docs(campaign): comments-boundary gate — format redesign before model A/B

docs/planning/v1.10-campaign-log.md

@@ -33,6 +33,34 @@ Chosen-model smoke (2026-06-10): `clean-titles --limit 5 --dry-run` on
 `claude-sonnet-4-6` rendered the dollar estimate (≈ $0.04) against the
 refreshed price table.
 
+## The comments-wave boundary gate (added 2026-06-11)
+
+The comments step does NOT start when the prior wave ends. When the
+campaign reaches the comments boundary (earlier waves swept or James
+calls it), **local Claude stops and tells James:** *"Comments boundary
+reached — go finalize the revised comments format before anything
+comments-related runs."* Two gates, strictly in this order:
+
+1. **Format redesign (James, external).** James intends to rework the
+   format and expectations of the comments section (a design session
+   outside this repo, likely a Claude/Fable conversation). The outcome
+   lands as a rules-content PR (`rules/comments.md` + its output-format
+   prompt files) with the standard prompt-content real-library smoke.
+   Prompt-touching → review tier before merge, per the merge-autonomy
+   policy.
+2. **Model A/B (the carve-out above), against the NEW format.** The
+   Opus-vs-Sonnet comparison runs only after gate 1 merges — comparing
+   models on a prompt that is about to be replaced answers nothing.
+   James may add a Haiku arm to the same ~$1 comparison if cost is
+   still a live question.
+
+Only then does the first comments wave run. Timing note, recorded
+because it's the cheap path: only ~24 books are comments-complete
+today, so the format change predates scale. Comments writes ARE
+audited, so those ~24 can be re-run against the new format afterwards
+with `regrade --step comments-enrich --before <format-PR-merge-date>`
+— a minutes-sized cleanup, not a sweep.
+
 ## Waves
 
 *(none yet — campaign starts after the v1.10.0 release and usage-tier