Essential session storage is required to sign in. Optional analytics remain off unless you allow them.
Deliver a complete, communicated data-science project that answers a real question. Include: (1) clear problem framing (target, success metric, supervised/unsupervised) and EDA; (2) a leakage-safe, validated modeling pipeline with tuning and honest held-out evaluation; (3) EITHER a designed A/B TEST or a causal analysis — state the hypothesis, compute and CORRECTLY interpret a p-value and confidence interval (p-value = prob of data this extreme IF the null is true, NOT prob the null is true), address power, and never claim causation from correlation alone; (4) a fairness/bias and limitations review (and interpretability, e.g. SHAP, if high-stakes); and (5) a stakeholder-facing writeup that leads with the insight and is honest about uncertainty and what the model can/can't do.
Sound framing; leakage-safe pipeline; tuned via CV; honest held-out evaluation with the right metrics — no training-set or test-set leakage.
A/B or causal analysis with a correctly interpreted p-value + CI and power; correlation-not-causation respected; no overclaiming.
Bias/fairness and limitations addressed; interpretability where stakes are high; PII/privacy considered.
Stakeholder writeup leads with the insight, honest about uncertainty and the model's boundaries; not just a high-accuracy notebook.
Use this README structure to make the work reviewable.
DevPath fetches a read-only, size-capped snapshot. Submitted code is never executed.
# An end-to-end DS project with an experiment or causal claim ## Goal Deliver a complete, communicated data-science project that answers a real question. Include: (1) clear problem framing (target, success metric, supervised/unsupervised) and EDA; (2) a leakage-safe, validated modeling pipeline with tuning and honest held-out evaluation; (3) EITHER a designed A/B TEST or a causal analysis — state the hypothesis, compute and CORRECTLY interpret a p-value and confidence interval (p-value = prob of data this extreme IF the null is true, NOT prob the null is true), address power, and never claim causation from correlation alone; (4) a fairness/bias and limitations review (and interpretability, e.g. SHAP, if high-stakes); and (5) a stakeholder-facing writeup that leads with the insight and is honest about uncertainty and what the model can/can't do. ## Acceptance criteria - [ ] Modeling & validation: Sound framing; leakage-safe pipeline; tuned via CV; honest held-out evaluation with the right metrics — no training-set or test-set leakage. - [ ] Experiment / causal rigor: A/B or causal analysis with a correctly interpreted p-value + CI and power; correlation-not-causation respected; no overclaiming. - [ ] Ethics & limits: Bias/fairness and limitations addressed; interpretability where stakes are high; PII/privacy considered. - [ ] Communication: Stakeholder writeup leads with the insight, honest about uncertainty and the model's boundaries; not just a high-accuracy notebook. ## Verification - [ ] Document installation and run commands. - [ ] Record automated checks and their exact commands. - [ ] Add representative output, screenshots, or a short demo where useful. - [ ] Confirm failure paths and known constraints. ## Decision log ### Decision title - Context: - Choice: - Alternatives considered: - Tradeoffs: ## Evidence - Link each rubric criterion to the file, test, or artifact that demonstrates it. ## Known limitations - List what is intentionally out of scope and what should be improved next.