Record the lessons of #623 through #641 in AGENTS.md - #645

Summary

This records five lessons from #623 through #641 in AGENTS.md. Each one is a defect that a check written down earlier would have caught. Following AGENTS.md's own rule, each names the PR and the input that showed it, and goes where the next person will meet it. Four amend existing paragraphs; one is a new Gotchas entry.

1. Workflow, guard tests. Re-take a recorded negative control after the last commit that adds tests, before merge. Three in a row went stale when a review round added rows: Merge the trailing run's particle runs in group, so S2's read needs no BOUND stop or report gathering #625 (27 → 30), Should John Smith, Jr., van der Berg Prof. read title 'Prof.' as John Smith, Jr., Prof. does? #636 (4 → 9) and Should Smith, abdul Dr. read title 'Dr.' as Smith, abdul Ali Dr. and abdul Smith Dr. do? #637 (5 → 7 → 11).
2. Gotchas, null results. A mutation that moves nothing shows code unexercised by your grids, not dead code. Remove group's chain re-ask after its own merges, which no longer changes a reading #638 was filed calling group's chain re-ask dead. The next review found 400 moves (Smith, Freiherr van Berg X.Y.Z.), and the issue closed the same day.
3. Release-log claims. A claim that a change restores the released reading has to be measured over the whole moved population. A population size is re-counted from its recipe, never copied from another entry. Merge the trailing run's particle runs in group, so S2's read needs no BOUND stop or report gathering #625 claimed its moves landed on the 1.4.0 and 2.3.0 readings from three names; measured, 0 of 258 did. The same entry copied Where does Freiherr St John Smith MA start: should a title that is also a particle open the name? #624's grid size (137,810) for a grid of 395,148.
4. Gotchas, new entry. A group-side predictor of a family comma's given part that asks tail_reading disagrees with the given part's own walk on three shapes: a shape-only numeral VI to X, V before a third comma part, and a title-particle that P6 attaches. This hit Should Smith, John van der Berg Prof. read title 'Prof.' as John van der Berg Prof. does? #623's chain bound and Should Smith, abdul Dr. read title 'Dr.' as Smith, abdul Ali Dr. and abdul Smith Dr. do? #637's join decline. Both were recorded rather than narrowed.
5. Sweep paragraph. The title-particles (freiherr, st) go in every head list in both bare and period-marked spellings. Should Smith, abdul Dr. read title 'Dr.' as Smith, abdul Ali Dr. and abdul Smith Dr. do? #637's grid had only the bare forms, and missed St. and Freiherr..

Docs only; the doc tests pass.

🤖 Generated with Claude Code