Evidence Gating for Small-Displacement Fault Attribution in Composite Coal-Seam Reflections

Small fault offsets in thin coal seams can resemble responses caused by tuning, impedance variation, processing effects, or poor data quality. We developed the Scale- and Dependence-Aware Evidence-Gating workflow (SDEG) to separate candidate detection from fault attribution, treat products from the same seismic volume as dependent evidence, and retain uncertain and not-evaluable outcomes through non-compensatory gates. Evaluation combined one- and two-dimensional modeling, a descriptive three-dimensional spatial-response transfer experiment, a reference-masked five-reader crossover study using a separately assembled 93-case numerical cohort, controlled rule perturbations, and a masked field experiment. Under the prespecified reader–case analysis, the prespecified reader-comparison criterion was met: coverage-adjusted balanced accuracy increased from 0.692 to 0.792 (paired difference, 0.099; 95% confidence interval, 0.006–0.191), while active interpretation time increased by 13.7%; robustness to parent–model–family clustering could not be assessed. Normalized throw organized model-side responses, but the prespecified scale-stability criterion was not met, and observable-scale routing remained unconfirmed. Rule perturbations supported uncertainty preservation and non-compensatory integration. In the common-complete analysis of 31 of 36 field objects, Gwet’s AC1 increased from 0.144 to 0.588, indicating greater inter-reader reproducibility within one survey, not geological accuracy. SDEG provides a transparent framework for recording attribution decisions, evidence dependence, and unresolved outcomes.