fix(parser): suffix entries follow the commas the writer typed (#436, #437)

[derek73]

The suffix view now separates post-nominals the way the writer did. Two consecutive suffix tokens render as one space-joined entry iff they sit in the same comma-delimited part of the name and nothing between them belongs to the name or separates entries; otherwise a comma. HumanName("John Smith MD PhD").suffix is MD PhD, as it has been for John Smith, MD PhD since 1.4.0 and for Smith, MD PhD since #429. Closes #436, closes #437.

What was wrong

rules.md#R1 already said it: "a run of post-nominals written with spaces renders with spaces, and one written with commas keeps them." Every example under it was a comma path, and the no-comma path violated the sentence at every release since 1.4.0. #429 fixed the family-comma path and recorded this one as "left alone"; it is also why round-tripping was unstable on the shapes #429 fixed, since str() of a fixed parse is a no-comma string.

The cause was where the entry boundary was decided: in group, from segment SHAPE — tail by index, ORed since #429 with a content verdict, gated per piece, sticky across an interleaved title, closed by a delimiter core. Three shape-derived answers to one question the input answers directly.

The change

A pass at the end of post_rules, after roles are assigned, walks the SUFFIX tokens in string order and tags a continuation joined iff the pair shares a comma bucket (comma_bucket, the function segment builds segments with) and nothing parts them. Transparent: a token that renders into another field — title, nickname, maiden — and a dropped maiden marker. Parting: a surviving name word, or a dropped delimiter core, read through the same delimiter_cores() helper group uses to drop it. group loses its between-piece marking whole and keeps core dropping and the within-piece Ph. D. heal. The tag stays the recorded form, so pickled names keep working.

Measured over the 1116 corpus names: 13 move, every one from a comma the writer never typed to the spacing they wrote; no name moves the other way; every list view other than suffix_list is byte-identical. Nine cases.py rows, four rules.md examples and eight older assertions carried the artifact and are re-pinned with notes. Round-tripping is stable on the shapes #429 fixed and a test pins those three; over the corpora it is a rendering, not a canonical form — 35 names failed to round-trip at the parent and 23 do now, 13 fixed and one newly visible, Smith, John PhD I., whose render used to hide a parse disagreement behind the inserted comma. The decisions entry carries the measurement.

What review caught

The first draft of the code commit made only a title transparent. Measured against the parent tree, that regressed Smith, MD "Doc" PhD and both spellings of Smith, MD nee Jones PhD from MD PhD to MD, PhD and left John Smith MD "Doc" PhD unfixed. Four stage tests and a case row now pin the transparent set, and the decisions entry records the miss. A four-reviewer pass over the PR then added a mixed-comma three-suffix test (the one mutant that had survived, measuring the comma bucket against the run's first suffix), pinned that a marker the policy itself names as a delimiter parts the run, and corrected five figures and sentences in the records. One limit is recorded rather than fixed: Parser.revise(n, suffix=n.suffix) is not the identity on a space-joined run, since the bare string has no comma to route by; pre-existing, and reachable on more inputs now (24 of the parent's corpus names, 38 here).

Records

R1 gains its no-comma example and a sentence naming the mechanism. Six ledger rules at fourteen placements, each with its claim roster; the differential gate reads 360 / 254 / 166 / 28 intentional at the four baselines, up from 352 / 247 / 155 / 14, with 0 unexplained and exactly one name changing rule (abdul Smith Jr V at 2.x, whose diff grew past fix(#401)'s fields). A dated bullet under C1 in decisions.md carries the decision, the figures, the four declined shapes and the review finding; the #429 and #430 entries are amended where this made them history; a 2.3.0 release-log bullet records the deviation from 1.4.0.

Verification

Full suite green, ruff and mypy clean, gate exit 0 at four baselines with no finding block. Mutation controls: dropping the comma-bucket conjunct joins Smith Jr., Mr. Jr. across the writer's comma; dropping the parting conjunct joins PhD FACS across a delimiter; running the pass before assign collapses John A B Smith's middle names. Each is killed by a named test.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.66%. Comparing base (f5deae0) to head (9e3834b).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #510   +/-   ##
=======================================
  Coverage   98.66%   98.66%           
=======================================
  Files          45       45           
  Lines        3219     3219           
=======================================
  Hits         3176     3176           
  Misses         43       43           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.