# Weekly metrics review — reusable template

Specification/template, authored 28 October 2026. No meeting, interview, experiment or product improvement occurred for this article. Example counts are synthetic; no causal uplift is asserted.

## Record before reviewing

Owner; review date; product/model/prompt version; event-contract/schema version; acceptance-rule version; analysis unit; eligibility filters; extract cutoff; late-event policy; mature windows and pending windows. Choose a cadence that fits the product, not a universal 45-minute Tuesday meeting.

## Review sequence

1. Reconcile a bounded set of application tasks with events. Are duplicate, missing or late events changing counts? Preserve logical task and attempt IDs.
2. Confirm cohort membership, t0 and windows. Do not mix users/accounts, signup/activation anchors, Day 7/Week 1/on-or-after metrics.
3. Present counts and denominators. Keep undefined and pending distinct from measured zero.
4. Inspect the most consequential gap. Request-to-draft can involve queueing/cancellation/errors/capture; draft-to-acceptance can involve fit/quality/interface/delayed review/capture. Treat these as hypotheses until evidence supports a cause.
5. Review structured rejection reasons and authorized qualitative evidence. No raw customer documents in analytics by default; inspect permitted underlying evidence through controlled paths if needed.
6. Choose one change or investigation, with owner and expected mechanism. Record what is not known and the guardrail metrics.
7. Set when follow-up will be mature enough to review. Do not invent a successful next cohort or call an observed association causal.

## Synthetic discussion example

Cohort 1 has 20 activated users and 8 Week 1 returns: 40%. One additional person would change the rate to 45%. Cohort 3 has 26 activations out of 52 eligible users, still 50% activation like 20/40 inCohort 1. Those counts do not establish improved activation efficiency. Cohort 3 Week 1 is pending at the fixture cutoff, so no completed retention comparison is available.

Decision example: inspect a small authorized set of task outcomes to determine whether a delivery/acceptance gap reflects task fit or event capture. Owner: assign. Evidence: not_collected. Product change: not_decided. Outcome: not_run. Revisit after the defined window and evidence are available.

## Decision note

Question / observation with counts / mature population / alternative explanations / evidence references / action and owner / metric plus guardrails / expected review date / limitations / actual result when available.
