Programmable matching mechanisms decide which orders become data. A loss metric computed from a mechanism's own transaction tape therefore mixes execution quality with selection. We study when loss-per-service rankings from policy-selected marked transaction tapes can be sign-certified under hidden-opportunity and sampling uncertainty. The tape identifies a selected compensator, $e^{π_0}ν^\circ$, plus martingale fluctuation; it does not identify hidden opportunity composition. We give an exact native two-tape non-identification construction, sharp signed hidden-mass ranking regions, and projection-based confidence sets that certify a sign or abstain with a region-wise denominator screen. In a two-year WETH--USDC application, point estimates order pools but the descriptive cross-pool ranking is not sign-certified. The reported comparison is a selected-tape ranking, not a dynamic mechanism counterfactual.

翻译：暂无翻译