Repository navigation
Smith, Jr., Dr. Bart reports its last part as "consumed as suffix" when 'Dr.' is read as a title #629
Description
Activity
- added a commit that references this issue
on Oct 9, 2026 Notes from #626's fix (PR #628), for whoever picks this up:
field_tailis the wrong tool here. Word a report's field from the word's final role (#626) #628 lets a report leave the field it names for assemble to fill in from the word's final role (PendingAmbiguity.field_tail). That works because each of those reports is about one piece whose tokens share one role. A part past the second comma doesn't: its words land in two fields. InJohn Smith, Jr., Freiherr von Richthofen, 'Freiherr' is a title and 'von Richthofen' are suffixes. So naming "the first token's field" would print "a title" for a part that is mostly suffix. The fix needs either a wording built from the part's distinct final roles ("consumed as title and suffix"), or a wording that names no field.- Add the phrasing to the guard.
test_a_report_names_the_field_its_word_lands_in(tests/v2/test_cases.py) only checks wordings on its_FIELD_CLAIMSlist, and "consumed as …" isn't on it. That's why the sweep passes today on the existing rowJohn Smith, Jr., Freiherr von Richthofen. Whatever wording the fix chooses, add it there, then re-recordtest_the_field_sweep_sees_the_claims_it_checks. - Where it's tracked: decisions.md#A1 now carries
Open: #629, and mechanisms.md's AMBIGUITY-AT-THE-DECISION-SITE entry notes this report as the exception.
A part past the second comma that the parser doesn't recognise gets a
comma-structurereport saying it was "consumed as suffix best-effort". Since #603 (rules.md#C2), a title word in that part reads as a title, so the report describes a reading the parse didn't make:John Smith, Jr., Dr. BartSmith, Jr., Attorney Berg,, Attorney BergAt 2.3.0 the wording was true, because the whole part became suffix. It went wrong this cycle and has not shipped. The report is made in
segment(_pipeline/_segment.py, theCOMMA_STRUCTUREemitter), before anything reads the part. Its referents then land in two fields.Found by: #626's review sweeping every report's field phrase against where its tokens land.
Fix direction: word it from the part's final roles at assemble, as #626 does for named fields (for example "consumed as title and suffix best-effort"). Or reword it to say what is certain, that the part is beyond the recognised comma structures, without naming a field. Either way, add the phrase to
_FIELD_CLAIMSin tests/v2/test_cases.py so the sweep checks it.Open question: does a part holding a title still need the report at all?
segmentalready excuses a part made only of titles and suffixes (titles_and_suffixes), so 'Dr. Bart' reports only because of 'Bart'.