[derek73]

Follow-up to #761. This lands three loose ends the skeleton migration left behind. No rule text, example or parse result changes, and the corpus and snapshot do not move.

1. Citation comments quote the bullet they name

Fourteen comments named a rule's bullet by description and quoted nothing ("C14's Except", "R3's Accepted clause", "W4's interpunct Except", …), so tests/v2/test_doc_citations.py could not check them. These include the four found after the migration plus ten more that a grep for 's Except|Accepted|Precedence and "Then bullet" turned up. Each comment now has three parts:

• the bullet quoted next to its rule ID (an Except bullet quotes its whole bold lead, which bullet_lead_findings holds it to);
• a line saying which clause the code implements or the test pins;
• a decisions/ pointer.

The citation test now reads all of them. bullet_lead_findings counts 7 citations that name a bullet, up from 2.

Two comments had named the wrong bullet:

• test_parser.py's "M2's Accepted row" is M2's Except bullet on a marker in post-comma text.
• test_pieces.py's "no-comma peel's Accepted limit" is S7's.

Review found two comments next to new citations that were false about how names parse, measured on this tree:

• test_properties.py claimed two maiden-clause pairs differ. Both spellings of each pair have read alike since Rethink the maiden-marker rule (M2): a marker stands after the current name and takes words up to the trailing suffix run #601, so I deleted the claim.
• _extract.py illustrated the M3/M2 contrast with née V. M2 reads that as the maiden name too, so the example is now née PhD.

Not touched: the notes= strings in tests/v2/cases.py, and two en.toml notes that say "P6's Accepted clause". They are data strings rather than comments, and the citation sweep reads neither.

2. Design review, axis 3: check Except fence rows against their bullet

docs/design/AGENTS.md axis 3 is amended in place. A row in an Except fence must be one where Then does not hold. Read each row against its bullet's bold lead. A row Then still decides moves out of the fence: to the main fence, or to the Accepted bullet that records it as a cost. Nothing checks which fence a row stands in.

Two cases are on record, both caught in review:

• H1's given-name rows, PR Migrate the H rules to the skeleton format; split out H7 (#761, part 5) #779.
• S3's switched-off Msc.Ed. row, PR Migrate the S rules (S1–S9) to the skeleton format (#761, part 8) #783.

A third case, a T1 row in PR #782, was in my brief, but neither that PR's body nor its commits record it, so I left it out.

3. Retired rule IDs are guarded

#786 changed the tombstone wording to "A retired rule's decisions entry records the retirement", which left "never reused" unenforced. As decided:

• A retired rule's decisions file carries a line of its own opening Retired:.
• test_no_rule_reuses_a_retired_id refuses a rule file whose ID a decisions file records that way. The readers (rule_file_ids, retired_ids, reused_id_findings) are in tests/v2/design_docs.py.
• Two more cases are findings, because either would switch the refusal off silently: a Retired: line in a file keyed by no rule ID, and a line that misspells Retired: (bold, a list marker or number, lowercase, an indent).
• The convention is stated in both docs/design/rules/README.md and docs/design/decisions/README.md.

No rule is retired today, so the real-tree test refuses nothing. Its own assertions check only that both readers reach the folders.

The negative control is test_a_rule_file_reusing_a_retired_id_is_refused. It builds a synthetic decisions folder recording X9 as retired and a rules folder with X9-reused.md, then runs them through the same readers:

• It fails with the ID check disabled (measured).
• Its control half, the same files with Folded: in place of Retired:, passes.

The 1. Retired: misspelling row failed against the first version of the pattern before the fix.

Verification

• uv run pytest: 22197 passed, 123 skipped, 3 xfailed.
• uv run mypy: clean.
• uv run ruff check: clean.
• The design-docs reviewer ran on the branch. Its six findings are fixed in 9355e31, and a second pass over that fix commit found nothing.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.05%. Comparing base (9932592) to head (9355e31).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #788   +/-   ##
=======================================
  Coverage   99.05%   99.05%           
=======================================
  Files          47       47           
  Lines        4642     4642           
=======================================
  Hits         4598     4598           
  Misses         44       44           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.