INST-10 · Live cross-check coverage metric

> Item derived from the fact-checking cycle analysis (**2026-07-03, v2**). New item (companion proposal **ENG-7 / OPS-3**). ⚠️ Reconciled — read the note below before acting.

### Finding (reconciled — see D3)
The "coverage fell 84.6% → 66.5%" claim was a **CSV error** in the companion proposal, already self-corrected. Corrected:
- **complete-flow rose** 57% → 66% (all stages executed)
- **direct-path fell** 36% → 28% (resolved without cross-checking)
- Among **published `main` 2025–26, coverage is ~97%**.

So cross-check coverage is a **shrinking** problem, not a growing one.

### Action
Instrument `coverageRate` per cohort/hub as a **live** metric, exposed via INST-8 (#2498). **Scope = measure/monitor**, not "restore coverage" — the original OPS-3 loses urgency given the correction.

### Reconciliation (D3)
The "84.6% → 66.5% drop" was a CSV error (direction reversed). Corrected: direct-path **36% → 28% (fell)**, complete-flow **57% → 66% (rose)**. Confirmed against the analysis: among published `main` 2025–26, coverage ~97% (the "all tasks" denominator yields ~66–79% — the gap is denominator choice). Conclusion: coverage is improving; treat as a monitoring metric, not an urgent restoration.

---
- **Priority:** 🟡 Medium · **Effort:** P/M
- **Dependencies:** exposed by INST-8 (#2498).
- **Epic:** #2489