Retire the fields-only ledger rule, and forbid the shape

[derek73]

Closes #451.

expected_since_1.4.0.toml had one rule with no name_regex. Because classify() matches on fields alone when there is no regex, it claimed any name whose diff fit {given, family, suffix} — and _CORPUS_CLAIMS records a regexless rule's reach as the whole corpus, so it sat at its maximum and arrivals could never move it. test_every_rule_claims_the_recorded_share_of_the_corpus passed whether it explained 4 names or 40.

It drifted from the four names its prose described to seventeen at its high-water mark (decisions.md#H1 records the 14 → 17 step), and stood at fourteen across six unrelated behaviour families when this branch opened — every guard green throughout.

This gives all fourteen rules of their own, deletes the catch-all, and makes validate_rules reject the shape so it cannot return.

Where the fourteen went

Names	Rule	Commit
Mc Donald, Sir de Mesnil	ports of fix(#360) mc / fix(#367) sir de — rules the 2.x ledgers already had and this one never got	fc6e303
(, .,	fix(A2) content-free input. ., was on a comma rule, on nothing but its regex reaching it	7cc282f
John😀Smith	fix(emoji-boundary), a pinned copy of _tokenize.py's _EMOJI_RANGES	1b55eb1
Berg, Jan vd	port of fix(#380) — the name two comments promised would "arrive unexplained"	525f743
'Smitty' Jones Jr.	fix(N3), three tokens so it fits no two-token rule	7b2b2b6
Aishwarya Rai, Carod i	NOT WANTED rules for open v2.2 bugs #342 and #397	67ec9cb
Smith Jr., John V, Mohamad X, QC MP, Donald mc, Jack M.A.	four bounded rules; the catch-all deleted	ffd242a

8b06f48 narrows the rule's declared fields mid-way — it declared given while nothing it still explained moved that role, the fix(#424) shape decisions.md#H1 records.

Why four rules and not one

Forced, not chosen, and by three independent constraints:

• _LatinCopy carries one vocabulary and the alternation guard asserts one roster key per alternation, so jr (SUFFIX_WORDS) and mc/mp (SUFFIX_ACRONYMS) cannot share one.
• m\.?a\.? cannot be an alternation member at all — measured, it matches the corpus fragment 'M.A.', which _normalize leaves as 'm.a', not an entry. The ambiguous-surname-acronym rule already records the same finding.
• x is in no vocabulary set; it reaches suffix through _ROMAN, so its alternation is a _NOT_A_VOCABULARY_COPY declaration.

Two rules for readings nobody wants

Aishwarya Rai and Carod i lose their surnames, and both are open v2.2 bugs. The catch-all was classifying them as intentional 2.0 behaviour changes — and AGENTS.md names the 1.4 summary as the release log's source for Behavior Changes.

Each gets a rule saying the diff is classified because the cause is known and recorded, NOT because the reading is wanted — the standing decisions.md#P5 established for #369 — plus a delete-when-fixed note the dormancy check enforces. A [[never]] was rejected: an excluded name reports UNEXPLAINED and exits non-zero, which would block every release until both bugs are fixed.

What the reviews found beyond the issue

Each of these is the same defect one level up, and each is fixed here:

• The emoji rule's first draft was Ledger rules classify diffs they do not describe, and a rule that stops explaining anything is invisible #372 inside a fix for Ledger rules classify diffs they do not describe, and a rule that stops explaining anything is invisible #372. Its range spanned 1072 codepoints where _tokenize.py does not divide, so its own prose was false for every one of them. Now a pinned copy of _EMOJI_RANGES, with a guard proved to fail on the draft it replaced.
• A fields-only rule silently falsifies other comments. The 2.x idiom "such a name should arrive UNEXPLAINED" was false in this ledger for anything of the catch-all's shape. Measured over seven probes via classify(): zero arrived unexplained — four went to the catch-all, two to fix(comma-precomma-family), one to the neighbouring rule that exists to claim it.
• Copy-paste is how a ledger comment goes false without being edited. All three ledgers carried the same fix(#379) comment; four sentences were true only in the 1.4.0 file, including one citing fix(suffix-routing) — a rule no 2.x ledger has — and one saying a name "falls to the fields-only catch-all" twenty lines after the same file says none exists there (8237b54).
• The ban falsified six standing claims elsewhere, including mechanisms.md#LEDGER-RULE-SEPARATION, whose heading ("file order doesn't") is now exactly backwards (ad78f04).

Consequences worth knowing

• _sorted_rules is now the identity on every ledger that loads — the ban removes the only shape that could occupy its second tier, so file order decides every tie. Kept deliberately, with the docstring saying why: it is what makes the ban safe for any reader that does not call validate_rules first.
• A latent guard failure, found and repaired in passing. test_no_legacy_citations bans spec\s+[S§]?\d, and the 1.4.0 header's (spec S5) was passing only because the line wrapped between "spec" and "S5". Repointed. That guard is one reflow away from firing on any other wrapped instance — worth its own look.

Not in scope

• A ledger rule's explained set is printed but never checked, so a diff can shrink underneath it and every run stays green #452 — nothing counts what a rule explains, only whether it explains anything. Filed from this work; it closes the half a name_regex does not protect against.
• Mohamad X / the trailing-numeral ambiguity. By C-i a trailing X is borne as an ordinary name (Malcolm X). The fork does report SUFFIX_OR_NAME, while Aishwarya Rai reports nothing — so the live question is about SUFFIX_ACRONYMS_AMBIGUOUS and may belong on Rai is parsed as a post-nominal suffix, consuming a common South Asian surname #342. Recorded, not filed.
• A rules.md#P6 defect, recorded in decisions.md and not fixed: its Accepted clause promises comma-less shapes "keep their positional reading", and measured, Donald mc and Smith vd lose the family entirely. Its only example is de-shaped, which is why no test catches it.

Verification

All three baselines exit 0, unchanged diff counts: 1.4.0 229/0, 2.0.0 194/0, 2.1.0 102/0. Full suite 6062 passed, 156 skipped, 9 xfailed. Zero fields-only rules in any ledger (1.4.0 70 rules, 2.0.0 58, 2.1.0 51).

Note what is not the verification: a green gate. It was green throughout the absorption this issue is about. The per-step count movements are what was checked, commit by commit.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.64%. Comparing base (609f632) to head (ad78f04).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #453   +/-   ##
=======================================
  Coverage   98.64%   98.64%           
=======================================
  Files          45       45           
  Lines        3177     3177           
=======================================
  Hits         3134     3134           
  Misses         43       43           

☔ 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.