Twinned from https://github.com/cfsh/ninefold/issues/872 by tools/beads/import_github.py's reconcile pass.
Implement [design 007 — Reviewer verdicts written for their readers](https://github.com/cfsh/ninefold/blob/main/design_docs/007-reviewer-verdicts-for-readers.md) (doc PR: #871).
The recommendation in two lines: the reviewer's clarity problem is its output contract, not its model. Pin --model claude-opus-5 in dispatch.py (today the model is the CLI default by accident), and rewrite CHARTER.md's verdict format — plain-English lead, findings before bookkeeping, delta re-reviews, convention-check freeze after pass 1, lighter mode for docs-only diffs — keeping every paid-for anti-gaming check (verbatim ask quotes, CLAIMED-NOT-FOUND, B2, severity floor, regression exemption).
Shape of the work (task cards in the doc's §8; the session that takes this may split them into issues then):
- T1 — one mechanical PR: pin the model in dispatch.py + argv test. Independent, land any time.
- T2 → T3 → T4 — one stacked sequence of charter PRs (same file): verdict restructure, delta re-reviews, docs-only mode. Invariants in §6: the REVIEWER-VERDICT line format, test_charter_notes.py's notes list, and the label-writing block survive untouched.
- T5 — measurement after ~20 organic verdicts: pass-1 median ≤ 700 words, re-review median ≤ 400, zero lost ledger statuses; report on this issue.
All four design forks were answered by Aedan on 2026-08-13 (doc §0); the open questions in §7 carry shipping defaults.
None.
No comments.