Problem
The wrap-up Orchestration Retrospective's Q5 (variety divergence) computes its coverage denominator from three journal markers — agent_dispatch, review_dispatch.reviewers, and remediation — via count_task_b_dispatch_sites. But the peer-review verify-only re-review path dispatches Task-A/Task-B pairs (teachback-gated reviewer reuse) that:
- do emit
dispatch_variety (numerator side, task-metadata stamp), yet
- emit no dispatch-site journal marker (no
agent_dispatch, not in any review_dispatch.reviewers, no remediation event — remediation covers the fixer, not the verify-only verifiers).
Result: on any arc that includes verify-only cycles, stamped dispatches exceed counted sites and compute_variety_divergence returns reason="coverage_exceeds_unity" — the tripwire designed for denominator collapse fires on a healthy, fully-stamped arc.
Observed
Session 9fa374df (2026-07-02), comPACT-led arc for the PR #1090 runbook row: 7 Task-B dispatches stamped (probe, 3 reviews, 1 remediation, 2 verify-only), denominator counted 5 (1 agent_dispatch + 3 review_dispatch.reviewers + 1 remediation) → coverage 1.4, coverage_exceeds_unity, no ratio renderable.
Proposed fix (either arm)
- Emit a marker for verify-only dispatches — e.g. extend
peer-review.md's verify-only section to write a review_dispatch event with a verify_only: true flag (or a dedicated verify_dispatch type), and include it in count_task_b_dispatch_sites; or
- Count from the numerator population — treat
dispatch_variety task_ids as the site set when they exceed marker-derived sites, demoting coverage_exceeds_unity to the genuinely-broken case (duplicate stamps / marker loss).
Arm 1 keeps numerator/denominator sourcing symmetric (the documented "coupled pair" invariant in wrap-up.md Q5) and is likely the smaller change.
Provenance
Surfaced by the wrap-up retrospective of the PR #1090 arc; the tripwire advisory behaved exactly as documented — this issue is about making the verify-only path emit the marker the denominator expects.
Problem
The wrap-up Orchestration Retrospective's Q5 (variety divergence) computes its coverage denominator from three journal markers —
agent_dispatch,review_dispatch.reviewers, andremediation— viacount_task_b_dispatch_sites. But the peer-review verify-only re-review path dispatches Task-A/Task-B pairs (teachback-gated reviewer reuse) that:dispatch_variety(numerator side, task-metadata stamp), yetagent_dispatch, not in anyreview_dispatch.reviewers, noremediationevent —remediationcovers the fixer, not the verify-only verifiers).Result: on any arc that includes verify-only cycles, stamped dispatches exceed counted sites and
compute_variety_divergencereturnsreason="coverage_exceeds_unity"— the tripwire designed for denominator collapse fires on a healthy, fully-stamped arc.Observed
Session
9fa374df(2026-07-02), comPACT-led arc for the PR #1090 runbook row: 7 Task-B dispatches stamped (probe, 3 reviews, 1 remediation, 2 verify-only), denominator counted 5 (1agent_dispatch+ 3review_dispatch.reviewers+ 1remediation) → coverage 1.4,coverage_exceeds_unity, no ratio renderable.Proposed fix (either arm)
peer-review.md's verify-only section to write areview_dispatchevent with averify_only: trueflag (or a dedicatedverify_dispatchtype), and include it incount_task_b_dispatch_sites; ordispatch_varietytask_ids as the site set when they exceed marker-derived sites, demotingcoverage_exceeds_unityto the genuinely-broken case (duplicate stamps / marker loss).Arm 1 keeps numerator/denominator sourcing symmetric (the documented "coupled pair" invariant in
wrap-up.mdQ5) and is likely the smaller change.Provenance
Surfaced by the wrap-up retrospective of the PR #1090 arc; the tripwire advisory behaved exactly as documented — this issue is about making the verify-only path emit the marker the denominator expects.