[derek73]

A docs-only follow-up to #815 (PR #1107). No reading moves.

rules/M4 (a maiden marker makes the lone name word the family) said in its Except: "A word the vocabulary has claimed as a given name keeps that reading". The guard has only ever read the vocab:bound-given and initial tags (_NEVER_FLIPPED). When #445 wrote the sentence, the bound-given claim was the only given-name claim the vocabulary held. Plain given readings arrived later, from #752's phase 2i and from #789, and M4 flips those words:

abd née Jones     →  given 'abd'      (bound given name: kept)
Chas. née Jones   →  family 'Chas.'   (a given name written short: flipped)
Dean née Jones    →  family 'Dean'

The code has flipped these words since #445's own commit: Van née Jones and Di née Jones read family there. Chosen: keep the code, narrow the words, and state the flip as a cost.

Changes

• rules/M4, Except: now says "A bound given name keeps the reading its position gives it, and so does a word read as an initial" (P5, T4, T5). It drops "which is nobody's family name", which is false under both family-first orders.
• rules/M4, Accepted (new): a word the vocabulary reads as possibly a given name, without the bound-given claim, is the family name all the same, because the marker names the surname. Examples: Chas. née Jones, Van née Jones → family.
• mechanisms.md#TWO-LAYER-ASSIGN: says a claim there is a word's role. A given reading is a lean, and the field rules H1 and M4 read past it.
• _post_rules.py and _assign.py: the comment quoting M4 follows the new words, which the citation test requires. The paraphrase beside _WORD_ALREADY_CLAIMED changes to match.
• decisions/M4: a 2026-11-05 entry, a new Standing line, and the parse("Smith née Jones") reports no family name — a maiden clause empties a one-word name, where 1.4.0 kept it #445 carve-outs entry's head marked as superseded in words.
  • History: given readings were flipped from parse("Smith née Jones") reports no family name — a maiden clause empties a one-word name, where 1.4.0 kept it #445's commit on, and per-word given readings arrived from 2026-10-19 (Classify the particles by language and kind (#749) #765 … Derive the two name-clash markers from name readings (#752) #883).
  • Declined: widening the carve-out to given names written short. Measured over every corpus*.jsonl name under three orders, it moves one name, the entry's own example.
• rules_budget.json re-recorded, since M4 grew. The two new examples enter corpus_rules.jsonl and snapshot.jsonl as new entries; nothing moves.

uv run pytest, mypy and ruff pass.

Reviews

• design-docs-reviewer, on the first commit: four findings, all fixed.
  • The Chas. row sat in the Except fence though Then holds for it; moved to Accepted.
  • The Except's reason argued against the flip. The design question went to the maintainer, who chose A: narrow the words.
  • The history wrongly dated the gap from 2026-10-26.
  • "Nobody's family name" was false under the family-first orders.
• design-docs-reviewer, on the fix commit fff94a3: four findings, all fixed.
  • The PR list missed Derive the two name-clash markers from name readings (#752) #883 and three others.
  • mechanisms.md kept the universal wording.
  • The H1 analogy also covers initials (Dr. J. → family), so it is now support only.
  • "Would move readings" had no measurement; measured, it moves one name.

🤖 Generated with Claude Code