Background: Goodhart collapse describes a failure mode in which an optimisation system continues to optimise a metric after the proxy relationship between that metric and the intended objective has degraded. This restricted-access deposit records Paper 116 and companion technical notes from the AxoDen / Epistemic Fabric research programme on proxy-validity failure, metric gaming, and structural detection of Goodhart-style collapse. Methods: The work treats the proxy-validity assumption as an explicit object in a provenance-aware reasoning structure. It introduces a Goodhart Erosion Index for monitoring proxy degradation, a signature-vector formulation for separating external drift from system-induced gaming, and companion analyses covering noise stability, adversarial-bias bounds, empirical separability, and high-dimensional centrality via heat-kernel methods. Results: The bundle reports reproducible toy-model validation under the stated benchmark assumptions, including stable-noise rejection and drift or gaming classification in the seed-42 Monte Carlo setting. The empirical separability result is scoped to the toy-model regime. Analytical claims are labelled according to their supporting assumptions and open obligations. Conclusions: The deposit establishes a restricted prior-art record for treating Goodhart collapse as a structural proxy-validity failure rather than only a metric-monitoring problem. It is intended for controlled scholarly, diligence, or intellectual-property review. It does not claim production readiness, real-world deployment performance, or generalisation outside the stated formal and toy-model conditions.
Erkan YALÇINKAYA (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: