derek73 · Nov 13, 2026 · Nov 13, 2026 · Nov 13, 2026 · Nov 13, 2026 · Nov 13, 2026
diff --git a/README.rst b/README.rst
@@ -152,7 +152,7 @@ More shapes the parser reads:
     >>> ivan.given, ivan.patronymic, ivan.family
     ('Иван', 'Петрович', 'Сидоров')
     >>> locales.available()
-    ('bn', 'cs', 'es', 'es_ar', 'ga', 'gd', 'hi', 'hr', 'ja', 'mn', 'ne', 'pt', 'pt_br', 'ru', 'sk', 'sl', 'tr_az', 'vi', 'zh')
+    ('ar', 'bn', 'cs', 'es', 'es_ar', 'ga', 'gd', 'he', 'hi', 'hr', 'ja', 'mn', 'ne', 'pt', 'pt_br', 'ru', 'sk', 'sl', 'tr_az', 'vi', 'zh')
 
 When a reading is wrong for your data
 -------------------------------------

diff --git a/docs/design/decisions/family-without-an-inherited-name.md b/docs/design/decisions/family-without-an-inherited-name.md
@@ -6,7 +6,7 @@
 - The patronymic view keeps rules/R9's basis, filled only by a word or a rule that says patronymic, never by a position; it may overlap the family or be a part of it, and a father's name nothing in the text marks (Ethiopia) fills no view — 2026-11-12 #1253.
 - An Ethiopian or Eritrean name stays positional, its family under the default given-first order the grandfather's name where it is written and the father's where it is not; a caller matching such a list builds the matcher from a parser with `middle_as_family`, keeping the default parser for the fields (docs/matching.rst) — 2026-11-12 #1253.
 
-Open: #1188 (Kazakh), #1155 (`بن` chain), #1231 (Burma), each a value question under the two statements; #1210 (Hebrew `בן`), decided with #1257; #1165 (`S/O`, `W/O`), a decision of its own.
+Open: #1188 (Kazakh), #1155 (`بن` chain), #1231 (Burma), each a value question under the two statements; #1165 (`S/O`, `W/O`), a decision of its own.
 
 ## Log
 
@@ -19,7 +19,7 @@ Open: #1188 (Kazakh), #1155 (`بن` chain), #1231 (Burma), each a value question
     - #1187 Kyrgyz: with no surname, the father's name and its marker are the `family` and the patronymic view alike (the reading lands with #1187, in rules/O2).
     - #1188 Kazakh: an ending rule fills the view, as the East Slavic one does; which choice turns it on is a value question.
     - #1155 a family name after a `بن` chain: the view may be a part of the family, stopping before a family name a closed signal marks; which signal is a value question.
-    - #1210 Hebrew `בן`: decided with #1257.
+    - #1210 Hebrew `בן`: decided with #1257, a clan particle filling no view (decisions/vocabulary-collisions' 2026-11-12 #1257 entry).
     - #1231 Burma: a pack, the last word the `family` by position, the honorifics titles.
     - #1165 `S/O`, `W/O`: out of this decision; the words after the marker are another person's name, which the model has no place for. Malaysian `a/l` and `a/p`, where the father's name is part of the legal name, read as `bin` does and are a vocabulary question.
   - **Measured for #1203** (dated snapshots):

diff --git a/docs/design/decisions/group-suggestion.md b/docs/design/decisions/group-suggestion.md
@@ -154,6 +154,7 @@ Open: bylines written with the initials first (`GH Yoon`), which the capitals si
     | az_parliament (124), kz_mazhilis (96, Cyrillic), tr_parliament (592), the whole list | decided: 124 for / 0 against, 92 / 0, 565 / 0 | — |
 
     The East Asian row reaches 95% at 10 names, the whole famcaps population at 3 on the threshold (97.0% given-first, 95.0% family-first) and clear of it at 4: the short families are no evidence, and the floor is kept for them. The user-facing claims say "from 4 names" and "from 10".
+    Re-taken 2026-11-12 after #1257 made `ait`, `nait`, `ould` and `ouled` never-given particles and `el` an article that may be given (`suggest_measure.py --base <a checkout of 161ebd28>`, 300 groups a size, seed 20261110): five cells moved, four of the capitals table and one of the order table's, its orcid-givcaps anti-goal row, a list writing its given names in capitals that the capitals signal reads the wrong way round by design. (1) orcid-famcaps, family-first written, given-first base, deciding `capitals_as_family` at 2 / 3 / 4 / 5 names: 77.3 / 95.0 / 99.0 / 99.7 → 77.0 / 95.0 / 99.0 / 99.7; (2) the same written order on a family-first base, the same figures; (3) orcid-famcaps' families read as the box where decided: 352,086 names scored, 351,382 right with the switch and 312,999 without (99.8%, 88.9%) → 352,082, 351,378 and 316,781 (99.8%, 90.0%), a family written in capitals after the given name that opens with one of the particles (`Youssef AIT HADDOU`, `Sara EL HADDOU`) now reading right without it; (4) that anti-goal row, given-first written on a given-first base, groups suggested at 2 / 3 / 4 names: 26.7 / 8.3 / 3.0 → 26.7 / 8.7 / 3.0; (5) orcid-givcaps' families read as the box: 351,802 names, 1,704 right with the switch and 4,256 without → 1,704 and 4,260. The user-facing claims stand.
   - Variants declined, a dated snapshot of 2026-11-12 on master bda4a632, taken with `capvote.py` (the 3.1 cycle's session notes, `659-pr-b/`, not the repository: the vote as a function outside the library, before the title and lower-case clauses of #1369's review, run over suggest_measure's populations and ORCID's, 300 groups a size, seed 20261110):
 
     | population (capvote.py) | as shipped (L4, n2, > 2/3) | L4, n3 | L2, n2 |

diff --git a/docs/design/decisions/name-matching.md b/docs/design/decisions/name-matching.md
@@ -158,3 +158,4 @@ Open: a family name of initials alone against the same initials in another order
 - **2026-11-09 #1286 A BIRTH NAME OF PARTICLES ALONE IS KEYED AS THE CHECK COMPARES IT.** Under `across-roles`, `Jack de` / `Jack Smith (née de)` and `Jack B. al` / `Jack S. (née al)` read consistent, the family matched against the birth name whole, but shared no key: the keys added the birth name's base, its text less its particles, which is empty there (`de;j` against `smith;j`). Of #1286's two ways, the keys taking the text whole, particles kept, as the check compares it, or the check refusing a cross-role match of particles alone, #1261's keys already do the first (its keys entry above: under `across-roles`, each text the check compares the family with, whole), so the pair now shares `de;j`. The second way, refusing a cross-role match of particles alone, is declined below.
   - Measured, 2026-11-09: on master at 0e4bd46c, before #1261, each of `test_a_birth_name_of_particles_alone_is_keyed_as_the_check_compares_it`'s six rows reads consistent and shares no key, and from b23f5fcc on each shares one. A grid under `across-roles` of 16 particles and particle runs (`de`, `al`, `van`, `von der`, `de la`, `da`, `el`, `bin`, `ibn`, `di`, `du`, `le`, `la`, `ter`, `van den`, `al-`), each written `P` as a family in four shapes (`Jack P`, `P, Jack`, `Jack B. P`, `Jack P Jr.`) against a birth name in four (`Jack Smith (née P)`, `Jack Smith née P`, `Smith, Jack (née P)`, `Jack S. (née P)`), both ways, 512 pairs, all consistent: 480 share no key at 0e4bd46c and none at e1ae8086.
   - Declined (maintainer, 2026-11-09): refusing a cross-role match whose text is particles alone, #1286's second way, which it called the more conservative. The particle list cannot tell a clipped `de` from a surname that is also a listed particle, and in real names a family name of particles alone is mostly such a surname: in up to 300,000 distinct names per local corpus (each corpus's names deduplicated, then shuffled by `random.Random(1286)` and cut; jrc-names' Latin-script spellings only, tse-donors' filed and registry names both), each parsed by `parse()`, a text being particles alone where each of its folded words, hyphens stripped, is in `Lexicon._sets.particles`, the commonest are `das` (570 in nepal-candidates), `le` (190 in orcid-researchers, 61 to 75 in ctgov-officials, patentsview-inventors and psc-individuals), `du` (95 to 103 in ctgov-officials and patentsview-inventors), `do`, `graf`, `van`, `bar`, `di` and `prinz`, only tse-donors' `de`, `da` and `dos` reading as clipped names. Refusing would read `Mary Das` / `Mary Smith (née Das)` a conflict. And it would protect against no pair the data holds: birth names are rare there (385 in 2,437,885 names, each source's deduplicated: thirteen local corpora sampled as above, and every distinct name in `tools/differential/snapshot.jsonl`, 119 of its 6,982), one of them particles alone, `Smt. Bandana Dome nee Das`, the Bengali surname. The mutant is master at e1ae8086 with one test added to `Matching._across`: a match is skipped where either text is particles alone, as above, and the role is the family. `match_diff.py --base <master at e1ae8086> --tree <the mutant>`, run from master at fd5bed17 so its draw holds #1290's rows, moved 76 of 654,066 rows (19 pairs under each of the four settings with `across-roles`), every one a constructed `Jack` pair of the test above, family `across-roles` → `conflict`, and no corpus pair; the same mutant with the role test dropped, refusing a particles-alone text in every cross role (a given name against a middle name or a nickname too), moved the same 76.
+- **2026-11-12 #1257 THE NEW PARTICLES MOVE MATCHING THROUGH THE PARSE ALONE.** #1257 reads `el`, `ait`, `nait`, `ould` and `ouled` as particles by default and `אבו` and `אל` in Hebrew script, and changes no matching code; the pairs move because the family now holds the particle. `NAMEPARSER_TOOLS_CACHE=<a folder of its own, the corpora linked in> uv run python tools/differential/match_diff.py --base <a checkout of the merge base 73718fd9>` (2026-11-12) moved 1,029 of 655,452 rows, 49 pairs under each of the 21 settings, every one holding `el` or `El`: 22 pairs become consistent, two spellings of one person writing the article as `El` and `al` or with and without its hyphen (`Abd el-Fatah` / `Abd el Fatah`), and one becomes inconsistent only in the setting that takes `particles-aside` out of the accept set (`Ahmed Nashar` / `Ahmed El Nashar`, read alike by default through that outcome); the rest move a role's outcome or the keys alone (`Sara El Haddou` keyed with its particle). No consistent pair shares no key.
diff --git a/docs/design/decisions/v1-xfail-triage.md b/docs/design/decisions/v1-xfail-triage.md
@@ -8,7 +8,7 @@ The reconciled v1-style banks (`tests/test_*.py`) carried eight `@pytest.mark.xf
 
 - A trailing `I` before a further comma stays a middle initial, the commoner reading: `Maier, Amy I, Jr.` reads middle 'I', suffix 'Jr.', where `Maier, Amy I` reads suffix 'I' (rules/C9) — 2026-09-01 (Derek's triage)
 - `Dr King Jr` reads title 'Dr', family 'King', suffix 'Jr', and `King, Dr Jr` reads the same fields, with no report since `king` became a given name outside a title phrase (`Dr Pope Jr` keeps H3's floor and its report) — 2026-09-01 (Derek's triage); 2026-10-26 #752, phase 2j
-- `ben` stays out of the particles, colliding with the given name Ben where the particle claim acts: `Ahmad ben Husain` reads middle 'ben', family 'Husain' — 2026-09-01 (Derek's triage)
+- `ben` stays out of the default's particles, colliding with the given name Ben where the particle claim acts: `Ahmad ben Husain` reads middle 'ben', family 'Husain' (an Arabic or Hebrew choice reads it as a clan particle written `default = false`, `locales.AR` giving family 'ben Husain', decisions/vocabulary-collisions' 2026-11-12 #1257 entry) — 2026-09-01 (Derek's triage)
 - An input that is all titles takes its last title word as the name, a convention accepted rather than defended: `The Right Hon. the President of the Queen's Bench Division` reads family 'Division' — 2026-09-01 (Derek's triage)
 - The one surviving marker is `E.T. Smith`'s, cited by issue (#490 until 2026-11-09, #1296 since): dotted initials against dotted vocabulary; `U.S. District Judge Marc Thomas Treadwell` was fixed on 2026-11-09 (decisions/H10); #489's `Her Majesty Queen Elizabeth` (given 'Elizabeth') was fixed on 2026-09-08 — 2026-09-01
 - `juan garcia III` repairs to `Juan Garcia III` and its test is an ordinary pin — 2026-09-23 #492.