Definition
The weighted mental effort a technician expends to reach a correct diagnosis, combining diagnostic phase duration, navigation transition count and entropy, and an optional short workload self-report: normalized against diagnostic accuracy.
The Measurement Gap
Wrench-time studies, labor utilization, and MTTR were designed when diagnostic work was physical. AI-assisted diagnostics is cognitively demanding work that looks like "standing around", a technician sorting anomaly scores, factor importance displays, and maintenance history is doing the highest-value work of the shift, but it does not generate a wrench-time signature. A 45-minute MTTR can mask either skilled evaluation or automation complacency. OEE measures throughput. PM Compliance measures schedule adherence. None of these capture whether the diagnostic process is sustainable or whether the best troubleshooters are quietly burning out under loads no metric tracks.
Decision Supported:
Where to invest in interface design, explanation quality, training, or workload redistribution to keep the diagnostic process sustainable as AI assistance scales.
Formula
Plain language: Combine objective signals of cognitive effort (how long the technician spent investigating, how many screens they consulted, how chaotic the navigation pattern was) with an optional short self-report, then divide by diagnostic accuracy. Two technicians producing the same correct answer may have very different cognitive costs.
Notation:
CLDR = (w₁ · T_diagnostic + w₂ · N_transitions + w₃ · TLX_micro + w₄ · H_navigation) / D_accuracy
Variables:
| Variable | Definition | Unit |
|---|---|---|
| T_diagnostic | Elapsed time from work order assignment to diagnostic decision (pre-action phase) | Minutes |
| N_transitions | Distinct screens or information sources consulted during diagnosis | Count |
| TLX_micro | 3-question NASA-TLX micro-survey (mental demand, time pressure, frustration) at WO close | Score (0–60, normalized) |
| H_navigation | Shannon entropy of UI navigation sequence: H = −Σ p(i) · log₂(p(i)) | Bits |
| D_accuracy | 1 if failure report complete and no 30-day repeat; 0.5 if repeat; 0 if reopened | Ratio |
| w₁–w₄ | Regression-calibrated weights summing to 1.0 | Weights |
Proposed formula, pending validation pilots.