[derek73]

Part of #762, the first batch. These are the gates that phase 2 of the vocabulary model (#752) will re-read: S3 (credentials with interior periods), S4 (the ambiguous acronym needs words to spare), S6 (the case signal at the suffix slot), S10 (the trailing roman numeral), H3 (the title run's floor) and P5 (bound given names). Each file now opens with what is in force and the entry that decided it, and leaves the exempt list (88 → 82). Landing them before #752 starts gives its spec the current answer per gate, and keeps the two efforts out of each other's files.

What changes

• Six Standing blocks. Every example a line cites was parsed on the tree, and every pointer resolves to the entry it means; review printed and read all 68 pointers across the 13 files that carry a block.
• Partial-supersession markers on entry heads (README edit kind 1), where a later decision replaced part of an entry and the log never said so:
  • S6's accepted abdul Smith Jr Ma, reversed by S7's Should a credential run after a comma keep its comma when the last word is an ambiguous acronym? John Smith, PhD MEng reads first PhD, last John Smith #544;
  • S4's family-comma maiden examples and the predicate the maiden walk shared, superseded by M2's Rethink the maiden-marker rule (M2): a marker stands after the current name and takes words up to the trailing suffix run #601;
  • P5's abdul Smith Jr Ma loses the family name — P5's reserve does not mirror S2's bare-acronym fork #425 view comparison and its Sheik abdul Ma application of the Sheik abdul salam splits the bound given name that abdul salam ahmed joins #369 precedent, superseded by Rethink the credential-run rule (S2): read a name's trailing run once, over name units, and bind it before any join #614;
  • three S10 entries, by Should john v read V as the family name, as JOHN V and john x do? #721 and Read a one-case one-letter numeral as its capital does (#721) #722.
• One edit-kind-2 note in P5. Two paragraphs under Rethink the credential-run rule (S2): read a name's trailing run once, over name units, and bind it before any join #614's bullet belong to the 2026-08-22 abdul Smith Jr Ma loses the family name — P5's reserve does not mirror S2's bare-acronym fork #425 entry (PR refactor(group): one trailing peel for assign and the bound-given reserve (#425) #426); Rethink the credential-run rule (S2): read a name's trailing run once, over name units, and bind it before any join #614's bullet was inserted between them on 2026-10-07. The note says so and moves no text.
• The pointer grammar now resolves every pointer to exactly one entry. Before, an issue pointer could name several entries of one date and issue, and two S10 lines landed on the wrong one with nothing failing.
  • Where entries share a date and issue, the pointer goes on with the words its entry's head continues with (— 2026-10-18 #718 review, — 2026-09-14 (Derek)).
  • A bare pointer names the entry whose head ends there.
  • Where two heads are the same words, the pointer goes on with the words its entry's text opens with (P8's — 2026-09-13 #383/#479, AMENDS).
  • A pointer still naming more than one entry is a finding.
  • A parenthesized supersession note in a head is not read, so marking an entry doesn't break its pointers.
  • A test pins each pointer's target, not only that it resolves.
• The decisions README's Standing bullet now says to look for what reversed a line elsewhere before writing it, naming the cases Require a Standing block in new decisions files, and backfill the five split out on 2026-10-19 #776 and this batch found.

Review

• design-docs-reviewer, three rounds before the PR opened. Round one had 8 findings, round two 3, round three none.
• A three-reviewer PR review (code, tests, comments), then four narrow rounds on its fix commits. They found, and these commits fix:
  • an unmarked reversal in H3 (Dr Smith V, reversed by S10's Should "John V" and "Malcolm X" read V and X as the family name rather than a generational suffix? #718), missed when the block was written;
  • a typo after a pointer falling back to the bare entry in silence;
  • an entry whose head another's extends (review, review round) being unreachable;
  • a period or date inside an entry's head words cutting the words or being read as a second pointer;
  • an issue in prose after a date-only pointer;
  • a supersession note's case;
  • test gaps found by mutation.
• One fix narrowed what it widened. Limiting the date-to-issue window stopped a head copied whole from reading its issue. The next round caught it and the window was restored.
• Every check on the final tree:
  • every new row fails on the commit before it, or is one of the reviewers' mutation-catching rows;
  • the real Standing lines resolve exactly as at the first commit;
  • every head copied whole as a pointer resolves to its own entry, except the two pairs of identical-head twins, which fail as ambiguous.

Found, not fixed

Log claims that went stale with no entry saying so. They're in dated entries rather than Standing lines, so they're left for whoever revisits each file:

• S4's 2026-09-18 entry: Smith V, Ma under strict mode now reads family 'V', suffix 'Ma'.
• P5's 2026-08-21 Abd van der Berg née Jones reports no family name — the bound-given join counts words the maiden name takes away #411 entry: Berg, abd née Jones now reads given 'abd'.
• P9's Why does SMITH VD MA, JOHN read family SMITH MA with VD as a suffix? #573 entry: DOE, JANE PHD VD PROF. MA, noted in Require a Standing block in new decisions files, and backfill the five split out on 2026-10-19 #776.

Test plan

• uv run pytest: 22,238 passed, 123 skipped, 3 xfailed, on 3.11 and on 3.15. CI caught two stdlib deprecations a 3.11 run cannot see: a positional re.split maxsplit (3.13) and os.path.commonprefix (3.15). The trap is now in AGENTS.md's Gotchas
• uv run mypy, uv run ruff check
• Negative controls on the new rows and the target test

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.05%. Comparing base (e766bf1) to head (6cc44b3).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #795   +/-   ##
=======================================
  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.