@@ -28,7 +28,7 @@ Three committed contributor docs carry the parser's normative rules and their re
 
 **Counting claims.** A bare count in prose is either an assertion or a liability, keyed by who observes its staleness: asserted counts (a test holds the number) fail CI at change time — the useful kind; dated snapshots ("51 sites at spec time") cannot go stale; standing present-tense prose counts are the forbidden class — promote to an assertion, add a date, or state the invariant and let a test count. After changing how many times something runs, sweep for counts, not for the thing's name.
 
-**Release-log claims.** Quantified or universal behavior claims in release bullets must come from the differential gate's classified summary, be verified against rules.md examples, or -- for a view the gate cannot see -- carry a recompute recipe stored with the design entry the bullet cites; never write one from memory. The classified summary covers the CONTRACT tier plus whatever radar diffs a rule classifies; a radar corpus's unmatched diffs are listed under UNCLASSIFIED (radar) and are not in it, so a claim quantified from the summary alone is silent about them. The gate compares the seven role fields, `_ambiguities` and -- since #484 -- `initials()`, under the `_initials` pseudo-field; so an initials-only change DOES show in the classified summary. But that pseudo-field sees only the names whose initials moved WHILE EVERY ROLE AND EVERY AMBIGUITY KIND STAYED PUT -- main()'s roles-identical guard keeps it out of any diff a role or a report is already in -- so a count taken from it is a FLOOR on initials movement, not the population: measured 2026-09-02 at 2.1.0 → tree on the v2 surface, 83 of the 1120 compared ENTRIES (1116 distinct names; seven entries carry a declared order rather than the default, and three strings are compared under more than one) changed their `initials()` string and only 28 were visible under `_initials`, the other 55 having moved a role as well. A bullet about how many names' initials changed still needs the recompute recipe. `capitalized()` and any other render view stay invisible to the gate (decisions.md#R4, #R3), and for those the first two sources still cannot reach a claim: a gate run is byte-identical across the change, and an example line witnesses an output without counting anything. A recipe names the corpus files, the policy sweep, and -- the part that is easy to omit and fatal -- THE COMPARATOR, which must be something the shipped tree is not: #408's first recipe said to compare `initials()` against a folded-first partition, which is what `initials()` now IS, so it reproduced 0 where the bullet claimed 660 and was the only stated provenance for the number. Run the recipe as written before shipping the bullet. Cross-version numbers (a released wheel, the pre-change tree) are dated snapshots under Counting claims, since nothing in the repository re-runs them. Per-rule ledger toml comments asserting PARSER behavior cite rule IDs under the excerpt discipline; free prose is for ledger mechanics only (owned by tools/differential/README.md).
+**Release-log claims.** Quantified or universal behavior claims in release bullets must come from the differential gate's classified summary, be verified against rules.md examples, or -- for a view the gate cannot see -- carry a recompute recipe stored with the design entry the bullet cites; never write one from memory. The classified summary covers the CONTRACT tier plus whatever radar diffs a rule classifies; a radar corpus's unmatched diffs are listed under UNCLASSIFIED (radar) and are not in it, so a claim quantified from the summary alone is silent about them. The gate compares the seven role fields, `_ambiguities` and -- since #484 -- `initials()`, under the `_initials` pseudo-field; so an initials-only change DOES show in the classified summary. But that pseudo-field sees only the names whose initials moved WHILE EVERY ROLE AND EVERY AMBIGUITY KIND STAYED PUT -- main()'s roles-identical guard keeps it out of any diff a role or a report is already in -- so a count taken from it is a FLOOR on initials movement, not the population: measured 2026-09-02 at 2.1.0 → tree on the v2 surface, 83 of the 1120 compared ENTRIES (1116 distinct names; seven entries carry a declared order rather than the default, and three strings are compared under more than one) changed their `initials()` string and only 28 were visible under `_initials`, the other 55 having moved a role as well. A bullet about how many names' initials changed still needs the recompute recipe. `capitalized()` and any other render view stay invisible to the gate (decisions.md#R4, #R3), and for those the first two sources still cannot reach a claim: a gate run is byte-identical across the change, and an example line witnesses an output without counting anything. A recipe names the corpus files, the policy sweep, and -- the part that is easy to omit and fatal -- THE COMPARATOR, which must be something the shipped tree is not: #408's first recipe said to compare `initials()` against a folded-first partition, which is what `initials()` now IS, so it reproduced 0 where the bullet claimed 660 and was the only stated provenance for the number. Run the recipe as written before shipping the bullet. Cross-version numbers (a released wheel, the pre-change tree) are dated snapshots under Counting claims, since nothing in the repository re-runs them. Per-rule ledger toml comments asserting PARSER behavior cite rule IDs under the excerpt discipline, and since 2026-10-10 tests/v2/test_doc_citations.py checks them as it checks code. A ledger comment is copied from ledger to ledger and reworded by nothing: measured 2026-10-10 by running the check as it now stands over the tree before it (8e6e524a, master after #631), 27 quotes were stale and 5 colon citations quoted nothing, beyond the five #631's review found by hand -- most in a shape the colon-only check could not read (`rules.md#X -- "..."`, `rules.md#X's Accepted clause ("...")`, a second quote chained by "and"). A rule's statement and its `Accepted:` clauses are quotable, its example lines are not; `rules.md#H` names the H section, whose Background is quotable; a decisions.md citation may paraphrase, but what it quotes must be verbatim; and a reference with no quote must still name something that exists. The check's reach is a quote within four words of the ID or chained to one that is: a quote further off reads exactly like a name written in double quotes, so it is NOT checked, and the second review of #632 found 14 such quotes stale by hand. Keep a quote beside its ID. Quotes of AGENTS.md itself are not checked at all. Free prose is for ledger mechanics only (owned by tools/differential/README.md).
 
 **Writing the user docs (docs/*.rst) has its own AGENTS.md too.** `docs/AGENTS.md` carries the style the docs are written in — tables as indexes, subheadings per task, measured claims, recipe doctests — distilled from PR #588. It loads the same way as the docs/design/ one below; if your tool does not do nested discovery, read it before editing a `.rst` file under docs/.
 
 
@@ -715,9 +715,9 @@ def _split_surname_site(state: ParseState) -> ParseState:
            for j in state.segments[0]):
         return state
     # No try/except around the call: rules.md#A1's Accepted clause
-    # ("a user-supplied segmenter's own error propagates"). The two
-    # checks below are that same doctrine, curated,
-    # and they are where the line this module draws is easiest to state:
+    # ("a user-supplied segmenter's own error, which propagates"). The
+    # two checks below are that same doctrine, curated, and they are
+    # where the line this module draws is easiest to state:
     # a PROTOCOL VIOLATION BY THE SEGMENTER AUTHOR RAISES, while an
     # ADAPTER'S DEFENSE AGAINST ITS LIBRARY DECLINES. Both checks here
     # are the first kind -- a wrong answer TYPE and an answer indexing
 
@@ -426,7 +426,7 @@ def test_the_walks_own_leading_piece_never_anchors_what_follows_it(
 
 def test_a_reserve_kept_leading_piece_beside_a_genuine_family_loss(
 ) -> None:
-    # decisions.md#S2's Accepted boundary ("an unambiguous suffix is
+    # rules.md#S2's Accepted clause ("an unambiguous suffix is
     # consumed even when that leaves no family name at all", 'Smith
     # Jr.' -> family='') applies just the same when the LEADING piece
     # is itself listed suffix vocabulary rather than an ordinary name: