Short industrial compressed-air audits can contain thousands of samples while representing only a few dependent operating episodes. Autocorrelation therefore makes nominal sample count an unreliable basis for deciding whether a low-flow regime is sufficiently supported for an energy-efficiency or maintenance action. This paper proposes an audit-to-decision procedure separating three claims: that a low-flow regime exists, that it is measured reliably enough, and that its relative scale justifies action. hv-block selection and minimum-dwell reconstruction are combined with episode repeatability, effective intra-regime size, resampling uncertainty, metrological distinguishability, and relative non-productive consumption in a six-level action scale (D1–D6). The procedure was applied to 29 records from three operating plants. Five PLC-labeled records provided a limited external evaluation of the operational meaning of the identified low-flow state. Nominal BIC and hv-block selection differed in 20 of 24 audit records. Median deviation fell from 32.2–33.9% for nominal-selection baselines to 2.9% when the HMM used the hv-block-selected regime count. Outcomes were 3 × D1, 15 × D2, 7 × D3, and 4 × D4. D-codes were unchanged in 23/24 threshold-tested and 14/16 pressure-augmented records. In the five-record PLC-labeled evaluation, no action unsupported by the labels was triggered. The strongest time-series-only outcome is a mandatory diagnostic work order, not an automatic repair decision.
Limited diagnostic resources require maintenance tasks to be prioritized. Under compensatory multi-criteria decision analysis (MCDA) alone, a machine with larger flow-based criterion values can outrank one for which the available evidence prescribes a different action. This study couples a previously developed evidence-sufficiency procedure with compensatory prioritization through a lexicographic rule. The D-code fixes the admissible action for each machine, and MCDA ranks only tasks of the same type. In eight short machine-level records from one plant, the MCDA-only order places two D2 tasks ahead of the single D4 diagnostic task, which the two-stage order places first. This conflict persists at all 256 combinations of the interval bounds of the flow-based criteria. A retrospective ultrasonic survey localized 12 static leaks on four machines. In a retrospective replay with a common localization modality, the two-stage order covers less at one visit, 33.3% versus 41.7% of the leaks and 23.0% versus 50.8% of the semi-quantitative estimated equivalent flow. At two visits, it covers 75.0% versus 50.0% and 73.8% versus 68.5%. At nominal inputs, the selected sets coincide at three and four visits. Within D2, MCDA did not outperform a flow-only ranking. Within the proposed framework, the admissible action is established before compensatory prioritization.