Split rules/S2 and decisions/S2 into one rule per piece of evidence

[derek73]

Closes #682.

rules/S2 had grown to 324 lines holding seven rules, each with its own evidence. This splits it with no behavior change: the 89 example lines are the same set, corpus_rules.jsonl rebuilds byte-identical, and the snapshot matches.

ID	Rule
S2	a suffix word ends a name and never opens the string (also keeps the trailing read-once clause the others rely on)
S3	credentials written with interior periods (unchanged, keeps its ID)
S4	the ambiguous acronym needs words to spare, and the slots that report
S5	the given part's trailing slot after a family comma
S6	the case signal at the suffix slot
S7	the company of an unambiguous credential
S8	the credential's trailing run (#602)
S9	unlisted words that join the ambiguous class by shape

The issue's open question is answered by the repo's own constraints: rule keys are letter+digits (design_docs._RULE_KEY_RE) and IDs are never renumbered, so S3 stays put and the new rules take S4–S9.

Commits

1. Rules split. Each statement piece, example and Accepted clause moves to the rule it witnesses. Every rules/S2, bare S2 and in-rule (S2) reference in code, tests, other rules and AGENTS.md is retargeted by what it meant. Four pointer citations became quoted ones, so S5, S6, S7 and S9 each have an implemented: list the citation check holds.

2. decisions/S2 split. The 40 entries go to the rule each one decided:

   • S2 keeps Rethink the credential-run rule (S2): read a name's trailing run once, over name units, and bind it before any join #614, Should S2's trailing read count the particle chain's units instead of merging a copy of them? #620, Merge the trailing run's particle runs in group, so S2's read needs no BOUND stop or report gathering #625.
   • S3 gets two entries whose history had been filed under S2.
   • S4 through S9 take the rest.

   Entry text is unchanged except 16 positional references the move broke, which are now key-form pointers. Each file ends with a dated note on the split.

3. Design-docs review fixes:

   • nine #S2 / "under S2" short-form pointers to moved entries
   • R4's caps credential points at S9
   • S4 now states that case gives way to the count at the first slot after a comma
   • S4's rationale is scoped to the count by position
   • C2 and R5 pointers

4. Lesson recorded in docs/design/AGENTS.md: an ID split must grep all five citation spellings, not only the three the citation check reads.

5. PR review fixes (comment analyzer + AGENTS.md reviewer): decisions/S3's closing note (S3 only shared S2's log, it was never part of it), the test_design_docs fixture now pairs the case-signal name with S6 and probes S99 as missing, two count-and-case comments cite S4 and S6, one citation moves to decisions/S6.

Verification

• Full suite: 20,897 passed. mypy and ruff are clean.
• Snapshot matches.
• The design-docs reviewer ran axes 1–6, 9 and 10, and skipped 7, 8 and 11 as inapplicable. It confirmed nothing was dropped or duplicated in the statement, the examples or the decisions entries. Its six findings are fixed in commit 3.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.03%. Comparing base (d791bb6) to head (36b86a1).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #716   +/-   ##
=======================================
  Coverage   99.03%   99.03%           
=======================================
  Files          46       46           
  Lines        4557     4557           
=======================================
  Hits         4513     4513           
  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.