Readout and Calibration-Update Audits for First-Order Probes: False Alarms, Detection Sensitivity, and Decision Coverage under Source Shifts and Calibration Contamination

Sensors measuring the same input can produce different normal records because their response times differ. Physical-model residuals help account for these differences, while alarms from finite records additionally require thresholds calibrated to normal fluctuations. When thresholds are updated after a change in source statistics, fault contamination in the candidate data can reduce both false alarms and detection sensitivity. For a pair of calibrated first-order probes, we construct an evaluation procedure that separates readout, candidate-update auditing, and normal-condition validation of the adopted calibration. The procedure uses a signed moment residual, a cross-relation residual, and their fixed linear fusion. Analysis of the response model, simulation-based mechanism studies, and a confirmation experiment with 50 independently regenerated reference–calibration–evaluation replicates establish three findings. First, retaining intermediate representations makes it possible to trace alarm suppression at the upper limit of finite ranks. Direct calibration avoids this restriction in some conditions, but also incurs detection losses in others. Second, positive- and negative-gain contamination can cancel in a mean residual. Adding private-input records—records with an independent signal entering only one probe—can make a candidate more likely to be withheld from threshold updating while further reducing the sensitivity of thresholds fitted to that candidate. Third, decision deferral and recovery based on independent normal data affect false-alarm evidence, sensitivity, and decision coverage separately. In the primary recovery condition, a clean calibration provided decisions in 45 of 50 replicates; 10% of records remained without a decision. Relative to continued use of the old calibration, all-record detection of positive and negative gain faults decreased, while the difference in all-record private-input detection remained unresolved after multiplicity adjustment. The study provides a traceable inspection procedure and evaluates its benefits and costs separately under specified conditions.