Brief
Ship a churn model on SaaS churn/LTV data: boosting model, SHAP story, threshold policy tied to retention economics. WE W3-W4 bar.
Dataset
SaaS churn/LTV train/holdout; hidden retention-cost matrix for grading.
Starter guardrails (88)
- Pipeline skeleton + leakage checklist (date/id/group handling).
- SHAP story slots: global plot + top-5 business sentences.
Deterministic checks
Holdout PR-AUC >= 0.72; threshold policy cites cost matrix; SHAP artifacts present.
100-pt rubric
- Pipeline + leakage checklist: 20
- Model quality vs bar: 20
- SHAP story (top-5 + honesty): 20
- Threshold policy economics: 20
- Holdout bar met: 10
- Reproducibility: 10