## 1. The Empirical Case for Negative-Space-Aware Pattern Recognition ### 1. 1 Why this section comes first This paper specifies a Pattern Recognition Engine (PRE) whose architectural commitments — context-light Coalescence–Entropy Boundary detection, autonomic operation, seamless void-positive substrate handling, family-aware null routing, recognition-as-trajectory — are not aesthetic preferences. They are the engineering response to a measured failure mode in standard interpolation and pattern-recognition methods. The failure mode is *destructive interpolation*: the systematic destruction of negative-space structure by methods that treat absence as something to be smoothed over rather than as load-bearing geometric content. The destruction is measurable, severe, reproducible, and worsens monotonically with the sophistication of the interpolation method used. Section 11 v2 of the empirical testbed (Destructive Interpolation Paper v2) demonstrates this failure mode across five interpolation/recognition methods on a controlled substrate with typed nulls and known structural features. The results establish, empirically, that: (a) Standard interpolation methods (Nearest, Linear, IDW p=4, Gaussian RBF) fail to preserve negative-space integrity at all tested removal levels (10%, 30%, 50%) ; (b) The most sophisticated standard method (Gaussian RBF) is the worst offender, falling well below the negative-space integrity floor (N ≥ 0. 7) and exceeding the warm-water risk ceiling (W ≤ 0. 4) at all removal levels; (c) NSEV2 — the Negative-Space-Aware engine prototype that operationalizes a predecessor subset of the principles formalized in this paper — preserves negative-space integrity at N = 1. 0 and warm-water risk at W ≤ 0. 10 across all tested removal levels; (d) NSEV2 is the only method that promotes true features without promoting distractors; (e) NSEV2 alone preserves structural features (cliffs and ridges) below the warm-water ceiling. These results justify the architectural commitments that follow. They are not validation-after-the-fact. They are the *reason* for the design. ### 1. 2 The empirical setup The Section 11 v2 experiment evaluates five methods against a controlled substrate containing: - Positive scalar content (the standard signal that conventional methods are designed to handle) - Pit interiors (regions of typed-null content where the architecture's negative-space integrity must be preserved) - Structural features (cliffs and ridges — sharp transitions between distinct regions that must not be smoothed) - True features and distractors (signal content that must be promoted vs. noise that must not) Removal levels of 10%, 30%, and 50% simulate the substrate degradation that any realistic pattern recognition engine must handle. The tuple-aware variant (used in v2) ensures that the test is sensitive to the typed structure of the substrate rather than only to scalar values. Two metrics anchor the evaluation: **N (Negative-Space Integrity): ** the fraction of typed-null content that survives the recognition pass with its family membership and ε/Z structure intact. The integrity floor is N ≥ 0. 7; methods falling below this floor are destroying void structure in ways the architecture cannot tolerate. **W (Warm-Water Risk): ** the degree to which the method smooths over distinctions that should remain sharp. The ceiling is W ≤ 0. 4; methods exceeding this ceiling are generating warm-water contamination — the smoothing-as-corrosion failure mode named in PRE v3. 1 and operationalized here. ▿ The N floor (0. 7) and W ceiling (0. 4) values are inherited from PRE v3. 1's empirical calibration. Phase B testing should validate whether these thresholds remain appropriate on the post-ROSA, post-ADR-003 substrate, or whether tighter bounds are now achievable. (Open: empirical calibration of threshold values. )
KIMBERLEY LAVERNE ASHER (Wed,) studied this question.