[derek73]

Closes #1000.

What changes

_vocab.split_pair merged Ph. + D. into a credential by a fixed pattern, with no vocabulary lookup. So removing phd from a Lexicon turned off the joined PhD but left the spaced pair a credential. Nothing could turn the pair off, and Lexicon.empty() and a strict choice whose files leave phd out read the two spellings differently.

The pair now follows the lexicon:

lexicon	split_pair("Ph.", "D.")	John Smith Ph. D.
lists phd as unambiguous suffix vocabulary, a credential (the default, every plain choice) or a suffix word	"suffix", as before	suffix 'Ph. D.' (unchanged)
no unambiguous phd: removed, marked ambiguous as a credential (add("phd", kind="family")) or as a suffix word, Lexicon.empty(), strict pt/es	None, the two words	middle 'Smith Ph.', family 'D.'

Where the lexicon holds no phd, the spaced pair and PhD agree that neither is a credential, and where it lists phd as unambiguous suffix vocabulary of either kind, that both are. Where phd is marked ambiguous, the pair is its two words while PhD is left to the count, so the two can still part (John Smith, Ph. D. reads title 'Ph.', given 'D.', where John Smith, PhD reads suffix 'PhD'). They cannot read identically otherwise, since unmerged the pair is two name words where PhD is one. The dotted Ph.D. stays a suffix by its shape (rules/S3) under every lexicon. Where the pair does not merge, an overlapping pair whose fold is listed merges instead: Jane Doe Dr. Ph. D. reads suffix 'Dr. Ph. D.', taking Dr. Ph. as drph.

Measured

The counts are in the #1000 comment, taken on master b2b75948.

• 687 corpus entries (684 distinct names) write the pair spaced, 624 of them in ctgov-officials, which also holds 100,908 joined PhD / Ph.D..
• Moves, master against this branch over those 687 entries (684 distinct names), comparing the seven role fields:
  • 491 entries (489 names) under strict pt, strict es and Lexicon.empty() (ctgov 441, fec 13, orcid 12, cz 1, committed 24);
  • 425 entries (423 names) with phd removed;
  • 0 under the default.
• tools/differential/snapshot.py: matches the snapshot, so no default parse moved.

Review

The independent review found that the first version answered "split" for an ambiguous phd, unlike every other ambiguous fold, and lost the pair's precedence over Dr. Ph.; ecc8e5a leaves such a phd to the other pairs' path and fixes the prose it flagged. The second review of that fix found the gate asked suffix_acronyms alone, so phd listed as a suffix word parted the spellings again, and four prose claims wrong when run; e5c0cab widens the gate to unambiguous suffix vocabulary of either kind and fixes them. The third review found that a phd marked ambiguous as a suffix word still fell through to the split-credential path, which admits it as a counted member; 00e28b2 has the pair answer directly, "suffix" or None, and its rows fail against e5c0cab.

Docs

• rules/S12: the Except bullet for the Ph./D. pair now says it is a credential only where the lexicon lists phd as unambiguous suffix vocabulary, and states the shape reading of Ph.D. and the overlap. The rule grew 674 → 777 words, and rules_budget.json is re-recorded. The new clause reads the vocabulary the rule already reads, so it adds no new signal.
• decisions/phd-merge: a Standing bullet and a dated Log entry, which records the declined option (b).
• Stale "whatever the vocabulary" passages amended:
  • decisions/S12's Standing;
  • decisions/vocabulary-exclusions' ph bullet;
  • vocabulary-model's 4i-d item (3), which opened this issue;
  • classify's pre-check comment.
• Release log: a Behavior Changes bullet.

Tests

• test_phd_pair_is_no_credential_where_the_lexicon_lists_no_phd: seven shapes under five lexicons without an unambiguous phd, one of them a family-first strict choice: neither word of the pair is a suffix.
• test_phd_pair_reads_its_two_words_where_it_does_not_merge: the two-word reading and Ph.D.'s shape reading, with phd removed or marked ambiguous.
• test_phd_pair_outranks_an_overlapping_pair_only_where_it_merges.
• test_phd_listed_as_a_suffix_word_reads_both_spellings_alike: phd listed as a suffix word reads the pair as PhD reads.
• test_an_empty_lexicon_names_no_file: restores coverage of the empty-lexicon branch of the reading-files lookup, which only the old Ph. D. reading under Lexicon.empty() reached.
• Negative controls: every new _vocab.py row fails against master's (test_an_empty_lexicon_names_no_file excepted, which tests _lexicon.py); 6 ambiguous rows fail against the first-round fix (1bf9f0b), while the ambiguous rows for John Ph. D., John Ph. D. Smith and the overlap read the same in both versions; the 3 suffix-word rows fail against ecc8e5a's suffix_acronyms-only gate.
• Updated tests that pinned the old behavior:
  • test_the_ph_d_pair_merges_whatever_the_vocabulary_says, renamed test_the_ph_d_pair_reads_by_the_vocabulary;
  • the strict half of test_reading_languages_under_a_strict_choice_name_the_chosen_files.
• test_layering's recorded count of unguarded _sets reads is 112 → 117, the gate's four new reads and one sets = binding.

Not in scope: whether phd and other credentials used across languages belong in mul.toml, and whether a strict choice should read mul.toml. That gets its own issue.

🤖 Generated with Claude Code