Sonnet reviewer trial: re-baseline cost/latency figures and amend design 007 D1 nf-t56 ← Beads

Twinned from https://github.com/cfsh/ninefold/issues/900 by tools/beads/import_github.py's reconcile pass.

Follow-up to PR #899, which switched the reviewer to claude-sonnet-5[1m] as a speed trial (Aedan's ask). Two pieces of bookkeeping once the trial has data — both deliberately left out of #899 (design_docs are perimeter-denied, and the figures need real Sonnet passes to exist first):

• [ ] Re-baseline the reviewer cost/latency figures (review R2 on #899). notes/reviewer.md ("~12 minutes and ~$4–6") and CLAUDE.md ("~12 min and ~$5") are Opus measurements — the numbers this trial exists to move. After a handful of Sonnet passes, read total_cost_usd and wall time from the reviewer logs (~/.ninefold-reviewer/logs/pr-*.log) and update both files. Design 007 §2.2 estimated Sonnet ≈ 0.6×; check whether that survives the 1M-variant pricing.
• [ ] Amend design 007 D1 (review R3 on #899). The D1 row reads "do not pay 2× for Fable or trial Sonnet now" against shipped code that now runs Sonnet. Precedent: #881 took #879's amendment the same way. If the trial is reverted instead, close this item with a note.

Unclaimed — not any session's work until triage activates it.