Spun out of #597 (D4).
The #599 redesign deliberately added NO numeric question budget — a fixed cap invites gaming, and the breaker plus the closure checklist own convergence. The honest observable is telemetry: rounds-per-spec (question rounds between Phase 3 entry and Phase 5 approval), trending like the three existing /ca:metrics figures.
Investigate:
- Data source: nothing currently records question rounds — gate-events lines, the sprint log, or spec frontmatter stamped at approval; whichever is chosen must be append-only-compatible and cheap enough to never be skipped.
- Whether the metric is per-spec or windowed like the existing three (20-commit window with trend arrow), and what a "bad" trend actually indicates (spec inflation vs. model regression vs. genuinely harder features).
- Guard against the metric becoming a target: if it ever gates anything, it recreates the cap the redesign rejected — decide up front that it stays observational, and write that down.
Done when: a data source exists and /ca:metrics renders the figure with a trend, or the metric is declined with the reason recorded.
Spun out of #597 (D4).
The #599 redesign deliberately added NO numeric question budget — a fixed cap invites gaming, and the breaker plus the closure checklist own convergence. The honest observable is telemetry: rounds-per-spec (question rounds between Phase 3 entry and Phase 5 approval), trending like the three existing /ca:metrics figures.
Investigate:
Done when: a data source exists and /ca:metrics renders the figure with a trend, or the metric is declined with the reason recorded.