We present the Threshold-Locked Spectral Imager — a computational framework for geometry recovery from noisy spectral signals, built on the ITT persistence and closure language of Pre-Veil Mechanics Volume I (Knight 2026). Reflected signals are treated as compressed packets carrying recoverable structural information (coordinate sub-keys kₓyz). Recovery is reformulated as a threshold-crossing problem: a reconstruction is accepted only when retention dominates loss (Sₛel >= 1) and gradient, curl, and curvature closure metrics each cross minimum thresholds — corresponding to the three irreducible operators of the Collapse Genesis Stack (nabla Phi, nabla x F, nabla² Phi). Testing across four synthetic scene types (plane, ridges, crater, Chladni-like) at three noise levels (sigma = 0. 12, 0. 22, 0. 32) demonstrates that threshold-controlled decoding consistently outperforms a fixed baseline decoder. At the highest noise level, the baseline gate-pass rate collapses to 0% while the threshold controller recovers feasible solutions in 75% of trials. RMSE improvement widens as noise increases (29% improvement at sigma = 0. 32). The plane scene fails the gate by design — a deliberate feature exposing the framework's internal logic: the feasibility rule requires curl closure >= 0. 45, but a planar surface has near-zero curl structure, causing cosine similarity to degenerate. This failure is diagnosed precisely and the regime-conditional gate fix is specified. The prototype is a falsifiable computational bridge between the ITT sub-key language and practical spectral geometry decoding. Full Python code included and reproducible.
Armstrong Knight (Sat,) studied this question.