Tracking Endogenously Moving Marginal-Stability Surfaces: Local Tracking Reserve, Patch-Localized Escape, and Feedback-Reactive Boundary Motion

Moving-boundary failure is established territory in rate-induced tipping, transition-path analysis, and safety control. This paper asks whether independently calibrated geometry, restoration, service margins, and feedback channels improve prospective prediction of first-failure time and location. We distinguish subsystem spectral marginality from finite-time growth and from service failure. Constructive and hidden-mode counterexamples delimit physical identification. Conditional comparison arguments treat moving margins, retained memory, and a restricted stochastic first-hit bound; calibration separates boundary position from its derivatives. Spatial evaluation decomposes joint first-event scores so timing gains cannot masquerade as location gains. Direct-input deletion, nominal-schedule replay, and parameter clamping remain distinct policies. Exact controls, reproducible simulations, and external quantitative comparisons support these distinctions, but establish neither empirical GGT4 validity nor predictive superiority. The contribution is a testable representation and intervention protocol whose practical value requires equally informed, budget-matched comparisons. Evidence status. Conditional analysis, reproducible implementation checks, and a targeted external quantitative comparison. External physical experiments and published predictor scores are attributed to their authors, not to GGT4. The proposed M1-M3 comparisons on learned AI or organizational systems have not been run; their practical effect sizes remain unestimated. Note on this release. Version 1.0 working paper, dated 19 September 2026 (about 14,900 words). The upload contains the manuscript and a supplement archive with the reproduction scripts and their reference outputs: reproduce.py and external_checks.py (all 393 numeric entries reproduce the preserved results), the analytic controls, and five additional check scripts covering the mathematical, memory, stochastic, empirical, and physical statements, together with an external-source evidence ledger, README, MANIFEST.json, and SHA-256 checksums. These are implementation and identity checks of the declared models; no external raw-data fitting, model retraining, or physical experiment was performed, and reproducibility is not evidence of empirical validity. Series Paper IV of the Governance Geometry Theory (GGT) papers within the author's Deficit-Fractal Governance (DFG) framework. The moving-boundary question originates in the author's GGT parent note, which remains unpublished at the time of this release; this manuscript does not inherit that note's asserted equivalences, fractal generation, or cascade exponents.