OutageMatch: Equal Missingness Does Not Preserve Selective-Sensing Decisions

Two sensor-loss timelines can hide the same number of observations while presenting different decisions: independent misses repeatedly reset cache age, whereas a recoverable burst restricts when stale evidence can be refreshed. OutageMatch turns this distinction into an evaluation protocol by representing availability, evidence age, and reacquisition legality, declaring the trajectory statistics preserved by each counterfactual, and separating process transformations from state-observability ablations. Its operator sequence first exposes where Bernoulli matching fails: exact-count and joint-state controls recover $0.8$ MUSIC-AVQA points by restoring realized counts and same-step co-failure statistics before the persistent path is considered. In the direct attribution test, replacing recoverable bursts with rate-matched Bernoulli masks lowers accuracy by $2.9$ points (95% CI [2.70, 3.10]), while controller removal lowers it by $0.8$ points ([0.60, 1.00]); the resulting control-attribution gap (CAG) is $2.1$ points ([1.92, 2.28]). Habitat ObjectNav yields the same comparison in SPL, with a CAG of $0.013$ ([0.0121, 0.0139]). As a process-aligned selective-sensing system, OutageMatch reaches $67.2\pm 0.3$ accuracy and $0.429\pm 0.006$ SPL, improving over a shared-interface POMCP planner by $1.0$ point and $0.012$ SPL with fewer requests. Thus the missingness rate alone does not define a valid control when sensing value depends on the path to recovery.