[derek73]

Under locales.ES and locales.ES_AR, rules/O6's given-name veto matched María Begoña López but not Maria Begona Lopez. An accent folds when the list is matched; ñ does not. Data from systems without the letter writes Begona, so the veto missed it.

The change is in the data. tools/es_given_names.py now writes a second tuple, _N_SPELLINGS, beside _RANKED:

• It holds each listed word written with ñ, again with n.
• A spelling is kept only where it passes the list's own share test, with given and surname uses summed over every spelling the n stands for.
• ES gains begona, inaki, benat and zurine. ES_AR gains those plus inigo, garbine, enaut, argine, eguzkine and ekine.
• martiño's n spelling is refused by construction: martino is 24.0% given once summed, against ES_AR's 50% share, because Martino is mostly a surname.

Neither the list's matching (given_name_key) nor _normalize changes. The fold therefore runs one way: a surname written with ñ (Oña, Mariño, Niño) never matches an entry written with n.

parser_for(...)	before	after
ES, Maria Begona Lopez	family Begona Lopez	middle Begona, family Lopez
ES_AR, Jon Inigo Perez	family Inigo Perez	middle Inigo, family Perez
ES, Juan Oña Pérez	family Oña Pérez	unchanged (ona is listed, oña is not)
ES_AR, Luis Martino Gil	family Martino Gil	unchanged (refused by the share test)

Measured

Measured against master 2d91f6b, under default, plain:ES and plain:ES_AR, with config_diff.py diff --readings compared line by line between the two trees:

• All 13 local corpora, restricted to names that could move (those holding ñ, an n spelling, or a salutation word the declined fold would reach): 12 names move, each an n spelling leaving the family: 2 under ES, 10 under ES_AR, none by default.
• Checked against the filers' boxes: 6 land in a given box; 3 are Inigo filed as a surname. Iñigo is also a surname, but it passes ES_AR's share at 78.8%, and ES lists no iñigo.
• No other change: the committed corpora, every cldr_agreement.py view, and the full runs of tse-donors, dof-signatures, ukec-individuals and ORCID's six Spanish-speaking countries are all unchanged.
• Score: config_diff.py score over ORCID and PatentsView in those six countries fixes one PatentsView Spain record under ES and moves nothing else.
• Same as the prototype: the built change gives readings byte-identical to the prototype measured before the decision.

Declined, with numbers in decisions/O6:

• Folding ñ in _normalize. Senor and Senora become titles by default. Under ES and ES_AR, Dona becomes a title too, which pt.toml refuses because it is a given name: 258 names a pack. es.toml's spellings have to be rewritten, and 3.0 pickles holding an ñ edit stop loading.
• A two-way list fold. It moves 28 (ES) and 149 (ES_AR) surnames written with ñ out of the family, plus 20 Martino; in ORCID, 135 of the 162 moved words are filed in the family box, 27 in a given box. Its fixes are the same as this change's.

Docs

• rules/O6: one clause in the "What is taken must be one name word…" bullet, plus two examples. tests/v2/rules_budget.json is re-recorded for it (870 → 903 words); the clause reads no new signal, only the list it already read.
• decisions/O6: the Open line becomes a Standing line, with a dated entry holding the Declined and Excluded items and a recompute recipe. The tool's new --words flag, and the n rows it prints with summed uses, produce every quoted figure.
• User docs: a locales.rst paragraph with doctests, a sentence in customize.rst, and a sentence plus :issue: link in the release log's ES bullet. It goes there rather than in a bullet of its own because both packs are new in 3.0.
• Corpus files: the rules corpus and snapshot gain the two example names.

Checks

• Suites on 3.11 and 3.15, the serial benchmarks, mypy and ruff.
• sphinx-build -b doctest, -W -b html, and python -m doctest README.rst.
• A design-docs review and a code review of the first commit. Their findings are fixed in the second commit: the share condition is added to the rule clause, the tool's tallies replace raw INE counts, and a test covers the given-side sum. A review of that fix commit found three more problems, fixed in the third: the ona/rene shares are restated per list, --words gives the remaining counts, and refused rows go to stderr. All figures in the third commit were re-checked by running the commands it adds.

Closes #1081

🤖 Generated with Claude Code