# Synthetic cohort definitions — v1

Teaching fixture only, authored 28 October 2026. Counts are invented aggregates, not Vantaige analytics, customer data or reconstructed raw events. No telemetry SDK or app was run.

Unit: distinct eligible individual user, not workspace/account. t0: first eligible signup timestamp. Weekly cohort membership is a UTC signup range with exclusive end. Exclude declared staff/test traffic consistently in a real pipeline. All rates here use acceptance rule mark_usable_v1: the user explicitly marks a draft usable; no independent factual-accuracy claim.

Windows per user (elapsed days, one day = 24 hours): first-week attempt/activation [t0,t0+7d); Day 7return [t0+7d,t0+8d); Week 1return [t0+7d,t0+14d); Week 2return [t0+14d,t0+21d). Return counts require an activated user and a distinct task from initial accepted task; repeat download does not qualify. Activated users accepted within first 7 days. Each user counted once per metric. Denominators: attempt/activation use eligible signups; return uses activated users, not all signups. Zero denominator=undefined, incomplete full-cohort observation=pending.

Cutoff: 2026-10-27T00:00:00Z. Conservative close bound=cohort_end_exclusive +7 days for activation, +8 for Day 7, +14 for Week 1, +21 for Week 2. Only display a full-cohort metric when cutoff>=close. This conservative convention avoids prematurely finalizing a cohort containing late signups. A partially observed subgroup can be reported separately with its exact eligibility/denominator.

Maturity checks:
- Sep 28–Oct 4 cohort end Oct 5: all windows closed by Oct 26.
- Oct 5–11 cohort end Oct 12: Week 1close Oct 26, Week 2close Nov 2.
- Oct 12–18 cohort end Oct 19: activationclose Oct 26, Day 7close Oct 27, Week 1close Nov 2, Week 2close Nov 9.
- Oct 19–25 cohort end Oct 26: activationclose Nov 2; later windows still pending.

Aggregates: 40 / 48 / 52 / 50 eligible users; activated: 20 / 24 / 26 / pending. Week 1 returns: 8 / 10 / pending / pending. Week 2 returns: 6 / pending / pending / pending. Percentages are rounded to one decimal at display; no intermediate ratio rounding. First cohort8/20=40%; one more would be 9/20 = 45%, illustrative only. Rule/date changes must produce a separately versioned calculation. Real extracts need event-time/ingestion cutoff and late-event policy.
