iter 15 — threshold-axis transfer: resolution-localized redundancy discovered

BREAKS → DEFENDED → CLOSED 2026-07-10 · the transfer trilogy's last axis (substrate · time · resolution) · a real market-microstructure property found · driving-cell attribution added · clean 0/3

Claim under attack (assumed true only to destroy it)

"The §B worked-example verdicts are resolution-stable — every single-threshold slice of the grid (100/250/500 dbps) reproduces the declared verdicts." Iter 1 proved substrate transfer, iter 9 time transfer; every aggregation so far POOLED bar resolutions.

ATTACK RESULT — falsified, by a real property of markets

`price_impact \| volume` (declared BAN)	@100	@250	@500	@750
ξ_worst per slice	0.9992 → BAN	0.8974 → WATCH	0.8206 → WATCH	0.7446 → WATCH

Redundancy can be resolution-localized. The declared BAN is driven entirely by the 100-dbps slice — mechanically sensible: at fine bars price_impact ≈ f(volume); coarser aggregation decouples them. All 6 other pairs reproduce in every slice (19/21 checks; PASS pairs PASS everywhere including 750). The §B worst-cell machinery behaves exactly as declared ("redundant in ANY cell ⇒ flag") — the v1 oracle demanded an invariance the declaration never promises (iter-14 oracle-class precedent, corrected honestly).

DEFENSES

Driving-cell attribution (mechanism, iter-13 diagnosability class): aggregate_worst_cell now emits xi_worst_cell per candidate — the operator can SEE when a flag is resolution-localized instead of it being buried in an aggregate.
Sound oracle (certification stability): the durable resolution guarantee is that no single-threshold slice CERTIFIES (PASS) a pair the full grid flags, nor FLAGS a pair the full grid certifies.

Proof

Check	Result
G1 — certification stability, 28 slice checks (750 included)	28/28: no slice certifies a flagged pair; no slice flags a certified pair
G2 — attribution through the probe's own path	`xi_worst_cell = AAVEUSDT@100` emitted for the producing kernel (2-cell aggregate run)
§B declaration self-test	PASS

Evidence

File	What
a15_threshold_transfer_attack.py / a15_results.json	v1 — per-slice §B readings for all 7 pairs × 4 thresholds (the discovery data).
a15_threshold_transfer_attack_v2.py / a15_v2_results.json	v2 — sound oracle + attribution verification; 0 breaches (final).
a15_stderr.log · a15_v2_stderr.log	Run logs under the 2 CPU / 2 GB scope.
regression_tests/test_a15_threshold_transfer.sh	Permanent regression test — refreshes measurements, asserts the sound oracle.

Consequence

The transfer trilogy is complete: substrate ✓ (iter 1), time ✓ (iter 9), resolution — now understood precisely: certifications transfer; flags may be resolution-localized, and the probe now says where. This materially informs the terminal head-to-head (a candidate's redundancy may live in one resolution only — the legacy probe, which pools less granularly, cannot see this). Clean rounds: 0 of 3.

iter 15 · threshold-axis transfer · verdict BREAKS → DEFENDED → CLOSED · one atomic commit: iter 15: threshold-axis transfer — BREAKS→DEFENDED · append-only.