Should Maria das Graças Silva read middle das Graças, family Silva, as María del Carmen García does? #1118

rules/P10 (a given name a particle completes) reads a Marian or devotional phrase behind a one-word given name as a middle name: María del Carmen García → middle del Carmen, family García. The words it reads are 23 marked after_particle, all in es.toml. A few of them also cover Portuguese names (do Socorro, de Fátima, de Lourdes).

The Portuguese-only phrases still read as the start of the surname:

Why they were left out. #1057 measured Portuguese-speaking countries on ORCID, capped at 20,000 records a country. It compared the given box ending in the phrase against the family box opening with it. No Portuguese word reached the bar of at least 90% given with ten or more given records: carmo 21 to 6, rosário 21 to 8, conceição 25 to 25 (decisions/O7). Graças, Dores, Glória and Penha weren't in that measurement.

Why look again. The --boxes index now covers the whole ORCID extract (888,163 Brazilian and 103,654 Portuguese records), not a capped sample. These word-level counts (the word anywhere in the box, not yet the phrase shape) show some given-leaning candidates:

(tools/name_usage.py --no-wikidata --country BR --country PT --json, 2026-11-06.) These are not the phrase measurement #1057's bar uses, so they show which words to measure and decide nothing.

Proposal, in rule shape

• When: a Portuguese given name of one word is followed by a particle run and a word that a Portuguese reading marks after_particle, in a name read given-first.
• Then: P10 reads the run and the word as a middle name, unchanged. Only the vocabulary moves: pt.toml gains given readings written after_particle for the words that pass.
• Because: Brazilian and Portuguese compound given names take the same devotional shape as Spanish ones (Maria das Graças, Maria das Dores, Maria da Glória), and filers who split the name put the phrase with the given names.
• Except: a word below the bar stays unmarked. carmo, conceição and penha are borne as surnames often enough that marking them by default would break José do Carmo Souza-shaped surnames.
• Accepted: a marked word borne as a surname loses it to the middle name, as de Fátima and del Valle already do (P10's Accepted).

Open questions

1. Re-measure at phrase level on the full extract, Portuguese-speaking countries only, with Read a particle phrase as the second given name; leave Kaur and Singh to neither two-name switch (#704) #1057's recipe (the given box ending in particle run + word, against the family box opening with it and holding more words).
2. Where the marking lives. One option is a default reading, which needs Read a particle phrase as the second given name; leave Kaur and Singh to neither two-name switch (#704) #1057's 90% bar. The other is a reading with default = false in pt.toml that only a Portuguese choice (locales.PT, locales.PT_BR) turns on, under a lower bar, since that caller has said the names are Portuguese. The second fits the opt-in-config principle, and it would let carmo and conceição in for a Portuguese caller without moving anyone else's parse.
3. Agreement and reach: local_agreement.py orcid-researchers --base for agreement, and snapshot.py --local tse-donors --base for how many Brazilian names move.

Not proposed: aparecida. It is mostly written without a particle (Maria Aparecida Silva already reads middle Aparecida), so marking it changes nothing. Also not proposed: neves, luz and anunciação, which lean to the family box at word level (11%, 16% and 13% given).