ADR 0025: Speed is measured, never claimed: event-anchored metrics and baselines

Status

Accepted

Date

2026-09-15

Deciders

Founder

Context

The founder expects twins to reduce the elapsed time of a stage. Marketing and product copy must not say so until it is measured, and the measurement must compare like with like.

Decision

• Metrics anchored on source timestamps and computed identically for the phase-2 baseline and the twin cohort: notebook to record, SOP to check, hypothesis to protocol, slip to open item (p50 and p90 per program per week); reported separately: proposal to decision, hand-off latency, stage cycle time, writes without a human (zero by construction), decisions per persona-week with decline and edit rates.
• Baseline before twins: two full weeks on the phase-2 console per partner program, registered before the enable route works (409 baseline_missing); no aggregate published from fewer than three programs; definitions and SQL registered in CLAIMS-REGISTER.md before the window opens.
• words.speedClaims lint on site, docs and deck; interim copy “Every decision stays human and recorded. Timing claims wait for measured data.”

Consequences

• Phase-2 “question to decision” is not comparable across cohorts and is not used.
• Per-person rows live only in the organisation audit export, named in the DPA and the light validation package.

Alternatives considered

• Quote a partner’s prior cycle time: only with the partner’s written, dated baseline; otherwise rejected.

Follow-ups

• Register the first cohort results (R5 exit).