[derek73]

Closes #792
Closes #771

The design question

Should one rule cover any phrase inside any known delimiter pair? The framing:

• delimiters set a word or phrase apart;
• a delimited clause is a nickname by default, or a maiden name where the caller declares the pair;
• where the words inside are known credentials, titles and the like, the delimiters group them apart from the name, and the clause is neither.

Option B was decided: set apart a clause of unambiguous credentials and title heads, each word by its vocabulary, with the exclusions below.

Why it was inconsistent. rules/S1 didn't make a clause a suffix. It dropped the brackets and let the words join the name.

Input	origin/master	This PR
John Smith (PhD, MBA)	nickname (1.4.0 through 2.8.0)	suffix PhD, MBA
John Smith (MBA, PhD)	nickname	suffix MBA, PhD
John Smith (Dr)	nickname (2.2.0 through 2.8.0)	title Dr
John (Dr.) Smith	middle Dr.	title Dr.
John Smith (PhD) Withers	suffix PhD Withers (2.4.0 through 2.8.0)	family Withers, suffix PhD
Mr Hugh (Snr) Boyle	middle Snr	suffix Snr
Justin (JP) Stearns	middle JP	nickname JP
John Smith (Sir), (Mr)	nickname	nickname (lone title of address)
Alex Green, FRCS (SN)	nickname	nickname (a credential's qualifier: #1065)
John Smith (Hon), MRCP (UK)	nickname	nickname (modifiers and styles only)

#792's headline example, W. Russell (Jr.) Withers, had already moved back to middle Russell Jr. when #1039 gated Jr.. It now reads suffix Jr., family Withers.

The rule (rules/S1, rewritten)

A clause wholly of words the vocabulary reads as no name is set apart. It is no nickname and no maiden name, and the words around it read as they would beside a nickname. Each word takes the role its vocabulary gives it: a title is the title, and a credential or generation is the suffix. A word that is both is the title in front of the name's first word and the suffix elsewhere. A member of the ambiguous credential class (S4) joins the clause only where its writing settles it after the name (S13).

Exclusions, each of which keeps its pair's reading:

• A clause of only title modifiers and styles: (UK), (Hon).
• A lone title of address: (Sir), (Mr).
• Titles alone straight behind a credential: FRCS (SN), which is a credential's qualifier.
• A credential written as initials before a name word of its comma part: (JP), (R.N.). A title/suffix dual counts here only with a period after each letter, or in capitals in a mixed-case name. This exclusion came from reading the moves, not from the options as first put.

Other content ending in an ASCII period keeps v1's escape: the brackets are dropped and the content reads bare.

#771

rules/M2's Precedence bullet claimed a marker inside a delimited clause made the span the maiden name "whether or not the pair is a configured maiden delimiter". That is false for M3's Except cases. The bullet is now scoped to clauses that read as the maiden name. No reading moves.

Mechanism

• _extract._set_apart: asks the vocabulary half and extracts the clause instead of dropping its brackets. _suffix_shaped became _period_escape.
• _post_rules._set_apart_clauses: gives the clause's words their roles once the name's are read, and asks the exclusions that need roles. It trims a comma off a word's span; R1's entry pass parts suffix words at that comma.
• _classify: tags a clause word without the comma written beside it.

mechanisms.md#CLAUSE-CONTENT-OVERRULES-DELIMITER records the split.

Measured

snapshot.py --local <source> --base <origin/master 84252268>, re-taken after rebasing onto master (the first measurement, against c620754, counted FEC 16):

Source	Entries moved
ctgov-officials	26
fec-individuals	17
psc-individuals	10
orcid-researchers	79
govuk-people, ukec-individuals, cz-candidates, dof-signatures	0

decisions/S1 records the moves by class. Two ORCID nickname clauses gain an S13 suffix-or-nickname report on MEng,. The seventeenth FEC entry is an interaction with #877, merged since: TOBIN, MARY (V) keeps suffix V but loses the suffix-or-name report master gives it, because the clause is set apart after assign's walk reports. A letter its writer bracketed is not a middle initial in doubt, so rules/C9 now lists a bracketed letter among the slots that settle it (Smith, John (V) → no ambiguity).

CLDR agreement: one Czech rendering moves. In Machová, Alexandra Zuzana (paní, Ph.D.), the given names now agree with CLDR, and the title gains paní, which CLDR leaves out.

Accepted costs (S1's Accepted bullets):

• FEC's state codes (ST) and (ME) read as titles.
• (Lady) reads as a title, because lady is in neither address set.
• John Smith, MD (Dr.) reads as a nickname.
• (MPH) alone leaves the name empty.
• Col. (Ret.) Smith reads suffix Ret., where 1.4.0 through 2.8.0 read title Col. Ret.. This re-pins two 1.x "known limitation" tests.

Reviews

Four reviews ran: code, design-docs, test coverage and comments. Then a review of the fix commits.

Code defects found and fixed:

• A word with a comma stuck to it lost its tags, so (MBA, PhD) and (PhD, MBA) read differently and (Jr, PhD) lost its generation.
• A leading comma stayed in the output.
• Duals escaped the initials exclusion, and the first fix caught (Lt).
• The widened R1 comma test joined (Jr.,) III.
• One loop never ran.

Docs fixed:

• S1's Then contradicted its Jack (PhD) example (N3).
• The lead example (MBA) was S13's case.
• The release spans were wrong.
• The corpus counts had no recipe.
• customize.rst, usage.rst and the _policy.py comments described the old escape.

The test review's unpinned branches are now rule examples. The suite passes on 3.11 and 3.15, and the serial benchmarks pass, with two new scaling rows.

🤖 Generated with Claude Code