Simplified models of complex systems routinely remove small-scale structure to enable tractable computation. While filtering reduces resolution, it does not necessarily reduce underlying dynamics. This creates a structural risk: models may stabilise representation without stabilising reality. This paper introduces the Structural Honesty Gate (SHG), a quantitative diagnostic that evaluates whether removed layers remain dynamically subordinate to resolved behaviour. The gate defines a Filter Integrity Ratio that compares hidden-layer feedback to resolved dynamics. When hidden-layer influence exceeds a defined threshold, continued inference from the simplified model becomes structurally inadmissible. The SHG does not replace governing equations and does not introduce new physics. It functions as an integrity overlay that provides a principled stopping condition for inference under filtering, parameterisation, or regularisation. Applications include turbulence modelling (LES), climate simulation with subgrid physics, regularised inverse problems, data assimilation frameworks, and neural network training stability.
Andrew John Paton (Thu,) studied this question.