same CV + NLP splits; rebuild uses a fresh seed.
Starter guardrails (88)
- Protocol template: metric bars, freeze statements, leak checklist, cost log.
- Rebuild diff scoped to the weakness named in the defense.
Deterministic checks
Both mini-lab bars re-verified (acc >= 0.70 CV, F1 >= 0.78 NLP); protocol sections complete.
100-pt rubric
- Weaker-lab diagnosis honesty: 20
- Rebuild quality (bar re-met): 25
- Protocol completeness: 20
- Defense clarity (per-class / per-slice command): 20
- Reproducibility (fresh seed): 15