Merge the tail's particle runs before S2's read (#625)

[derek73]

Summary

Since #614, S2's trailing run is read before P2's particle chain runs, so the read sees particles the chain hasn't merged yet. #620 made up for that in the reader:

• _chain_units marked a particle standing straight behind another as BOUND, and the peel stopped at it.
• credential_run gathered an absorbed run of those particles into one report.

Now group merges each run of two or more adjacent particles past the chain's leading position into one piece. It does this at the trailing site, after P3's connective joins and before the read (_pieces.merge_particle_runs). The read counts, stops at and absorbs the run as the one piece the chain makes, so the BOUND mark, its peel stop and the gathering are gone.

Two details the measurements forced:

• The merged piece keeps prefix and drops title. A P3 join can carry the title tag in from a word that is both a title and a particle (St und Berg). On a merged run that tag made the whole run read as a leading title.
• The chain's PARTICLE_OR_GIVEN report needed a new trigger. It keyed on the chain taking something past the particle (j > k + 1). A merged run has already done that, so the chain claims nothing more (Freiherr von der).
  • The report now also fires on runs the merge made (premerged).
  • Keying instead on "the piece holds several tokens" also catches P3-joined phrases (Dr. St Do und), and moved 11,358 reports on the grid below.

The title run's end and the chain's lead are read once, before the merge, and handed to the read, which used to ask for both again.

The rule text is unchanged: rules.md#S2 already counts a particle run as the one name word the chain makes, duals straight behind a particle included. The code grows by about 45 lines, docstrings included; what shrinks is the number of mechanisms. The read loses a mark and two branches and re-creates nothing of the chain; group gains the merge.

Measured (against 505d6de, py3.11, import asserted on each side; recipes in decisions.md#S2)

• Particle-run grid (395,148 texts, 1,185,444 parses over three orders, all seven fields and every report): 774 role moves and no report-only moves.
  • Every move contains St und. Master read Freiherr Do St und Berg MA as title 'Freiherr Do St und Berg', family 'MA'.
  • It now reads title 'Freiherr', family 'Do St und Berg', suffix 'MA', so the name part reads the same with and without the trailing 'MA'.
  • This is not a released reading. 1.4.0 reads first 'Do St und Berg', last 'MA', and 2.3.0 reads family 'Do St und Berg MA'. Over the 258 moved texts under the default order, master matches 2.3.0's full reading on 152 and the tree on none.
• S2 stress grid: no role moves, and one text's report changes under all four of its configurations.
  • The text is de Mesnil Ma do mc and Ph. D., and it no longer reports do.
  • P3 joins mc and Ph. D. into one suffix piece and the merge folds do into it, so the peel takes the piece whole.
  • adf6da8 didn't make that report either; Should S2's trailing read count the particle chain's units instead of merging a copy of them? #620 added it.
• No change on Where does Freiherr St John Smith MA start: should a title that is also a particle open the name? #624's title-particle head grid, the corpus names and case texts, or Where does Freiherr St John Smith MA start: should a title that is also a particle open the name? #624's comma-tail grid.
• Five differential gates: all exit 0, with reports identical line for line to master's.
• The issue's 2026-10-08 figure (80 moves, Dr. mc mc) no longer applies: Where does Freiherr St John Smith MA start: should a title that is also a particle open the name? #624 already reads Dr. Mc Mc with suffix 'Mc Mc'.
• The review found more reach. It is all garbage input and gets no rows (Derek's call). No move went the other way. The classes are recorded in decisions.md#S2:
  • The St und class is any P3 join whose left word is both a title and a particle. Example: Freiherr Do Freiherr and Berg MA.
  • A mid-name title and a dotted word after a particle run. Example: Freiherr van der Dr. G.J. MA now reads family 'van der Dr. G.J.', suffix 'MA'. That matches how Freiherr van Dr. G.J. MA already reads.
  • H3's title give-back before a run of dual particles. Example: Dr. vd Jr. mc mc now keeps title 'Dr.' where master handed it back as the given name.
• Frames: +3 on the reference row, which contains de la, and +1 on any name of three or more words read at the trailing site. Baselines are now 307 on 3.11 and 288 on 3.12–3.15 (decisions.md#parse-cost).

Tests

• test_a_merged_particle_run_is_a_particle_and_no_title (new): checks the merged piece's tags, and that the name's leading particle stays out of the merge (P4).
• _chain_units and read helpers in test_pieces.py: now merge first, as group does. The marks test reads van mc as one piece.
• test_the_read_takes_the_title_run_as_written_before_the_merge (new, from the review): Dr. mc Jr mc mc.
• New rows two_particle_runs_merge_each and a_second_run_of_duals_is_the_family (parity, from the review): two particle runs in one name. An off-by-one in the merge's bookkeeping had passed the whole suite; these rows catch it. The merge's unit test gains the same two-run half.
• Negative controls (recorded in decisions.md), each guard removed in turn:

Guard removed	Tests that fail
The merge itself	30 in tests/v2 (32 over the whole suite)
The premerged report trigger	2 case rows
Starting the merge past the lead (starts at the second piece instead)	13
Dropping title from the merged piece	only the new unit test
Reading the title run again after the merge, in the read or in group	the new read test (its parse assertion holds group's half)

Review

• 647e6c7: fixes the review's findings. It adds the two-run tests and the pre-merge title-run test, removes the false parity claim, corrects the grid size, records the extra move classes, and updates a stale case note.

• Review of the fix commit: corrects the merge-off control count, adds a parse assertion so the read test holds group's half too, and fixes a docstring word, the P3-join class's connectives and a row note.

Docs

• decisions.md#S2: new Merge the trailing run's particle runs in group, so S2's read needs no BOUND stop or report gathering #625 entry, plus superseded notes on Should S2's trailing read count the particle chain's units instead of merging a copy of them? #620's entry.
• decisions.md#parse-cost: new bullet for the frame change.
• mechanisms.md: UNIT-PARTITION updated.
• AGENTS.md: the decision-site bullet's "claimed something" check.

Closes #625

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.02%. Comparing base (505d6de) to head (9bf225d).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #634   +/-   ##
=======================================
  Coverage   99.01%   99.02%           
=======================================
  Files          46       46           
  Lines        4487     4495    +8     
=======================================
+ Hits         4443     4451    +8     
  Misses         44       44           

☔ 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.