TRY FIRST — use copies; keep original fixtures unchanged.
1 Write a funnel and retention metric contract using the template.
2 Reconcile45 event rows to41 accepted. Find the retry; create a conflicting duplicate in a copy.
3 List each user's furthest funnel step. Move U06 exercise to firststart+72h. Predict new counts.
4 Create the three-row cohort matrix. Distinguish calendar week from product age.
5 Calculate mature pooled day 7; show why3/12 is wrong. Compare a weighted/unweighted paper example.
6 Complete an experiment plan before reading output. CompareITT 20/100,28/100 with exposed-only 20/90,28/80.
7 Calculate absolute/relative effect, interval, planning size andSRM 130/70. Explain uncertainty.
8 Write a250-400 word decision memo and diagnose 'retention 25%,lift 8%,launch now'.

EVIDENCE LOG: task | predicted count | exact command | actual result | limitation
All results use fictional data. No live feature experiment or tracking deployment.
Mini-project: add a SECOND scenario by copying the fixtures to your own folder. Preserve
baseline controls. Change exactly one defined feature (e.g.a return at an interval endpoint),
add an independent expected answer/test, and explain why the result changes. Do not call
manufactured records empirical evidence or a personal commercial result.
