tooling(differential): every measured contest has an owner (#498)

[derek73]

Every measured contest in the differential ledger now has an owner. A new run-time check, UNPINNED CONTEST, fails the gate where a diff two or more rules admit is won by neither narrow-first nesting nor a precedes_narrower declaration and no _RECORDED_DIFFS row pins it; the fourteen names that lacked one are adjudicated and pinned. Closes #498.

What was open

#382's undeclared_contests refuses a WIDER rule sitting ahead of a NARROWER one it shares a corpus name with, unless the wider rule declares why. Nesting is sufficient for an order-decided contest and not necessary: two rules whose fields merely intersect, or are equal, both admit any diff inside the intersection, classify() returns the first, and the static check looks past the pair in either arrangement. The docs said _CROSS_RULE_WINNERS "stays the instrument" for that class. Nothing checked that the instrument had been applied.

Measured 2026-09-05 by driving main() at all four baselines with classify() spied (recipe in the decisions entry):

baseline	diffs two or more rules admit	already pinned	owed a pin
1.4.0	47	18	14
2.0.0	3	3	0
2.1.0	1	1	0
2.2.0	0	0	0

The fourteen come from seven (winner, loser) pairs over fifteen name-pair rows. Seven of the names sit in EQUAL-fields pairs — six under fix(#325) against the lone-post-comma rule, and Bob Jones, compositeur — the clearest case of the class the docs named the roster for, and none was pinned. Four are contract tier: the issue's three nakaguro names and Smith, Ph. D. Jr..

Why run-time, and what the 2026-09-02 decline declined

A pin is a shape beside a winner, and the shape is what a run measured. Without a diff the population is the static reach of every pair the check can report, 71 pairs (60 overlapping, 11 with equal fields), and the roster nobody writes; with a diff it is 14 names. So the check has no static form. The recorded decline of run-time contest detection was of REPLACING the static check, on coverage; this adds to it: the static check covers the hazard that has a declaration site, this covers the hazard that has none. Speed was never the reason — a whole gate run is a fraction of a second, and the "multi-minute" phrase was withdrawn 2026-09-03.

The commits

1. The fourteen pins. A researcher adjudicated each of the fifteen name-pair contests against both rules' prose and the measured 1.4.0 old/new values; all fifteen current winners stand, none by accident. Each gets a _RECORDED_DIFFS row and a _CROSS_RULE_WINNERS winner with a per-pair argument that says what the pin buys (a reorder moves 7 / 3 / 9 / 15 / 2 / 3 / 8 of the 352 classifications, pair by pair). Five names move from _WATCHED_DIFFS; every 31-row figure in the tree becomes its 45-row successor, re-measured, with the dated records keeping their numbers and gaining a clause.
2. unowned_contests, a pure function behind hermetic tests, with _UNOWNED_WHY rendering its kind tokens the way _DORMANT_WHY does. Mutation-tested: dropping the order narrowing, the sorted, any continue, or returning early each fails exactly one named test.
3. UNPINNED CONTEST joins the exit code, fatal on both tiers, rows grouped by pair, after the MOVED SHAPE blocks; main()-level tests including that a _WATCHED_DIFFS row does not silence it and a radar name fails as a contract one does. Mutation control on the real ledger: deleting Smith, Ph. D. Jr.'s row in memory exits 1 naming the name, its winner and its loser; unmutated, no block prints at any baseline.
4. Records. A dated decisions entry with the population figures, the seven verdicts, the scope decision (the predicate stops at Should file order decide between rules whose fields overlap without nesting? (マイケル・ジャクソン goes to whichever of two rules comes first, and the contest check never sees the pair) #498's class; the wider narrow-first one costs 9 more names, 23 in all, and is declined with Six contested diffs in the 2.x ledgers have no pin, and the empty rosters read as if none existed (MD, PHD goes to whichever rule comes first) #501's gap (ii) left open by choice), and the coverage argument; the two 2026-09-02 bullets each gain a sentence pointing here; README's "What counts as a contest" carries the second predicate; mechanisms.md's cross-rule pins section says the roster's completeness is now checked rather than stated.

Scope decision

The predicate stops where the winner beats a loser with equal or non-nested fields. A pin for every undeclared contest, narrow-first ones included, would also have forced the argument #501's regex accident lacked, at the cost of pins the declaration-free default already justifies. Recorded as declined, with the figure.

Verification

Intentional counts 352 / 247 / 155 / 14 byte-identical at every commit (a pin moves no classification; a check that fires on nothing prints nothing), 0 unexplained, 0 radar-unclassified, no finding block at any baseline. Full suite green; ruff and mypy clean. Every commit went through a spec-compliance review and a quality review; the review rounds found and fixed one stale present-tense roster total, two mutation survivors, a fail-open .get, and four comment sentences a measurement contradicted. A final five-reviewer pass over the whole branch then closed two more: the pin gate now applies only to default-order comparisons, since a recorded shape is a default-order shape and cannot own a declared-order contest, and the function sorts its rules internally as dormant_rules does.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.66%. Comparing base (fca8bf4) to head (24fad28).
⚠️ Report is 5 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #507   +/-   ##
=======================================
  Coverage   98.66%   98.66%           
=======================================
  Files          45       45           
  Lines        3219     3219           
=======================================
  Hits         3176     3176           
  Misses         43       43           

☔ 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.