nx_scene_through_aperture.nx

nx_scene_through_aperture.nx -- Tier 6 background-coherence detector. User feedback 2026-05-16: "the car isnt real and the background is a mess". AI image-gen frequently fails the SCENE-THROUGH- FRAMED-APERTURE check: when looking through a window or door, the content visible beyond should follow physical rules: - sharper foreground, blurrier background (depth-of-field) - cooler / hazier colors (atmospheric perspective) - consistent perspective (vanishing lines align with frame) - recognizable objects, not arbitrary blobs (edge-density distribution typical of real scenes) AI puts random fragments through windows because the gen model has no scene-coherence prior beyond local pixel statistics. Substrate's answer: measure the four physics-grounded signatures and emit a sealed verdict. Inputs: caller pre-extracts the APERTURE region (window/door bounding box content) and the IMMEDIATE FOREGROUND region (interior wall / dashboard / face beside window). Computes: sharpness_ratio_q10 = sharpness(aperture) / sharpness(foreground) color_temp_shift = mean(aperture_b - aperture_r) - mean(foreground_b - foreground_r) edge_density_q10 = edge_count(aperture) / aperture_pixels edge_uniformity_q10 = std(local_edge_density) / mean(local_edge_density) Physics expectations for real-photo background: sharpness_ratio in (0.2, 0.8) -- background blurrier color_temp_shift > 0 -- background bluer (atmospheric) edge_density in (0.05, 0.30) -- some structure, not flat, not noise edge_uniformity > 0.4 -- non-uniform distribution Sealed verdict: COHERENT -- all 4 checks pass DEPTH_FAIL -- sharpness ratio outside physical range ATMOSPHERIC_FAIL -- no color-temp shift FRAGMENTED -- edge metrics outside real-scene distribution IMPLAUSIBLE -- 3+ checks fail (likely AI background) genealogy_id: substrate_scene_through_aperture_2026_05_16