derek73 · Nov 3, 2026 · Nov 3, 2026 · Nov 3, 2026 · Nov 3, 2026 · Nov 3, 2026
diff --git a/...ns/R8-generation-and-credentials-views.md → ...ation-credentials-and-honorifics-views.md b/...ns/R8-generation-and-credentials-views.md → ...ation-credentials-and-honorifics-views.md
@@ -1,11 +1,12 @@
-# R8 — the generation and credentials views
+# R8 — the generation, credentials and honorifics views
 
 ## Standing
 
-- `ParsedName.generation` and `ParsedName.credentials` partition `suffix` by the reading each word took, a generation or an agnomen in the first and every other word in the second; they are views, not roles, so `suffix`, `as_dict()`, `replace()`'s fields and the repr are unchanged, item 4 of decisions/vocabulary-model's #754 output-model entry — 2026-10-29 #754, sub-phase 4b.
-- A listed suffix word's kind is the lexicon's, compiled with its sets (`Lexicon._sets.generation_words`, `credential_words`), which on the default is the shipped vocabulary's — 2026-11-02 #754, sub-phase 4i-b, superseding 4b's read from the reading index.
+- `ParsedName.generation` and `ParsedName.credentials` partition `suffix` by the reading each word took, a generation or an agnomen in the first and every other word in the second (the honorifics leaving the second for a view of their own since 2026-11-03, the bullet below); they are views, not roles, so `suffix`, `as_dict()`, `replace()`'s fields and the repr are unchanged, item 4 of decisions/vocabulary-model's #754 output-model entry — 2026-10-29 #754, sub-phase 4b.
+- The honorifics, a listed suffix word whose kind is `honorific` or `professional_honorific` (`さん`, `씨`, `先生`), leave `credentials` for a third view, `ParsedName.honorifics`, and read as `ReadingKind.HONORIFIC`; a courtesy (`Esq.`) and a status (`ret.`) stay in `credentials`, read as `post-nominal`, and the offices CLDR files as credentials (`議員`, `国会議員`, `太平紳士`) are credentials by kind — 2026-11-03 #326 (Derek).
+- A listed suffix word's kind is the lexicon's, compiled with its sets (`Lexicon._sets.generation_words`, `credential_words`, and since 2026-11-03 `honorific_words`), which on the default is the shipped vocabulary's — 2026-11-02 #754, sub-phase 4i-b, superseding 4b's read from the reading index.
 
-Elsewhere: the field's composition and the 2026-08-16 decision to keep it one field, which still stands, decisions/suffix-field-composition; whether `credentials` should leave the honorifics out, CLDR having no third field, is a value question under the RFC's Phase 4, Left open (docs/design/vocabulary-model-rfc.md).
+Elsewhere: the field's composition and the 2026-08-16 decision to keep it one field, which still stands, decisions/suffix-field-composition.
 
 Elsewhere: the gated credential words `kc`, `pc` and `dl`, a credential written as a word beside a name reading, read as credentials, decisions/S11's 2026-10-30 #965 entry.
 
@@ -19,6 +20,11 @@ Open: none in this file.
   3. **Each word is placed by its own reading, and separated as the writer separated it.** A `joined` mark is not only the second word of a split credential (`Ph.` `D.`, which takes the first word's reading) but every word of a suffix run written without commas (`Jr. III PhD`), so a view places each word by its reading. Two words of one view are separated by a comma where the suffix holds one anywhere between them, R1's separator (a typed comma, or a semicolon, hyphen, slash or delimiter C10 drops: `MD / Jr. / PhD` gives credentials 'MD, PhD'), and by a space where not, so `MD Jr. PhD` gives credentials 'MD PhD'; the first draft started a new comma-joined entry wherever a word of the other view stood between, writing a comma nobody typed ('MD, PhD'), found by the design-docs review. A connective goes with the word before it (`PhD and MD`, `Jr. and III`), so one joining a generation to a credential dangles at the end of the generation (`Jr. and MD` gives generation 'Jr. and'), accepted rather than given a rule of its own, no corpus name holding the shape.
   4. **A suffix word with no reading is a credential.** `replace()` and `Parser.revise()` write tokens with no `Token.reading` (4a), and a `ParsedName` unpickled from 2.x has none either, so `replace(suffix="Jr.")` reads credentials 'Jr.' and the 2.8 pickle of `Dr. Juan Q. de la Vega III` credentials 'III': nothing records what the word was read as, and the catch-all is the view that claims least. A name word a run took into the suffix by position is a credential too, 4a's reading of it (`Smith, Jr., John` gives credentials 'John').
   5. **The views stay out of `as_dict()`, `--json` and the repr**, as `family_base` and the other derived views do: `as_dict()` is the seven fields `replace()` takes back. They are `render()` keys and assertable fields of a rule's example lines.
-  6. **The snapshot records them** (`tools/differential/snapshot.py`, a `RECORDED_LATER` pair): `generation` where the suffix holds one, and `credentials` beside it where the suffix holds both, so a row with neither is a suffix of credentials only, or none. No other key moved on recording.
+  6. **The snapshot records them** (since 2026-11-03 it records `honorifics` too, so a row with neither generation nor credentials holds credentials and honorifics only, or none: the 2026-11-03 #326 entry below) (`tools/differential/snapshot.py`, a `RECORDED_LATER` pair): `generation` where the suffix holds one, and `credentials` beside it where the suffix holds both, so a row with neither is a suffix of credentials only, or none. No other key moved on recording.
   - Measured 2026-10-29 on the 4b branch over the committed snapshot (`tools/differential/snapshot.jsonl`, 6,570 readings, rules/R8's own examples included): 1,077 rows hold a suffix, 488 of them a generation, 142 a generation and credentials both. Recompute: count rows with `suffix`, `generation` and `credentials` in the snapshot. Over `cldr_fields.jsonl`, 576 distinct (text, order) renderings carry CLDR's `generation` or `credentials`; 65 parse to exactly CLDR's suffix words, and all 65 split into CLDR's two fields. Recompute: parse each distinct text under its order (`givenFirst` as `GIVEN_FIRST`, `surnameFirst` as `FAMILY_FIRST`, any other order the default), keep those whose `suffix` words, commas removed, equal CLDR's `generation` and `credentials` words together, and compare each view's words with its field's. The RFC's 615 and 75 (2026-10-28) counted renderings, not distinct (text, order) pairs.
 - 2026-11-02 #754, sub-phase 4i-b — A LISTED SUFFIX WORD'S KIND IS THE LEXICON'S. SUPERSEDES items 1 and 2 of the 4b entry above. A caller now states a word's kind (`add("secundus", kind="generation")`), and a view reading the index instead would ignore it in silence, so the compile gives the kinds the sets do not record, `GENERATION_WORDS` and `CREDENTIAL_WORDS`, and classify reads them off the lexicon (rules/R8 amended). On the default they are the index's by the compile invariant (`test_the_default_kind_sets_are_the_shipped_index_s`), so no default parse moves; one test's caller-configured reading did: `Lexicon.default().add("st", kind="suffix")` reads `St` as a post-nominal where 4b read the cs file's generation, written `default = false`. An agnomen needs a name reading, as the build's check 11 requires of a file, so `add(w, kind="agnomen")` alone is refused. decisions/vocabulary-model's 2026-11-02 #754, sub-phase 4i-b entry has the rest.
+- 2026-11-03 #326 (Derek) — THE HONORIFICS HAVE A VIEW OF THEIR OWN. #326's last open question was whether the honorifics after a name belong in `credentials`, CLDR's catch-all, CLDR having no third field. Three options were weighed: keep them in `credentials`; give them a third view and narrow `credentials`; or move them out of `suffix` into `title`, rules/W3's "no part of the name" argument. The third would move 2.x callers' `suffix` text, and the render default, which writes the title first, would render `김민준 씨` as `'씨 김민준'`, so it was declined. The second was chosen. It had to be decided before 3.0 ships `credentials`, the view being new there, so that narrowing it later is not a change to a released view; CLDR's `en` pattern writes `{surname} {generation}, {credentials}`, which with the honorific in the credentials would render `Kim Minjun, ssi` once a pack carries patterns. What was built:
+  1. **The kinds.** `honorific` (19 words on 2026-11-03: `さん`, `様`, `씨`, `님`, `氏`, `जी`, `साहब`, ...) and `professional_honorific` (11: `先生`, `教授`, `선생님`, `교수님`, `박사님` and the job titles `社長`, `課長`, `部長` with their simplified twins) compile into `HONORIFIC_WORDS`, `Lexicon._sets.honorific_words`, as `CREDENTIAL_WORDS` holds a credential written as a word (4i-b); classify tags such a word `vocab:honorific-word`, `_vocab.suffix_kind` reads it as `ReadingKind.HONORIFIC`, and assemble drops the tag from the published token, as it drops the other kind tags. `courtesy` (`esq`, `rinpoche`, `שליט"א`, `महाराज`) and `status` (`ret`, `vet`, `ז"ל`) stay `post-nominal` and in `credentials`: CLDR's own examples of its credentials field include `Esq.`, and a status says something about the bearer rather than addressing them. A courtesy said to the bearer (`Rinpoche`, `महाराज`, `মহারাজ`) stays there with `Esq.`, by its kind, accepted rather than re-kinded (rules/R8's Accepted).
+  1a. **The offices left the honorifics** (Derek, 2026-11-03, on the design-docs review's finding): the first build gave the whole `professional_honorific` kind (17 words) to the new view, and the rule's Because said CLDR's credentials field does not hold a form of address. `tools/differential/cldr_fields.jsonl` (CLDR release 48) holds both sorts: it files `先生` (12 renderings), `教授` (9), `さん` (3) and `님` (3) as `title`, a form of address, and `議員`, `议员`, `国会議員`, `太平紳士` and `太平绅士` (3 each) as `credentials` (`雅婷婷婷王二世先生，議員` credentials `議員`). So those six offices, the five and `國會議員`, the Diet member's traditional spelling, take the `credential` kind written as a word (`acronym = false`), as `博士` is, in `ja.toml` and `zh.toml` alike, a word having one suffix kind in every file (the build's check 4); their glued peel is the `attached` property's and is unchanged (`田中議員` suffix `議員`). The job titles `社長`, `課長` and `部長`, which CLDR's samples do not hold, stay `professional_honorific`, their notes calling each a form of address after a name. Recompute: for each word of `HONORIFIC_WORDS` and the six, count the fields of `cldr_fields.jsonl` whose value is the word or holds it as a comma-separated item.
+  2. **What moved.** No field of any parse: `suffix` keeps every word, and `as_dict()`, `replace()`'s fields and the repr are unchanged. In the committed snapshot (`tools/differential/snapshot.jsonl`, each name read under its declared order), 73 of 6,793 readings hold an honorific in the suffix and 21 of them other suffix words besides (`田中さん, PhD`: credentials `PhD`, honorifics `さん`), each moving its views only, and the snapshot records the new view where it is not empty (`RECORDED_LATER`). `Token.reading` of the 30 words reads `honorific` where it read `post-nominal`, and so does `Ambiguity.alternative` where a lone one reports the suffix-or-name fork: six on 2026-11-03, the Devanagari and Bengali `जी`, `साहब`, `साहिब`, `साहेब`, `বাবু` and `সাহেব`, which 2.7.0 and 2.8.0 named `post-nominal`. A caller's `add(w, kind="honorific")` puts the word in the view; `add(w, kind="courtesy")` keeps it in `credentials`.
+  3. **The rule.** rules/R8's Then names three views and its file is renamed for them; the decisions file follows it. Recompute the kind counts of item 1 by tallying `nameparser._reading_index.suffix_kind_of(w)` over `Lexicon.default()._sets.honorific_words`, and item 2's lone-word alternatives by parsing each word of that set alone and keeping the reports whose `alternative.kind` is `ReadingKind.HONORIFIC`. Recompute the snapshot counts: count the rows of `tools/differential/snapshot.jsonl` recording `honorifics`, and of those, the rows whose `suffix` holds words the `honorifics` value does not.
diff --git a/docs/design/decisions/README.md b/docs/design/decisions/README.md
@@ -127,7 +127,7 @@ Entry conventions:
 - [R5 — the case-repair gate](R5-case-repair-gate.md)
 - [R6 — the default rendering follows the written name](R6-default-rendering-follows-the-written-name.md)
 - [R7 — no initials in a script that writes none](R7-no-initials-in-a-script-that-writes-none.md)
-- [R8 — the generation and credentials views](R8-generation-and-credentials-views.md)
+- [R8 — the generation, credentials and honorifics views](R8-generation-credentials-and-honorifics-views.md)
 - [R9 — the patronymic view](R9-patronymic-view.md)
 - [R10 — the second-surname view](R10-second-surname-view.md)
 

diff --git a/docs/design/decisions/suffix-field-composition.md b/docs/design/decisions/suffix-field-composition.md
@@ -4,9 +4,9 @@
 
 - `suffix` stays one field holding generational words, credentials and postnominal honorifics; the composition is recorded, with the command that re-counts it, instead of splitting the field — 2026-08-16 #326.
 
-Elsewhere: generation and credential are separate kinds in the vocabulary, though they still compile into the one field, decisions/vocabulary-model's 2026-10-15/16 entry (item 3). The field stays one, and since 3.0 two views partition it by the reading each word took, `generation` and `credentials`, the honorifics in the second (#326, decisions/R8).
+Elsewhere: generation and credential are separate kinds in the vocabulary, though they still compile into the one field, decisions/vocabulary-model's 2026-10-15/16 entry (item 3). The field stays one, and since 3.0 three views partition it by the reading each word took, `generation`, `credentials` and `honorifics` (#326, decisions/R8).
 
-Open: [#326](https://github.com/derek73/python-nameparser/issues/326) — whether the honorifics belong in `suffix` at all, deliberately undecided here (the views' side of it is decisions/R8's).
+Open: none in this file. Whether the honorifics belong in `suffix` at all, left open here by the 2026-08-16 entry, was decided on 2026-11-03: they stay in the field and have a view of their own, decisions/R8's 2026-11-03 #326 entry.
 
 ## Log