derek73 · Nov 13, 2026 · Nov 13, 2026 · Nov 13, 2026 · Nov 13, 2026 · Nov 13, 2026
diff --git a/AGENTS.md b/AGENTS.md
diff --git a/LICENSE-nam_dict b/LICENSE-nam_dict
diff --git a/README.rst b/README.rst
@@ -302,9 +302,6 @@ under other licences, which apply to those parts:
 
 - the familiar forms from Old Dominion University's nicknames list:
   Apache License 2.0, in ``LICENSE-nicknames``;
-- the group suggester's common given names from Jörg Michael's
-  nam_dict: GNU Free Documentation License 1.2 or later, in
-  ``LICENSE-nam_dict``;
 - ``locales.ES`` and ``locales.ES_AR``'s given names from Spain's
   Instituto Nacional de Estadística: Creative Commons Attribution 4.0,
   credited in ``nameparser/locales/_es_given_names.py`` and
@@ -314,7 +311,7 @@ under other licences, which apply to those parts:
   ``nameparser/locales/zh.py``.
 
 The package's licence expression is
-``LGPL-2.1-or-later AND Apache-2.0 AND GFDL-1.2-no-invariants-or-later AND CC-BY-4.0 AND OGDL-Taiwan-1.0``.
+``LGPL-2.1-or-later AND Apache-2.0 AND CC-BY-4.0 AND OGDL-Taiwan-1.0``.
 
 .. |Build Status| image:: https://github.com/derek73/python-nameparser/actions/workflows/python-package.yml/badge.svg
    :target: https://github.com/derek73/python-nameparser/actions/workflows/python-package.yml

diff --git a/docs/design/decisions/group-suggestion.md b/docs/design/decisions/group-suggestion.md
diff --git a/docs/design/decisions/shipped-data-licences.md b/docs/design/decisions/shipped-data-licences.md
@@ -6,12 +6,13 @@
 - A dataset under CC0, a corpus the package measured counts on without copying its words, and a source that states no terms for reuse add nothing to the expression; the last kind is listed in this file's Log as left out, by name, so the next source of its kind is weighed against it — 2026-11-12 #1370 THE LICENCE EXPRESSION.
 - Data the project derives by measuring (counts, rankings, cutoffs and filters of its own over its sources) is the project's own, its sources credited as a courtesy; a curated list copied as a whole, one compiler's judgements of which entries belong, keeps its licence, and a list the core will not carry under that licence ships as an optional extra package, as the Japanese segmenter does, or not at all — 2026-11-12 #1370 FACTS WE MEASURE.
 - The familiar forms copied from nam_dict are dropped before 3.1, keeping the ODU list's (done, 2026-11-12 #1370 NAM_DICT'S FAMILIAR PAIRS LEAVE THE PACKAGE), and rebuilt from pairs the project finds in CC0 data and reviews by hand (not yet done) — 2026-11-12 #1370 FACTS WE MEASURE.
+- No shipped data derives from nam_dict since 2026-11-12, so the GNU FDL and `LICENSE-nam_dict` left the package before any release named them: its familiar pairs were dropped (#1370 NAM_DICT'S FAMILIAR PAIRS LEAVE THE PACKAGE) and the suggester's common given names are ranked from ORCID (CC0) — 2026-11-12 GIVEN NAMES FROM ORCID.
 - No shipped data derives from Unihan since 2026-11-12: `locales/zh.py`'s simplified forms come from the simplified-Chinese labels of Wikidata's family-name items and `_order_names.py`'s pinyin surnames from those items' Hanyu Pinyin statements, both CC0, so `Unicode-3.0` and `LICENSE-unicode` left the package; 2.5.0 through 3.0.1 shipped Unihan-derived forms under the expression naming the LGPL alone — 2026-11-12 UNIHAN REPLACED.
 
 ## Log
 
-- **2026-11-12 #1370 NAM_DICT'S FAMILIAR PAIRS LEAVE THE PACKAGE; `LICENSE-nam_dict` NAMES THE GROUP SUGGESTER'S COMMON GIVEN NAMES ALONE.** The first half of the FACTS WE MEASURE ruling below, carried out: the 736 familiar-form pairs `tools/vocab/familiar.py` wrote from nam_dict into 20 vocabulary files are removed, with the generator's nam_dict path, before any release ships them (decisions/name-matching, 2026-11-12 ENGLISH ONLY UNTIL REBUILT). `mt.toml` and `sq.toml`, which held nothing else, go with them. `LICENSE-nam_dict`'s header and README's License bullet now name `config/_order_names.py`'s `_COMMON_GIVEN` block alone, the one part of the package still derived from nam_dict until the suggester's common given names are drawn from ORCID; `GFDL-1.2-no-invariants-or-later` stays in `license`, `license-files` and `tests/v2/test_licensing.py`'s `_THIRD_PARTY` until then, and the change that removes the last of nam_dict's data deletes the file and the three entries. The counts in vocabulary notes that quote a nam_dict frequency band (`nam_dict band 5 in Sweden`) stay: a band is a fact the project measured, not a list it copied.
-- **2026-11-12 #1370 THE LICENCE EXPRESSION NAMES EVERY DATASET THE WHEEL SHIPS, AND NAM_DICT AND UNIHAN GET THE LICENCE FILES THEIR TERMS ASK FOR.** Until this entry `license` read `LGPL-2.1-or-later` alone and `license-files` held `LICENSE`, `AUTHORS` and `LICENSE-nicknames` (decisions/name-matching's 2026-11-09 THE NICKNAMES LIST). Five third-party datasets under stated terms reach the package, read from each module's provenance docstring and the tools that write the generated blocks, on master at bda4a632:
+- **2026-11-12 #1370 NAM_DICT'S FAMILIAR PAIRS LEAVE THE PACKAGE; `LICENSE-nam_dict` NAMES THE GROUP SUGGESTER'S COMMON GIVEN NAMES ALONE.** (The licence file it leaves superseded by 2026-11-12 GIVEN NAMES FROM ORCID below, which ranks those names from ORCID and deletes `LICENSE-nam_dict` and the GFDL term.) The first half of the FACTS WE MEASURE ruling below, carried out: the 736 familiar-form pairs `tools/vocab/familiar.py` wrote from nam_dict into 20 vocabulary files are removed, with the generator's nam_dict path, before any release ships them (decisions/name-matching, 2026-11-12 ENGLISH ONLY UNTIL REBUILT). `mt.toml` and `sq.toml`, which held nothing else, go with them. `LICENSE-nam_dict`'s header and README's License bullet now name `config/_order_names.py`'s `_COMMON_GIVEN` block alone, the one part of the package still derived from nam_dict until the suggester's common given names are drawn from ORCID; `GFDL-1.2-no-invariants-or-later` stays in `license`, `license-files` and `tests/v2/test_licensing.py`'s `_THIRD_PARTY` until then, and the change that removes the last of nam_dict's data deletes the file and the three entries. The counts in vocabulary notes that quote a nam_dict frequency band (`nam_dict band 5 in Sweden`) stay: a band is a fact the project measured, not a list it copied.
+- **2026-11-12 #1370 THE LICENCE EXPRESSION NAMES EVERY DATASET THE WHEEL SHIPS, AND NAM_DICT AND UNIHAN GET THE LICENCE FILES THEIR TERMS ASK FOR.** (The suggester's common given names in its nam_dict row, and ORCID's part in them, superseded by 2026-11-12 GIVEN NAMES FROM ORCID below: ranked from ORCID's given-name boxes, CC0; no release shipped the nam_dict-derived table.) Until this entry `license` read `LGPL-2.1-or-later` alone and `license-files` held `LICENSE`, `AUTHORS` and `LICENSE-nicknames` (decisions/name-matching's 2026-11-09 THE NICKNAMES LIST). Five third-party datasets under stated terms reach the package, read from each module's provenance docstring and the tools that write the generated blocks, on master at bda4a632:
 
   | Source | Shipped in | First release | Licence (SPDX) | What it asks | How it is met |
   |---|---|---|---|---|---|
@@ -34,3 +35,4 @@
 - **2026-11-12 UNIHAN REPLACED — THE SIMPLIFIED SURNAME FORMS AND THE PINYIN SURNAMES COME FROM WIKIDATA, AND THE UNICODE LICENCE LEAVES THE PACKAGE.** Of the two shipped derivatives the 2026-11-12 #1370 entry lists from Unihan, neither ships now; that entry's table stays the record of what 2.5.0 through 3.0.1 shipped. This settles FACTS WE MEASURE's Open item for `Unicode-3.0`: with no Unihan-derived data in the package, the term leaves the expression whatever that item decides for derived facts.
   - `locales/zh.py`'s simplified forms. `tools/zh_surnames.py` gives a ranked form the simplified form that the simplified-Chinese labels (`zh-hans`, `zh-cn`, `zh-sg`, `zh-my`) of the Wikidata items labelled with it give, where they give exactly one in the Basic Multilingual Plane, the rule Unihan's variant table was read by; the items are every family name (Q101352) and Chinese family name (Q1093580), at any rank, since the items for 關, 顏 and 遲 are family names but not Chinese family names. Of the 19 forms Unihan alone had added (the #1370 entry's list), 12 are attested so (东云兰关区娄盘练迟阙颜风) and 7 are not (买厐盖篮芈诸谌), whose traditional forms no family-name item labels (買, 龎, 籃, 諶) or whose items' simplified labels give no other form (蓋, 諸, 羋); the block holds 317 forms where it held 324. Measured over every name holding a Han character in the committed corpora and the local corpora, read under `plain:ZH` in this tree and in the tree before it (`tools/differential/config_diff.py diff --committed --local all --match '[\u4e00-\u9fff\u3400-\u4dbf]' --readings <file> default plain:ZH` in each, the two files compared line by line): 9 of 322,895 names moved, all transliterations of foreign names opening with 盖 (`盖尔哈尔德`, `盖达尔 阿利耶夫`), which the pack had cut after 盖 and which now read as the default reads them; `cldr_agreement.py`'s records, both locale views included, did not move. A Chinese name with one of the 7 surnames in simplified characters (`谌龙`, `盖丽丽`) is no longer divided; written in traditional characters (`蓋麗麗`) it still is. Weighed and not taken: crediting the form most of an item's bearers write their own simplified labels with (P734), which attests 盖 and 诸 but reads `马` for a European surname item labelled 蓋 and attests none of the other five.
   - `config/_order_names.py`'s pinyin surnames. `tools/vocab/order_names.py` takes every Hanyu Pinyin reading that the Wikidata items labelled with a `locales.ZH` single-character surname state, by three statements (P2440 qualified by P459 Hanyu Pinyin, Q42222; the P1721 qualifier of P1705; the P1705 of the item P460 names under the criterion P1013 Q42222), and leaves out a surname no item gives one for, naming each in its run: 35 of 463 on 2026-11-12 (偕储儲兵农刁力塗奚安宫宮寇浦留瞿祝籃繆缪芦蘆褚諶諸買辜農鄞闕阙髙龎𡍼𨶒). The block holds 200 spellings where it held 206: `bu`, `nian`, `shan`, `zeng`, `zha` and `zhai` joined, the surname readings of 卜, 粘, 单, 曾, 查 and 翟, which Unihan's first `kMandarin` reading gave as `bo`, `zhan`, `dan`, `ceng`, `cha` and `di`; `an`, `bing`, `bo`, `ceng`, `cha`, `chou`, `dan`, `diao`, `jie`, `kou`, `nong` and `que` left, the readings of the surnames left out and the readings that were not the surname's (`an` stays in the table through the Korean list). The suggester's figures moved as decisions/group-suggestion's 2026-11-12 PINYIN FROM WIKIDATA entry records.
+- **2026-11-12 GIVEN NAMES FROM ORCID — THE SUGGESTER'S COMMON GIVEN NAMES NO LONGER DERIVE FROM NAM_DICT.** `config/_order_names.py`'s `_COMMON_GIVEN` ranks ORCID's given-name boxes by the country each holder states (CC0) where it ranked nam_dict's country columns by frequency band; the filter over ORCID's two boxes is unchanged. The ranking, cutoff and filter are the project's measurement, the first kind FACTS WE MEASURE names, so the table carries no licence of its own. The table and its moves are decisions/group-suggestion's 2026-11-12 GIVEN NAMES FROM ORCID. No release shipped the nam_dict-derived table. With #1372's familiar pairs gone before it, no nam_dict data is left in the package, so this change deletes `LICENSE-nam_dict`, README's bullet, `GFDL-1.2-no-invariants-or-later` in `license` and the file in `license-files`, and `tests/v2/test_licensing.py`'s `_THIRD_PARTY` row; the expression is `LGPL-2.1-or-later AND Apache-2.0 AND CC-BY-4.0 AND OGDL-Taiwan-1.0`, and no release named the GFDL.
diff --git a/docs/getting-started.rst b/docs/getting-started.rst
@@ -87,7 +87,7 @@ it, and ``suggestion.base`` is the parser you started from:
     Suggested for 5 names (base: Parser(Lexicon(default), Policy())):
       name_order  FAMILY_FIRST  4 for, 0 against, 1 no evidence
           surname FAMILY_FIRST  3
-          given-name FAMILY_FIRST  1
+          given-name FAMILY_FIRST  2
       patronymic_rules  unchanged (none)  no evidence
       initials_as_family  unchanged (False)  no evidence
       capitals_as_family  unchanged (False)  5 for, 0 against, 0 no evidence
@@ -98,7 +98,8 @@ so they are evidence for leaving ``capitals_as_family`` off. Four names
 vote for family first: ``Kim``, ``Wang`` and ``Nguyen`` by
 the romanized surname in front, ``Garcia Emily`` by the common given
 name at the end (``Garcia`` is not one of the surnames ``suggest``
-reads). ``Taylor Morgan`` gives no evidence, so the suggested
+reads). ``Wang Xiaoming`` carries both kinds, so the surname and
+given-name rows count it once each. ``Taylor Morgan`` gives no evidence, so the suggested
 parser reads it family first only because the rest of the list is
 written that way; nothing in the name itself can confirm that reading.
 A name written the other way round that carries evidence (``Emily

diff --git a/docs/release_log.rst b/docs/release_log.rst
@@ -4,7 +4,7 @@ Release Log
 
     nameparser 3.1 is under development. It refuses a ``Lexicon``, ``Parser`` or locale pack pickled by 2.x, which 3.0 loads with a ``DeprecationWarning``.
 
-    - **Change the package's licence metadata to name the licences of the data it ships.** The ``License-Expression`` is ``LGPL-2.1-or-later AND Apache-2.0 AND GFDL-1.2-no-invariants-or-later AND CC-BY-4.0 AND OGDL-Taiwan-1.0``, where every release through 3.0.1 gave ``LGPL-2.1-or-later`` alone; the code stays LGPL-2.1-or-later, and ``LICENSE-nam_dict`` joins ``LICENSE-nicknames`` in the wheel. README's License section names each source. :issue:`1370`
+    - **Change the package's licence metadata to name the licences of the data it ships.** The ``License-Expression`` is ``LGPL-2.1-or-later AND Apache-2.0 AND CC-BY-4.0 AND OGDL-Taiwan-1.0``, where every release through 3.0.1 gave ``LGPL-2.1-or-later`` alone; the code stays LGPL-2.1-or-later. No release shipped data derived from nam_dict (GNU FDL), which the 3.1 cycle's familiar forms and suggester table held until the familiar pairs were dropped and the table was ranked from ORCID. README's License section names each source. :issue:`1370`
     - **Change locales.ZH's simplified surnames to the ones Wikidata's family-name items write, no longer Unihan's.** ``parser_for(locales.ZH).parse("谌龙")`` gives family ``谌龙``, undivided, where 2.5.0 through 3.0.1 gave family ``谌``, given ``龙``; seven simplified forms no Wikidata item attests (买, 厐, 盖, 篮, 芈, 诸, 谌) left the list, and their traditional forms (``諶龍``) still divide. 2.5.0 through 3.0.1 shipped those forms from Unihan, whose Unicode License v3 their metadata did not name; 3.1 ships no Unihan-derived data. :issue:`1370`
     - **Remove loading a Lexicon, Parser or locale pack pickled by 2.x, which 3.0 deprecated.** ``pickle.loads`` of a ``Parser`` pickled by 2.8.0 raises ``ValueError`` saying to re-pickle it under 3.0, where 3.0.x loaded it with a ``DeprecationWarning``; the error for a 2.x ``Lexicon`` spelling now names :doc:`migrate`'s table rather than each word's calls. :issue:`1003`
     - **Change PolicyPatch.script_orders to merge by script.** ``Policy().patched(PolicyPatch(script_orders={Script.HAN: GIVEN_FIRST}))`` keeps the Hangul and hiragana entries beside its Han one, where 2.1.0 through 3.0.1 replaced the whole table with the patch's; an empty patch no longer clears the table, which ``Policy(script_orders=())`` on the base parser does. A pack can now add one script's order without erasing a caller's own, and ``parser_for`` warns where two packs set one script differently. :issue:`1237`

diff --git a/docs/suggest.rst b/docs/suggest.rst
@@ -59,7 +59,8 @@ votes do not decide keeps the base parser's value.
 
 Four names vote for family first: ``Kim``, ``Wang`` and ``Nguyen`` by
 the romanized surname in front, ``Garcia Emily`` by the common given
-name at the end. ``Taylor Morgan`` gives no evidence: it is read family
+name at the end; ``Wang Xiaoming`` carries both, so each signal's row
+counts it. ``Taylor Morgan`` gives no evidence: it is read family
 first because the list is, and nothing in the name can confirm that.
 
 .. _suggest-disagreeing:
@@ -77,7 +78,7 @@ written, so it takes ``Clark`` for the given name:
     Suggested for 6 names (base: Parser(Lexicon(default), Policy())):
       name_order  FAMILY_FIRST  4 for, 1 against, 1 no evidence
           surname FAMILY_FIRST  3
-          given-name FAMILY_FIRST  1
+          given-name FAMILY_FIRST  2
           given-name GIVEN_FIRST  1
       patronymic_rules  unchanged (none)  no evidence
       initials_as_family  unchanged (False)  no evidence
@@ -179,10 +180,11 @@ a name that writes no capitals disagrees with a decided switch:
 
 Where it stops. A group is read right only once it carries enough
 names with evidence. Measured on groups drawn from real names written
-in a known order (2026-11-10, re-taken 2026-11-11), a list written in the base's own order keeps it
+in a known order (2026-11-10, re-taken 2026-11-11 and 2026-11-12), a list written in the base's own order keeps it
 in at least 98% of groups from 2 to 200 names, but moving the base
 takes more names, from 5 for Chinese, Taiwanese, Korean, Czech and
-Polish names to 200 for Indian ones. Two lists keep their order less
+Polish names to 50 or 200 for Indian ones (US patent inventors and UK
+company filers). Two lists keep their order less
 often, the capitals in them read as a family name: bylines written with
 the initials first (``GH Yoon``), so a group of ten or more is usually
 suggested family-first, and names written with the given name in

diff --git a/nameparser/_parser.py b/nameparser/_parser.py
@@ -362,27 +362,30 @@ def suggest(self, names: Iterable[str]) -> Suggestion:
         an empty group, or one with no votes, suggests no change. A
         bare str, bytes, a mapping or a non-str item raises TypeError.
 
-        Measured 2026-11-10, re-taken 2026-11-11, with
+        Measured 2026-11-10, re-taken 2026-11-11 and 2026-11-12, with
         tools/differential/suggest_measure.py
         on groups drawn from real names written in a known order, 300
         groups a size (decisions/group-suggestion's 2026-11-11 TRAILING
-        INITIAL entry holds the table): a group written in this parser's own order kept
+        INITIAL entry holds the table, its 2026-11-12 GIVEN NAMES FROM
+        ORCID entry the cells that moved since): a group written in this parser's own order kept
         it in at least 98% of groups at every size from 2 to 200 names,
         under either base order, save two populations whose capitals the
         capitals signal reads as a family name: ORCID's credit
         names written as a byline with the initials first (``GH Yoon``)
         kept their given-first order in 96% of groups of two and 0% of
         groups of fifty, before this gate as after it, and ORCID holders
         who typed their given name in capitals (``ANNA Smith``) kept it
-        in 27% of groups of two and none of five (decisions/
-        group-suggestion's 2026-11-12 CAPITALS entry). Moving the base takes more names, and
+        in 30% of groups of two and none of five (decisions/
+        group-suggestion's 2026-11-12 CAPITALS entry, re-taken in its
+        GIVEN NAMES FROM ORCID entry). Moving the base takes more names, and
         how many depends on what the names carry: groups of Chinese,
         Taiwanese, Korean, Czech or Polish names (the romanized surnames
         and the common given names) were read right in 95% of groups
         from 5 names; German, Dutch, Italian, Spanish, British and
         Vietnamese from 10; French, US, Japanese and Turkish from 20;
         Nigerian, Romanian, Russian and Brazilian from 50; and Indian
-        only at 200. A group whose votes fall short of the thresholds keeps
+        from 50 for US patent inventors, only at 200 for UK company
+        filers. A group whose votes fall short of the thresholds keeps
         the base rather than guess, so a smaller group moves it in fewer
         of the draws. ``capitals_as_family``, measured 2026-11-12
         (decisions/group-suggestion's CAPITALS entry): ORCID holders who