R-426: one-register reads the leading verdict; decoys added, exemption removed

DONE/IDEA were searched over the whole state cell, so an open ROADMAP row whose
state named ANOTHER row's verdict (`READY — split out after R-86 CLOSED`)
escaped as shipped — R-87's shape from R-378, one file over. The gate now reads
register_table.leading_verdict(state); all 28 current rows classify unchanged.
Decoys: that shape (seen to PASS against the old gate), an id present in the
register only as prose (seen to PASS when `have` is loosened to any mention),
the existing suffix-id row; genuine: a row with a counterpart, and an `idea` row
asserted BY DESIGN — R-424's hole, accepted by the operator 2026-10-05.
EXEMPT drops `felhom.eu/one-register`.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS

scripts/one_register_gate.py

@@ -21,7 +21,8 @@ WHAT THIS GATE CANNOT SEE — the residual holes, named rather than implied:
 
   1. **A finding filed with the state `idea` escapes.** The state column is a human judgement, and a
      defect written under `idea` looks exactly like a proposal to this gate. It is the same shape as
-     the problem it fixes, one level up.
+     the problem it fixes, one level up. **Accepted by the operator 2026-10-05 (R-424, closed): no
+     mechanical fix exists; a person reads the roadmap.** test_gate_decoys.py asserts it BY DESIGN.
   2. **A finding written in prose with no `R-` identifier escapes entirely.** This gate matches ids.
      A survey that enumerates a gap and never numbers it is invisible here — that is the previous
      session's sweep territory and the PROMPT-TEMPLATE rule "an enumerated gap becomes a row".
@@ -45,8 +46,9 @@ ROADMAP = os.path.join(ROOT, "documentation", "backlog", "ROADMAP.md")
 REGISTER = os.path.join(ROOT, "documentation", "backlog", "OPEN-ITEMS.md")
 
 ROW = re.compile(r"^\|\s*\*{0,2}R-(\d+[a-z]?)\*{0,2}\s*\|")  # 2026-10-03: a suffix id (R-27b, R-27c, R-50b) was skipped unchecked
-# DONE and IDEA are matched against the row's STATE cell only, never the whole row: the body of a
-# finding routinely contains the word "shipped" while describing something else.
+# DONE and IDEA are matched against the LEADING VERDICT of the row's STATE cell only, never the whole
+# row (the body of a finding routinely contains the word "shipped" while describing something else) and,
+# since 2026-10-06, never the whole cell either (a state can name another row's verdict).
 # BANKED and PROVEN-LIVE are this project's own done-words and were found by running the gate: a row
 # reading "BANKED in full" or "first slice PROVEN-LIVE" is shipped work, not an open finding.
 DONE = re.compile(r"\b(SHIPPED|CLOSED|KILLED|DONE|SUPERSEDED|OBSOLETE|WITHDRAWN|MERGED|DISCHARGED|"
@@ -77,10 +79,15 @@ def main():
     have = set(rid for rid, _, _ in rows(REGISTER))
     offenders, ideas, done, ok = [], 0, 0, 0
     for rid, state, line in rows(ROADMAP):
-        if DONE.search(state):
+        # R-426 (2026-10-06): the LEADING VERDICT of the state cell, not the whole cell — R-378's lesson,
+        # one file over. A cell reading `READY — split out after R-86 CLOSED` matched DONE on the other
+        # row's word and the open finding escaped as shipped (and `READY — was an idea` would escape as
+        # an idea). Decoy: test_gate_decoys.py `one-register/done-word-about-another-row`.
+        verdict = register_table.leading_verdict(state)
+        if DONE.search(verdict):
             done += 1
             continue
-        if IDEA.search(state):
+        if IDEA.search(verdict):
             ideas += 1
             continue
         if rid in have: