Record the #495 decline, and correct a tier figure that shipped in #499 - #500

Records the #495 decline, and corrects a figure that shipped in #499.

Why #495 closed as not planned

It asked whether ledger rules that only explain radar-tier names still earn
their place. Measured at baseline 1.4.0, 19 of the 72 explaining rules explain
only radar names — 48 of the 352 intentional diffs. That is not the five the
issue named; it is a SUPERSET of them, containing all three rules #495's own
option 2 proposed deleting.

What kills the criterion is what deletion actually does. Driving classify()
over each of the 48 diffs with its own rule removed: 25 report UNCLASSIFIED
and 23 are claimed by a DIFFERENT rule — nine of fix(cjk-comma-compound)'s
eleven land on fix(cjk-glued-honorific-peel), all seven of
fix(cjk-comma-honorific-peel)'s land on the compound rule, 'Carod i' lands on
the roman-numeral fix(suffix-routing). So for nearly half the population
deleting is not merely wasteful, it is a MISLABEL — the rule-broader-than-the-diff
absorption this file already names twice.

The criterion that does justify deleting a rule is that it explains NOTHING,
which dormant already names and the dormancy check already enforces.

The five precedes_narrower exemptions citing #495 described it as a live
question in the present tense; all five now read as decided. The #496 decline
that leaned on it still stands on its own reasoning — nothing can demand the twin
rule, which is not what #495 settled.

The corrected figure

expected_since_1.4.0.toml has said since #499 that the routing rule "explains
eight of which six are contract". At baseline 1.4.0 it is five.

Not a mis-map: six is what tiering by FILE MEMBERSHIP gives from a correct
_CORPUS_TIERS. 'John Smith, Dr.' sits in corpus_issues.jsonl (radar, no
shape) and corpus_shapes.jsonl (contract, shape 4, FAMILY_FIRST,
min_baseline 2.0.0). The dedup key is (name, order), so the shape-4 copy is a
SECOND entry rather than the dedup winner — verified at baseline 2.0.0, where the
run compares both — and at 1.4.0 it is among the seven the baseline-minimum skip
drops. The only entry a 1.4.0 run compares for that string is the radar one.

Tier, in any claim scoped to a baseline, is a property of the ENTRY that run
compared — not of the files the string appears in. Measured: 24 of the 352
explained names sit in files of both tiers, 2 of those carry a shape tag, and
exactly 1 bites at 1.4.0. File-membership tiering gets 1 of 352 wrong;
entry["tier"] gets it right for free. The radar-only count of 19 is identical
under either tiering.

Three traps in one measurement, all recorded in the RECOMPUTE recipe

Two of them produced wrong answers during this work:

1. Report truncation (names[:10], no "and N more" line) — harmless for this
   predicate, because contract files load first so contract names occupy the
   prefix; first-ten-only still gives 19. Recorded because it looks dangerous.
2. Naive repr parsing — gives 20. "'Smitty' Dr. Jones" renders with double
   quotes because the name contains single ones, so strip-the-quotes drops
   fix(#410)'s only contract name. ast.literal_eval is the fix.
3. File-membership vs compared-entry tier — the one above, and the only one of
   the three that changes an answer on this tree.

Verification

Full suite 6467 passed, 156 skipped, 5 xfailed; ruff clean; mypy clean on 111
source files. Gate at all four baselines, exit 0 each: 352 / 247 / 155 / 14
intentional, 0 unexplained, 0 radar-unclassified.

Prose only — no rule, name_regex, fields, orders or file order changed.
design-docs-reviewer ran the eleven axes over the first draft and returned seven
confirmed findings; this is the rebuild, plus the shape-tag correction found while
fixing one of them.

🤖 Generated with Claude Code