fix(assign,vocab): a one-word family comma reads its credential run as suffixes (#296, #325) - #428
Open
derek73 wants to merge 5 commits into
Open
fix(assign,vocab): a one-word family comma reads its credential run as suffixes (#296, #325)#428derek73 wants to merge 5 commits into
derek73 wants to merge 5 commits into
Conversation
…y by titles
The FAMILY_COMMA rule that segment 0 is wholly the family name rests
on the writer having said where the family name ends. A comma
followed by nothing but titles said no such thing -- 'John Smith,
Mr.' is 'Mr. John Smith' with the honorific moved -- so 2.0's merge
of the pre-comma run into one family name threw away a split the
writer gave us: family 'John Smith', title 'Mr.'. Segment 0 keeps its
positional read there, with the script-order table and the
particle-or-given report that the merged read suppressed ('Van
Johnson, Mr.' reports the fork again).
"Nothing but titles" is title-shaped AND not suffix vocabulary, which
the approved plan's helper (2026-08-01) did not say: the title list
overlaps the suffix sets until the audit in the next commit, and the
period-abbreviation inference reads 'Jr.' as a title regardless, so
the helper as planned read 'Smith, PhD Jr.' as the title run 'PhD
Jr.' and 'Smith, Mr. Jr.' as 'Mr. Jr.' -- both master's suffix 'Jr.'
(the 2.1.0 gate showed the first; the second was measured against
master). A suffix piece is read where it stands, whatever else the
word is. One pre-comma piece has no split to keep ('Smith, Dr.' is
unchanged), and a post-comma name still fixes the family ('John
Smith, Jones').
This is the repair the #291/#296 bundle plan orders FIRST, so that
the vocabulary audit -- which makes 'Dr.' non-suffixy and would
otherwise flip 'John Smith, Dr.' into this merge -- lands on a path
that already keeps the split. Two corpus names move: 'Bob Jones,
author' and 'Bob Jones, compositeur' read given 'Bob', family 'Jones'
at every baseline (1.4.0 read first 'Bob Jones'), under a rule
written ahead of the comma-precomma-family rule whose fields had
absorbed them (11 -> 9, the count telling); the contested-name record
follows. 'John Smith, Dr.' itself is unchanged until the audit.
Tests first: the split, its two-piece guard, the all-title run, the
title-then-name and name-only non-flips, and the fork report; case
rows for the split, the guard and the non-flip.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…, and read a post-comma credential run as suffixes v1's word lists encoded where the v1 parser needed words to be, not where words can occur, and the listing form paid for it twice. 'Smith, Jr.' read title 'Jr.' (#296): with one word before the comma C1 keeps the listing form, and the title peel -- 'jr' was title vocabulary, and the period-abbreviation inference reads 'Jr.' as a title regardless -- claimed the lone post-comma piece before the 2.0 deviation that routes a lone suffix piece to SUFFIX could. 'Smith, Ph. D. Jr.' read given 'Ph. D.' (#325): the space-split credential followed by another suffix was no longer the LONE piece, so it fell through to v1's walk and became the given name -- a 1.4.0 regression (v1 read suffix 'Ph. D.', title 'Jr.'). The two halves are one change because each alone is wrong (the 2026-08-01 plan's F1): reading the post-comma run as suffixes first would make 'Smith, Dr.' a suffix, since 'dr' was in SUFFIX_WORDS as v1 residue; cleaning the vocabulary alone leaves 'Smith, Jr.' a title by inference. So: the nineteen TITLES/suffix overlap words get the classes the approved 2026-07-30 audit table gives them -- 'do', 'jr', 'junior', 'phd' and 'se' leave TITLES (postnominal only; 'Do' is a Vietnamese name), 'dr' leaves SUFFIX_WORDS and 'sra' SUFFIX_ACRONYMS (v1 residue), 'ms' and 'sa' join the ambiguous set so the bare spellings read as the honorific and Special Agent and the perioded ones as the degree and the business form, and the rest keep dual membership with position deciding -- and in the listing form a post-comma segment that is nothing but suffix pieces is the credential run C1 describes and reads as suffixes, whole, before the title peel: 'Smith, Jr.' -> suffix 'Jr.', 'Smith, Ph. D. Jr.' -> suffix 'Ph. D., Jr.', 'Smith, Sr.' -> Senior where 'Sr. Garcia' is still Señor, 'Smith, Esq.' -> the postnominal where H2 still reads 'Esq. Smith' as a title. 'Smith, Dr.' never reaches the ordering, 'dr' not being suffix vocabulary; 'Smith, Dr. Jr.' keeps v1's walk (title, then the last piece as the suffix); 'Smith, John Jr.' keeps its given name. One word deviates from the approved table, on evidence the table did not have: 'md' KEEPS dual membership. Bare 'Md' before a name is the Bengali and South Asian Muslim abbreviation of Muhammad -- 'Md Abdul Karim' and 'Md. Abdul Karim' are #343/#345's corpus rows, and the drop read the first as given 'Md', middle 'Abdul' -- while 'MD' after the name is the degree. Position decides, as for 'sr'. Derek to confirm or overrule. Consequences, all measured: a leading 'PhD' or 'Jr' is a name word now ('PhD Smith' -> given 'PhD', a string nobody writes); 'Do Quang Minh' and 'Do Nguyen' read given 'Do' with the particle-or-given fork 'Van Johnson' reports, and 'Dr. Do Van Johnson, MD' reads given 'Do', family 'Van Johnson' where 'do' had stopped #367's transparency scan as a title-and-particle word (the fix(#367) rule for that name is replaced; 'St' and 'Freiherr' still stop it, and the #367/#424 examples spelled with 'Do' are respelled); a trailing bare 'Dr' is a name word ('John Smith Dr.' -> family 'Dr.'), which is what every other title-only word already does there (#316 is the open question) and what the #100 prefix-join tests relied on 'dr' NOT doing -- they use 'lt' now, a genuine dual; '田中さん, PhD' reads suffix 'さん, PhD'. The #367 ledger rule's alternation loses 'jr' (no longer a title), and 'Jr. Van Johnson' -- the same reading, by inference -- has its literal. rules.md#C1's #296 deviates-marker is a real example now, with the #325 shape beside it; assign cites C1. Tests first, at every level: the assign-level run, duals, non-flips and mixed runs; case rows for each audit disposition in leading, post-comma and trailing position, the ordering's shapes, the #325 shapes and the non-flips; test_post_rules' lexicon mirrors the shipped classes. Ledger rules at every baseline for each class that moves (literal where a word is involved; the #325 rule is the shape, since an alternation over the credential spellings would reach the spaced 'Ph. D.', which is no entry); the 1.4.0 catch-all drops 15 -> 14 as a name it had absorbed gets its rule. Two release-log bullets. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… read for what it is
rules.md#C1 says what the listing form does with the part after the
comma: a part that is nothing but suffix words is the credential
run and reads as suffixes, whole ('Smith, Jr.', 'Smith, Sr.',
'Smith, Ph. D. Jr.'); a part that is nothing but titles fixes no
family boundary, so the part before the comma keeps its positional
read ('John Smith, Mr.'); a name word makes it the given name, with
titles before and suffixes after ('Smith, Dr. Jr.', 'John Smith,
Jones'). The #296 deviates-marker is an example; #291's stays. H2
carries the one place the two meet as Accepted: after a family comma
the credential run is not an opening the abbreviation inference
runs on, so 'Smith, Esq.' is the postnominal where 'Esq. Smith' is
the title.
decisions.md#C1 records the bundle's three commits in the plan's
order -- the repair and what "nothing but titles" had to mean, the
audit and the ordering as one change and why, the one deviation
from the approved table ('md' keeps dual membership: bare 'Md' is
Muhammad, #343/#345's corpus rows) for Derek to confirm, the #325
widening, and the audit's measured consequences. The comma-suffix
arc is shipped less #291. The three places that named {do,
freiherr, st} as the load-bearing TITLES ∩ ambiguous set say it is
{freiherr, st} since the audit, and the C-i demonstration's 'Do
Quang Minh' reads given 'Do' with the fork now. AGENTS.md's
positional gotcha points at the audit.
The new C1 examples are corpus names (203); 'John Smith, Mr.' gets
its literal rule beside 'Bob Jones, author' at every baseline, and
the reach pins follow the seven new names.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #428 +/- ##
=======================================
Coverage 98.59% 98.59%
=======================================
Files 44 44
Lines 3053 3070 +17
=======================================
+ Hits 3010 3027 +17
Misses 43 43 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
… name count, the order, the duals, the gate Four reviewers on PR #428 (code, tests, comments, design docs). What they found, and what changed. The repair's condition is "no name word after the comma", not "all titles": 'John Smith, Mr. Jr.' -- a title and a postnominal -- had kept the merge the rationale argues against, and keeps its split now (given 'John', family 'Smith', suffix 'Jr.'). Its guard counts NAME pieces: the positional read peels a trailing suffix first, so 'Smith Jr., Mr.' -- two pieces, one name -- read positionally lost its family (the code review's critical); it keeps family 'Smith' at every baseline. The positional read records its ORDER, and post_rules' leading-piece scan keys on "assign records no order after a family comma" instead of on the structure, so P1's family- first fold reaches the pre-comma name as it does without the comma: 'de Mesnil Juan, Dr.' under a family-first order had read family 'de' (the test review). The dead segments-count guard the coverage check flagged is gone; the helper is called on the family-comma path alone. Two deviations from the 2026-07-30 audit table now, both measured. 'ms' and 'sa' do NOT join the ambiguous set: the gate is position- blind and the collision is not -- gated, 'John Smith, MS' lost its suffix-comma route and read title 'MS', and 'Smith, Ms.' passed the gate on its one period and read as a credential anyway. Both are genuine duals and position decides: 'Smith, Ms.' reads suffix 'Ms.' (title at every baseline; C1 carries it as Accepted, 'Smith, Ms. Jane' keeping the title), 'John Smith, MS' keeps suffix 'MS'. And the period gate means what S2 says: an ambiguous acronym counts when written with ITS periods, one after each letter ('M.A.', 'J.D.'), not on any period -- 'Smith, Ed.' had become a credential on a name's trailing period, and 'Jack Ma.' had read suffix 'Ma.' since 2.0 (family 'Ma.' now, as 1.4.0 read it; S2 carries the boundary). Recorded rather than fixed, with the reason: 'Smith, RN - CRNA' under extra_suffix_delimiters keeps given 'RN' -- a draft stepped over the core, and tests/test_suffixes.py pins "the delimiter must not affect parsing at all" outside the suffix-comma form as v1 parity; '田中さん, Dr.' keeps the honorific glued, joining master's '田中さん, Mr.' -- the peel runs in script_segment before group or assign can say the comma fixed nothing; group's chain emitter still reports no PARTICLE_OR_GIVEN on the family-comma path. The docs follow: C1's statement carries the no-name-word condition, the name count, the order and the three Accepted limits, with examples; H2's statement says where the shape does not outrank vocabulary; S2's says "its periods"; decisions#C1 records the round, 'Smith, Dr. Jr.' is attributed to the 2.0 deviation rather than v1, 'Jong, vd' is C1's by statement, the guard test is named by its real name; AGENTS.md and the release log carry the shipped counts (five words left TITLES, twelve duals remain, 694 title-only) and the shipped examples. The ledgers: the lone-credential rule sits behind the two comma-family rules so 'Smith, Dr.', 'Smith, Prof.' and 'Smith, Abd' keep the rules written for them; the #325 shape rule is two literal rules that cannot absorb 'Smith, Dr. Jr.'; the 'do is a name' rule states the baselines' reading and drops the title field that did not move; comment corrections from the comment review; the contested-name shape for '田中さん, PhD' re-recorded; reach pins follow the thirteen new examples (corpus 210 + 98). Tests first at every level -- the assign-level no-name-word read and the recorded order, the family-first rows (core-only, v1 has no order to declare), the duals in every position, the name-count guard's three shapes, the mixed runs, the title runs, the three- segment run, 'Smith, MD'; the pins the test review asked for. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…e word test_revise_sets_a_missing_unjoined_mark rested on 'Do' alone parsing as a TITLE, so that the sub-parse marked nothing and the recompute had to mark the bare particle it re-roled into FAMILY. #296's audit took 'do' out of TITLES, 'Do' alone is a marked given name now, and the test passed without reaching the recompute -- which is what Codecov's project check saw (one line of _types.py lost its witness; the project has no codecov.yml, so any drop fails). 'St' is still in both vocabularies; the Do case stays as the other path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #296, closes #325
Bundle 2, first PR: the first three commits of the approved #291/#296 comma-suffix bundle (spec 2026-07-30, plan 2026-08-01), in the plan's order, with the ordering step widened to cover #325 — plus a fourth commit for the review round. #291 (multi-word credentials,
SUFFIX_PHRASES) is the bundle's fourth commit and the next PR.The change
Smith, Jr.,Smith, PhD,Kim, Jr.Jr./PhD(#296)Smith, suffixJr./PhDSmith, Ph. D. Jr.,Smith, Ph. D. III,Smith, PhD Jr.,Smith, Jr., PhDPh. D.(#325, a 1.4.0 regression)Ph. D., Jr.etc. — the whole credential runSmith, Sr./Sr. Garcia;Smith, MS/Ms. SmithSmith, Esq./Esq. SmithEsq./ titleEsq.— H2's inference doesn't run on the credential runJohn Smith, Mr.,John Smith, Mr. Jr.,John Smith, Dr.,Bob Jones, authorJohn SmithJohn, familySmith, title (+ suffix) — a comma followed by no name word fixes no family boundary;de Mesnil Juan, Dr.keeps familyde Mesnilunder a family-first orderDo Quang Minh,Do Nguyen,Dr. Do Van Johnson, MDDoDo(with the particle-or-given fork) —dois a Vietnamese name, never a prenominalPhD Smith,Jr SmithJohn Smith Dr.,Smith DrDr.Dr.—drwas v1-residue suffix vocabulary; a trailing title-only word is a name part on the no-comma path today (#316)Jack Ma.Ma.(since 2.0)Ma.— the ambiguous-acronym gate needs the acronym's own periods (M.A.), not any period; as 1.4.0 read itSmith, Ms.Ms.— the cost ofmsas a dual, recorded as Accepted;Smith, Ms. Janekeeps the titleSmith, Dr.,Smith, Dr. Jr.,Smith, Mr. Jr.,Smith Jr., Mr.,Smith, John Jr.,John Smith, MS,Hardman, RN - CRNA,Md Abdul Karim,田中さん, Dr.Vocabulary audit (the approved table, two deviations).
do,jr,junior,phd,seleaveTITLES;drleavesSUFFIX_WORDSandsraSUFFIX_ACRONYMS; the rest stay genuine duals, position deciding. Deviation 1 —mdkeeps dual membership: bareMdbefore a name is the Bengali and South Asian abbreviation of Muhammad (Md Abdul Karim,Md. Abdul Karimare #343/#345's corpus rows; the drop read the first as givenMd, middleAbdul). Deviation 2 —msandsado not join the ambiguous set: the gate is position-blind and the collision is not — gated,John Smith, MSlost its suffix-comma route and read titleMS, andSmith, Ms.passed the gate on its one period anyway (the review found it). Derek: confirm or overrule either — each is one line of vocabulary and a case row.Mechanism. (1)
_segment_holds_no_name(titles and suffixes only, by the peel's own title test) sends segment 0 through_assign_mainwhen it has more than one name piece, and records the order, which post_rules' leading-piece scan now keys on (so P1's family-first fold reaches it). (2) In the FAMILY_COMMA branch a post-comma segment that is wholly suffix pieces reads as suffixes before the title peel. (3)suffix_as_written's period gate is the dotted-acronym form. The audit and the ordering are one commit because each alone is wrong (the plan's F1).Review rounds
Four reviewers on the pushed PR (code, tests, comments, design-docs); everything found is in commit 4, not deferred:
Smith Jr., Mr.lost its family (the guard counted pieces; the positional read peels the suffix first) — counts name pieces now.de Mesnil Juan, Dr.under a family-first order read familyde— the order is recorded and post_rules' scan follows it.Smith, Ms./John Smith, MSunder the table's ambiguous gate — the gate dropped forms/sa(deviation 2), and the period gate tightened to the acronym's own periods (Smith, Ed.,Jack Ma.).John Smith, Mr. Jr.— C1's condition is "no name word", not "all titles"; H2's statement now says where the shape does not outrank vocabulary; S2 says "its periods".Smith, RN - CRNAunderextra_suffix_delimiterskeeps v1's "the delimiter has no effect outside the suffix-comma form" (pinned in tests/test_suffixes.py);田中さん, Dr.keeps the honorific glued (the peel runs before the comma is read); group's chain emitter does not report on the family-comma path.Smith, Dr.,Smith, Prof.,Smith, Abd,Smith, Dr. Jr.) are ordered/literalized; counts in AGENTS.md and the release log are the shipped ones (five words left TITLES, twelve duals remain); a dead guard Codecov flagged is gone.Verification
test_revise_sets_a_missing_unjoined_markrested onDoparsing alone as a title, which the audit made false, so_types.py's recompute branch ran in no test. Commit 5 respells it withSt. (Nocodecov.yml: the project target is "no drop", so a reading change that orphans a witness fails it.)Commits
fix(assign)— the all-titles repair (plan commit 1), with the suffix-vocabulary clause the helper needed.fix(vocab,assign)— the audit (plan commit 2,mdexcepted) + the credential-run ordering (plan commit 3, widened forSmith, Ph. D. Jr.puts the credential infirst(regression from 1.4.0) #325), tests and ledgers.docs(design)— C1/H2, the decisions entries, the arc, AGENTS.md, corpus.fix(assign,vocab)— the review round: no name word, the name count, the order, the duals (ms/saexcepted), the period gate; docs, ledgers and pins.test(parser)— the revise recompute's witness respelled with a word still in both vocabularies.🤖 Generated with Claude Code