List credentials written in two chunks joined at a full stop (#998) - #1297

Closes #998

Lists the credentials written in two chunks joined at a full stop that #998 measured and the six local corpora show, so the vocabulary reads each word whole before rules/S3's shape does. This is step 2 of decisions/period-joined-words (#1291): the division of a word at its full stop has to come after this batch.

What changes

• 43 listed, each vetted on its entry (tools/vocab/query.py, tools/name_usage.py, counts in the note), each with a cased mask except msci and msi:
  • 26 in en.toml: bpharm, cpsych, mpharm, msci, mtech, dtech, dsci, mmath, moptom, and others.
  • 16 in id.toml: mkes, mkep, mkom, skom, mpd, spd, mhum, mkn, msi, and others.
  • menf in pt.toml.
• msi marked ambiguous by a family reading, Msi being a Xhosa surname. Refusing it reversed Jane Doe, MSi. Marking it costs ANDISWA ZANELE MSI its surname, which no local corpus writes; a row pins that as the accepted cost.
• 12 refused with excluded entries, because the spelling without stops is a name or another word:
  • peng, ceng: Chinese surnames.
  • barch, bchir, mendo, sked, shut, merg, mak, mor: borne as names.
  • bed: the Nepali given name.
  • ssi: the romanized Korean honorific 씨, as in Park Ji-hoon ssi.
• btech left unlisted on purpose. B.Tech is the example of an unlisted dotted credential in eleven files, including comments in _pipeline/_vocab.py and _render.py. Moving those examples is a follow-up of its own.

Measured (2026-11-09)

tools/differential/measure.py --base against master b89c548, then the corpora that moved again on the final tree.

• Snapshots: the committed snapshot and the CLDR, GND and Wikidata records do not move.
• ctgov-officials: 407 readings move, 122 of them in a field. Each one sampled is a credential leaving a name field or the title for the suffix.
• orcid-researchers: 81 move, 50 of them in a field, mostly Indonesian degrees that had read as titles.
• Other moves: cz-candidates 2; jrc-names 1 (Bernd Lange SPD: family 'Lange', suffix 'SPD'); psc-individuals 1, its report only (Mr Afikile Manduleli Msi).
• Agreement: orcid-researchers +9, −1 (a one-word name with credentials). cz-candidates +1. psc-individuals and patentsview-inventors unchanged.
• opensanctions-parliaments not measured: the shared cache predates the datasets the tool reads.

The division's cost, which decisions/period-joined-words' Open line on unlisted credentials asks for, is in that file's new Log bullet. Over the six corpora #998 counted:

• Cost: 18 unlisted credentials would read as name words (P.Eng 4, B.Chir 2, …). With second chunks of any length, the cost is 44 (M.Ed 8, B.Ch 5, …), 42 of them after a comma.
• Gain: 63 glued initials would get their surname back.

For a decision

bed is refused for the given part after a family comma (Thapa, Bed reads suffix 'Bed' with the word listed), though no local corpus writes that shape, against a gain of 6 readings (3 in a field). decisions/suffix-acronym-collisions' Standing line for ded says a given name is not a position the suffix claim acts on, and Thapa, Ded reads suffix 'Ded' today. That Standing line is corrected to say the claim reaches a given name after a family comma and under a family-first order (Luli Ded), for edd too, and both stay unmarked; whether to reopen them is yours.

Docs

• decisions/period-joined-words and decisions/suffix-acronym-collisions each gained a Log bullet.
• nameparser/vocab/AGENTS.md now says to vet a dotted credential's spelling without stops.
• docs/release_log.rst has a new bullet.
• No rule's stated limits move: rules/S3 still reads the unlisted shape.

Reviewed by design-docs-reviewer (twice), pr-test-analyzer and code-reviewer; their findings are fixed in the second and third commits. The period-joined-words Open line moves from #998 to #1255, since this PR closes #998.

🤖 Generated with Claude Code