[derek73]

Phase 2h of #752: generation readings gated by kind. Addresses #843's agnomens.

What changes

• A generation word that is also a name is gated. Such a word now reads as the suffix only where the name has a word to spare, the way a bare ambiguous acronym (Ma) does, and the parse reports the choice as suffix-or-name. A generation or agnomen reading beside a given or family reading derives the new SUFFIX_WORDS_AMBIGUOUS (Lexicon.suffix_words_ambiguous), the third derived marker after phase 2a's two. The rule is rules/S11, a suffix word that is also a name.

• The Brazilian agnomens (filho, filha, neto, neta, sobrinho, sobrinha in pt.toml, and netto in a new pt-BR.toml) get a new agnomen kind.

  • Each carries its name readings, with dated counts.
  • They are read by the count alone, because an agnomen is written in the case of the name around it. AGNOMENS (Lexicon.agnomens) lists them.
  • The count alone is the agreed design, and Derek confirmed it on 2026-10-25 after seeing the measurements below.

• French fils loses its exclusion and takes a generation reading beside its family reading. In a mixed-case name, lower case makes it the generation and Title case leans it to the name (rules/S6's new bullet).

• default = false now marks a reading that is recorded and checked but compiles to nothing, for a language pack to turn on in phase 3. It applies to Spanish h (hijo), German i for the regnal ordinal written I., and Czech st, whose exclusion it replaces. Each of these is written exactly as an initial is, so nothing in its writing tells the two readings apart.

• Reserved names. attaches stays reserved for phase 2, and prefer and strict for phase 3. bare_only, leading_only, trailing_only and separators move to phase 4.

• New build checks, each with a mutation test:

  • default = true is refused.
  • An agnomen must have a name reading.
  • An agnomen may not be an acronym.
  • A name reading beside any other suffix word kind is refused, since no gate would read it.

• Readings:

  Input	Reads	1.4.0 through 2.6.0
  Pedro Alves Filho	family Alves, suffix Filho	middle Alves, family Filho
  Alexandre Dumas fils	family Dumas, suffix fils	middle Dumas, family fils
  Ana Neto	keeps family Neto	same
  PEDRO FILHO	keeps family FILHO	same
  Marie Fils	keeps family Fils	same

• Case repair writes a suffix fils in lower case, so the repaired name parses the same way again.

• Lexicon guard. A gated word listed as an unambiguous acronym now raises ValueError. Without it the word would bypass its gate. The v1 shim subtracts the overlap instead, so a config that 1.4 accepted never raises.

• Pickles. Lexicon gains two fields, so a Lexicon or Parser pickled by an earlier release no longer loads.

Measured

Every local check ran with --base against a checkout of master 225f3d6. Figures are counts and token shapes only.

Check	Moves
Committed snapshot	4 names: Pedro Alves Filho, Marie Fils (a report), and CLDR's French … de Bouchart fils and its family-first twin
cldr_agreement.py	4 fields newly agree, 1 still-wrong field changes
PSC (local_agreement.py psc-individuals)	23 fixed, 20 broken, 8 changed
tse-donors	21,059 of 848,351 filed names, all in capitals
ctgov-officials	110
fec-individuals	12 reports only (the names are all capitals)
tse filed/registry pairs	the 405 that differ only by a trailing agnomen: 405 agree, against 1 on master
tse spelling consistency	604 pairs now alike, 258 now apart

• PSC: every moved record is three words or more, so no two-word name moves.
  • The 23 fixed: the filer typed two surnames, the second being the agnomen, in the surname box.
  • The 20 broken: the filer typed only the agnomen there.
  • Filers split on where the agnomen's companion surname goes.
• tse:
  • 12,811 names had the agnomen as the whole family name, and now the word before it is the family.
  • 8,198 had a longer family name.
  • 50 gain a report only. None is left with no family name.
• ctgov: most of the 110 have the agnomen joining the credentials after a comma (Neto, MD, as Júnior, MD reads). 4 are reports only.
• tse spelling consistency: the lean spike gave 604 now alike and 304 now apart. Most of the 258 apart are pairs master read alike only because the agnomen was the family name in both spellings. local_corpora.py pairs tse-donors --agreement reports in whole percents, which do not move for these counts. decisions/S11 gives the recipe.
• Read by shape. The four comma shapes, the three name orders, one-case names, the agnomen as the whole family, after a particle, after an initial, behind a title, with a trailing title, and in a maiden clause. The accepted costs are in rules/S11 and decisions/S11:
  • An initial straight before an agnomen is left as the family name (Pedro J. Neto, 568 tse names). Most of those initials abbreviate the surname.
  • A particle straight before one is left as a family name of its own (1).
• The floor holds before the case cue (Derek's call): a lower-case fils in a mixed-case name is the suffix only where the name keeps a family word, and at the first slot after a comma only behind rules/C4's two name words. Dumas fils, Pedro Junior fils, , Dr. fils, Dumas, fils, Mr. and Mrs., fils and Sir Alexandre, fils keep a family word as master does; Alexandre Dumas fils, Alexandre Dumas, fils and Dumas, Alexandre fils read suffix fils.
• What the walk takes in front of a gated word is no word to spare (fixed before merge, Derek's call): the count looks past a suffix word, a numeral read as the generation, a member its writing makes a suffix, and a member any of those carries (rules/S7), and a suffix word there is no company. So Pedro Junior Filho, Pedro Júnior Neto, Pedro Jr. Filho, Pedro Junior Filho, MD, PEDRO JUNIOR FILHO, Pedro VI Filho, Pedro MA Filho, Pedro Junior Ma Filho and Pedro Junior do Filho read as master does, the agnomen the family name. The scan is one pass, guarded by new scaling rows (agnomens, carried_agnomens, and the frame-counted agnomen_front).
• Left for Should José Silva Júnior read Júnior as a suffix, as José Silva Junior does, and where do Filho, Neto and Sobrinho belong? #843: the filed abbreviations still read as the family name, so their pairs stay apart: FH 229, NT 119, FO 99 and smaller ones.

Docs

• Rules: new S11; S2, S5, S6, S7, C4, C12 and T5 amended.
• Decisions: new decisions/S11; dated entries in decisions/vocabulary-model; decisions/generational-words' Standing lines on the agnomens and fils replaced.
• The vocab README's kinds, properties, derived markers and reserved names, AGENTS.md, customize.rst, migrate.rst, locales.rst and the release log.

Checks

• uv run pytest and -n0 -m serial pass, and so do --python 3.15, mypy, ruff, the sphinx doctest build and tools/vocab/build.py --check.
• Reviewed by a code review with the sweeps and a design-docs review, then a second review of the fix commits. Every finding is fixed, each with a test that fails without it.

🤖 Generated with Claude Code