Split rules/C1 and decisions/C1 into one rule per piece of evidence

[derek73]

Closes #728.

rules/C1 (the suffix-comma decision) had grown to 352 lines holding ten rules, each with its own evidence. This splits it the way #716 split S2. There is no behavior change: the 103 example lines all carry over (plus one policy-annotated line on an existing name), corpus_rules.jsonl rebuilds byte-identical, and the snapshot matches.

ID	Rule
C1	the suffix-comma decision (keeps the word count, the lenient/strict knob, the CJK-honorific Accepted clause, and a one-line map of C3–C11)
C3	counting the words before the comma (#575, #613(1); the Freiherr Accepted clause)
C4	the ambiguous credential class at the comma (#544, #554, #562)
C5	paired initials after the comma (#563; the G.J.R. Accepted clause)
C6	an unlisted all-caps word needs the name's case contrast (#564)
C7	a postnominal part is read before any join (#613(3), #617, #636)
C8	the listing form's part after the comma (#296, #325, #603; the Ms. Accepted clause)
C9	a one-character suffix word inside a comma part (#430, #432; the further-comma Accepted clause)
C10	a declared delimiter parts a trailing suffix part (#549; the maiden-marker and delimiter-core Accepted clauses)
C11	what the comma decision reports (#544, #562, #641)

C2 keeps its ID, so the new rules take C3–C11. Each new rule has a short Rationale drawn from its statement. So that every rule carries a boundary, C11's John Smith, PhD DO DO (reporting nothing) gained a · boundary mark. C10's boundary is its Accepted clause's own input run under the dash policy, a new example line on an existing corpus name. C7 declares no-boundary:, pointing at P5's Salam, abd Allah.

decisions/

• Split by rule. Each entry moved to the file of the rule it decided. C5 gets a pointer entry to decisions/S9 (Dotted initials after a two-word surname read as a credential: García Márquez, G.J. has no given name #563), where its decision has always lived.
• New decisions/R1. The parse("Smith, MD PhD").suffix returns "MD, PhD" — a comma the input never had #429, parse("John Smith MD PhD").suffix returns "MD, PhD" — a run without a comma renders with one #436/Should the suffix separator be derived from the comma the writer typed, rather than from segment shape? #437 and Should Parser.revise(name, suffix=name.suffix) be the identity? (John Smith MD PhD gives MD PhD, and revising with that string gives MD, PhD back) #511 entries decide R1's entry pass (one comma segment is one suffix entry), not the comma decision. R1's history: now points at them.
• Measurement paragraphs cut to one line each: population, counts, baseline commit, PR. That covers the WHAT MOVES, COST, BLAST RADIUS, VIEWS and RE-PINNED paragraphs of parse("Smith, MD PhD").suffix returns "MD, PhD" — a comma the input never had #429, parse("Smith, PSM I") reads "PSM" as a given name — a suffix run breaks on a trailing I or V #430, parse("John Smith MD PhD").suffix returns "MD, PhD" — a run without a comma renders with one #436, Should a declared suffix delimiter inside a joined connective run be dropped? Smith, John, Puig - i Soler keeps the - in the suffix #549, Why does De La Cruz, Ed lose its given name? #575, Should a comma part opening with a credential be a suffix comma? John Smith, PhD Jones reads family John Smith #603, Rethink the suffix-comma rule (C1): decide the comma's structure once, after the words are tagged #613 and Should John Smith, Jr., van der Berg Prof. read title 'Prof.' as John Smith, Jr., Prof. does? #636. The decisions, Declined and Accepted clauses, and the moved readings stay.
• Superseded passages collapsed to their decision plus what replaced it: the Why does John Smith, PhD DO DO read given 'PhD' and family 'DO DO John Smith'? #562 entry (replaced by Rethink the suffix-comma rule (C1): decide the comma's structure once, after the words are tagged #613), the "segment decides" sentence in Should a comma part opening with a credential be a suffix comma? John Smith, PhD Jones reads family John Smith #603, Why does De La Cruz, Ed lose its given name? #575's "segment runs before classify" paragraph, and parse("Smith, MD PhD").suffix returns "MD, PhD" — a comma the input never had #429's "Left alone" and ledger paragraphs.
• Word count (wc -w at HEAD): 17,548 → 15,289 across the eleven files, decisions/R1 holding 3,732 of them.

Retargeting

Every rules/C1, decisions/C1 and bare C1 reference was retargeted by what it meant. That covers code and test comments, tests/v2/cases.py notes, other rules' prose and interacts: lines, other decisions entries, mechanisms.md, AGENTS.md and two release-log pointers (now R1). References to the decision as a whole stay on C1. Three new quoted citations give C3, C7 and C9 an implemented: list the citation check holds: C3 and C7 in _comma.py, C9 in _assign.py. _RETIRED_EXCERPTS' C1 row now names C7, where that text went. One case note, bare_ambiguous_acronym (John Ed), credited C1 for a no-comma reading. It now cites S6, the rule that states it.

Open question (for Derek)

#611's entry spans C1, S2 and P6, and it stays in decisions/C1 for now. Should it stay there, or move to mechanisms.md's field notes?

Verification

• Full suite: 21,465 passed after merging master (Should john smith nee x x read like JOHN SMITH NEE X X? A shape-only numeral in a maiden clause #731 and its revert Revert #731's maiden-clause numeral fix; accept the shape in T4 #733). ruff and mypy are clean.
• tools/differential/build_rules_corpus.py rebuilds corpus_rules.jsonl with no diff.
• Design-docs reviewer. It ran axes 1–10 and skipped 11, since no rule's behavior changed. It confirmed that every old statement sentence, Accepted clause, example line and decisions paragraph lands exactly once, and that every condensed baseline matches its PR's merge parent. Its 13 findings are fixed in the third commit:
  • C10's first boundary had no policy annotation, so the rule never ran on it.
  • C7's waiver reason was wrong for P2 and P3.
  • C3's and C4's new Rationales overclaimed.
  • The split note misled where a moved pointer meant a piece now in another file.
  • One condensed count lost its non-hand-read mover.
  • Three collapses dropped clauses that are still in force.
  • The rest were interacts: and pointer touch-ups.
• PR review (comment analyzer and AGENTS.md reviewer): every retarget checked, and its findings are fixed in the fourth commit. One note read "C11's exception" where it meant C11's report. "C10's Accepted entry" now names which of its two Accepted entries it means. A C1 credit in a note now goes to C8, where the reading is stated. Older decisions pointers in S9, S7 and suffix-acronym-collisions that meant C4, C5 or C6 now name them. The word count above was corrected.
• Lesson recorded in docs/design/AGENTS.md's ID-split paragraph: retarget each pointer that names a piece, and read the new prose a split writes against its piece's exceptions.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.04%. Comparing base (09fc524) to head (f6b379e).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #734   +/-   ##
=======================================
  Coverage   99.04%   99.04%           
=======================================
  Files          46       46           
  Lines        4589     4589           
=======================================
  Hits         4545     4545           
  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.