Session Handoff [review-tooling]: Count-First Overview Reader Released as 2.0.1077, Shared Heading Reduction Next #3178

Next steps, in priority order

1. Read a Section Heading Through One Reduction in pr_review.py #3174: read a section heading through one reduction in pr_review.py. One reduction primitive should serve normal, unwrap, and heading_count, and one leading-count rule should serve both vetting and counting. The unrecognized_in vetted-but-unread check should also compare which summaries were read rather than how many. Done looks like: the form-feed case and the two-nested-sections case in the issue each read correctly, and no reading of an observed shape changes.
2. Read a Finding Entry Titled With the Suppressed Marker as No Section in pr_review.py #3173: a finding entry's own summary is never a section heading. This covers both the pre-existing SUPPRESSED case and the widening Read Copilot's Count-First Overview Sections in pr_review.py #3172 added for PREVIOUSLY_MISSED and sibling_heading, recorded in the issue's comment. Do it in the same change as Read a Section Heading Through One Reduction in pr_review.py #3174 if the reduction lands first, since both decide what counts as a heading.
3. Pair a Review Findings Entry With Its Thread Past the Line Suffix in pr_review.py #3176: pair a Review findings: entry with its thread past the :LINE suffix. It is small and independent. Every round in that shape currently reports all its entries as unthreaded, so each costs a manual pairing.
4. Finish the Incidental Count Sweep, Including Counted Section Titles #3169: finish the incidental-count sweep. Lower priority, carried from the last round.

External blockers

None.

Internal dependencies

• Read a Finding Entry Titled With the Suppressed Marker as No Section in pr_review.py #3173 depends on Read a Section Heading Through One Reduction in pr_review.py #3174's reduction, since "is this a heading" and "what count does it state" should come from the same reduced text. Land Read a Section Heading Through One Reduction in pr_review.py #3174 first, or do both in one change.
• Pair a Review Findings Entry With Its Thread Past the Line Suffix in pr_review.py #3176 is independent of both.

State

• Released 2.0.1077 from main at b6c53dbc, the promotion Promote the Incidental Counts Rule and the Count-First Overview Reader to Main #3175 merge. It carried Refer to an Enumeration Instead of Stating an Incidental Count #3168, Read Copilot's Count-First Overview Sections in pr_review.py #3172, and Name the Never-Tiered-Down Groups in the Plural in backlog-burndown #3177, and closed Refer to an Enumeration Instead of Stating an Incidental Count in Prose and Comments #3160, Read Copilot's "Previously Missed Issues" Overview Block in pr_review.py #3167, and Read Copilot's "Open Issue" Overview Block in pr_review.py #3170.
• This machine's Codex and opencode skills snapshot was refreshed from b6c53dbc and reads current. The host-safety kit reports CURRENT.
• The primary checkout is on develop, clean, and fast-forwarded to 8198ee0a. No task worktree or feature branch from this round remains, locally or on the remote.
• No other repository was touched.

The parked decision queue

The queue holds 0 open issues with the decision label, counted when this was written. Nothing is parked.

What the last round did

• Read Copilot's Count-First Overview Sections in pr_review.py #3172 taught pr_review.py Copilot's count-first overview summaries, N previously missed issues, N open issue(s), and the **No open issues** stated zero. heading_count is now the single reader of a heading's count. marker_blocks folds a <summary> split across lines and keeps count-first prose out of the section. unrecognized_in reports a vetted previously-missed summary no reader finds. The promotion's own Copilot round, 3 open issues, read correctly with the new reader.
• Name the Never-Tiered-Down Groups in the Plural in backlog-burndown #3177 fixed a singular noun Refer to an Enumeration Instead of Stating an Incidental Count #3168 left in backlog-burndown, which Copilot flagged on the promotion.
• Filed Read a Finding Entry Titled With the Suppressed Marker as No Section in pr_review.py #3173, Read a Section Heading Through One Reduction in pr_review.py #3174, and Pair a Review Findings Entry With Its Thread Past the Line Suffix in pr_review.py #3176. pr_review.py Passes a Clean-Verdict Copilot Round Whose Headline Sentence Names a Finding #2567 and pr_review.py: a GitHub Actions billing-failure notice can be misread as UNRECOGNIZED_REVIEWER_OUTPUT #1071 were triaged and left as their own issues: pr_review.py Passes a Clean-Verdict Copilot Round Whose Headline Sentence Names a Finding #2567 needs a design decision on reading headline prose, and pr_review.py: a GitHub Actions billing-failure notice can be misread as UNRECOGNIZED_REVIEWER_OUTPUT #1071 is a different shape that would need a new status.

What not to repeat

• The local strict-review loop on this parser did not converge. Four passes on Read Copilot's Count-First Overview Sections in pr_review.py #3172 each fixed the last pass's findings and introduced one or two new ones, on shapes no Copilot body has shown. At round four the maintainer chose to file the rest (Read a Finding Entry Titled With the Suppressed Marker as No Section in pr_review.py #3173, Read a Section Heading Through One Reduction in pr_review.py #3174) rather than run a fifth. In this file, a fix that widens one pattern needs every reader that shares the pattern's assumptions checked against it. The HEADING_OPENER share broke the badge assumption in sibling_heading, and the blank-line fold stop broke a summary set off by blank lines.
• Run pr_review.py from a checkout that carries the change under review. The promotion's first round read as exit 43 from a primary checkout still on the pre-Read Copilot's Count-First Overview Sections in pr_review.py #3172 commit. With the current script it read shapes=ok.
• Record each pass's receipt before editing. Once a fix moves the content, the receipt for the pass that found it can no longer be recorded.

New learnings

• None beyond the issues filed. The non-convergence lesson is already local-strict-review's rationale for its two-round edit budget, and this round is one more instance of it.