[derek73]

Removes the 736 familiar-form pairs copied from Jörg Michael's nam_dict (GFDL) before any release ships them, keeping the ODU nicknames list's English pairs. This carries out the rule recorded in #1370: a compiler's list copied whole keeps its licence, and the core will not carry the GFDL (decisions/shipped-data-licences, 2026-11-12 #1370 FACTS WE MEASURE).

Rebased onto master after #1370 merged.

What changes

• tools/vocab/familiar.py writes the nicknames list alone into en.toml. The nam_dict reader, the per-country placement, the bands notes and the de-duplication against nam_dict are removed.
• 563 pairs leave, the ones the nicknames list doesn't hold, across 20 vocabulary files. mt.toml and sq.toml held nothing else, so they are deleted, and hu.toml now holds exclusions only. FAMILIAR goes from 2,743 pairs of 2,665 names to 2,180 pairs of 2,015 names.
• familiar-form is English-only until other languages' familiar forms are rebuilt from pairs found in CC0 data and reviewed by hand. That rebuild is a separate follow-up PR.
• LICENSE-nam_dict's header and README's License bullet now name only _order_names.py's _COMMON_GIVEN block. GFDL-1.2-no-invariants-or-later stays in license, license-files and _THIRD_PARTY. Whichever merges second, this PR or the one replacing _COMMON_GIVEN from ORCID, deletes the file and those three entries.
• Vocabulary notes that quote a nam_dict frequency band as a bearer count stay. A band is a measured fact, not a copied list.
• Decisions:
  • name-matching gets a new entry, 2026-11-12 ENGLISH ONLY UNTIL REBUILT. It amends the Standing line and marks the 2026-11-08 and 2026-11-09 entries as superseded in part.
  • shipped-data-licences gets a Log entry recording the drop.
  • group-suggestion's Standing clause is corrected.
• Docs and the release log: docs/matching.rst, docs/customize.rst, docs/locales.rst, the vocab README and root AGENTS.md now use English examples. The Parser.matching release bullet says the shipped pairs are English only.

Measured

match_diff.py --base #1370's tip (848c00c), 31,212 pairs × 21 settings:

• The key guarantee holds: no consistent pair is left without a shared key.
• 4,444 of 655,452 rows move: 44 under each of the 17 settings that don't accept familiar-form, and 919–929 under each of the four that do.
• 99 rows go from consistent to inconsistent, all under settings accepting familiar-form (Masha/Maria, Józsi/József). 924 given rows read conflict where they read familiar-form. Keys move on 3,532 rows.

No parse moves, since no parse reads a familiar reading.

173 of nam_dict's pairs are also in the nicknames list, and they now ship as the list's readings. The list's review read only the 2,624 rows nam_dict didn't already hold. Both this change's reviewer and I read those 173 by hand and found none in a refused category.

Tests

• Full suite: 46,796 passed. ruff, mypy, sphinx-build -b doctest and the README doctest are clean.
• test_familiar_tool.py is rewritten for the list alone. A new test checks that every shipped familiar reading is the list's and sits in en.toml.
• test_match.py plants its Hungarian accent pairs in a caller's lexicon instead of relying on shipped pairs.
• A new planted-index test covers _reading_index.tags()'s filter for a file of familiar readings alone, whose last shipped examples were mt.toml and sq.toml. Negative control: it fails with the filter removed.

Reviews run

• design-docs-reviewer on the docs/design and AGENTS.md changes: five findings, all fixed.
• pr-review-toolkit code-reviewer on the change: one coverage gap and stale prose, all fixed.
• code-reviewer on the fix commit: no defects at 80 or above; three minor items applied.

🤖 Generated with Claude Code