[derek73]

What

Strict-gap batch C, decided on 2026-11-03: ca.toml takes de, del and la, the particles of Catalan names (de la Torre, del Castell). It held the conjunction i alone, so strict ca read the rest as name words. Each entry copies es.toml's kind and properties.

Measured

• Strict losses (tools/differential/config_diff.py gaps on master f48c88a, over the records of Spain and Andorra).
  • Strict ca broke 626 records on de, 208 on del and 197 on la. The language's 1,032 broken records fall to 188 (simulated with all five batches).
  • It also loses fixes, 182 to 25: records, mostly Spanish names, where strict ca without the particles happened to agree with the filed boxes.
• Default and plain choices. The compiled default is unchanged. Over a grid of 16 Catalan names under the three orders (each comma shape; cased, upper and lower spellings; i joining two surnames; a title and a credential beside them), the default, plain ca and strict ca + es move no cell.
• Strict ca. It moves 35 cells, toward the default's reading, with the class batch B (Add the Spanish-derived particles of Filipino surnames to fil.toml (strict-gap batch B) #1031) met:
  • del and la hold a given reading in es.toml alone, so under strict ca they read as never-given particles.
  • Del Castell reads family 'Del Castell', where the default reads given 'Del' and reports particle-or-given.
  • _reading_index.strict_ungated's reviewed table gains ca: del, la.
• Committed corpora. The snapshot matches. Reading.languages of de and la now names ca, so the release log, usage.rst, _types.py and two tests list it.

decisions/vocabulary-model gains the batch's entry.

Review round

An independent review found no defect. It reproduced the figures (198 broken and 26 fixed on this change alone, 188 and 25 with batch A), found plain ca byte-identical over 61,582 local names, and read the 160 lost fixes: nearly all Spanish names whose filer boxed del Carmen with the given name, now read as plain ca and the default read them, with no Catalan name worse. Fixed after it:

• The decisions entry gives this batch's own figures and what the lost fixes are.
• It states the comma form of the never-given class (Castell, Del → family 'Castell Del').
• It notes dels, which no file holds, as left out.
• It gains a recompute recipe.
• STRICT_UNGATED's comment and the 3j entry date their counts.

Checks

• uv run pytest -q: 35,787 passed. uv run mypy, uv run ruff check.
• sphinx-build -b doctest and -W -b html; snapshot.py.
• Not run: -n0 -m serial and Python 3.15.

Merge note: #1031 (fil) and #1033 (si) edit the same lists of files for de (and #1031 for la), and #1031 and #1034 the same reviewed table. Whichever merges later takes every entry (('ca', 'es', 'fil', 'fr', 'it', 'nl', 'pt', 'si') for de once all three land) and rebuilds nameparser/config/_index.py. This batch's entry points to batch A's entry (#1030): merge #1030 first.

🤖 Generated with Claude Code