Phase Sampling Domain Update¶
Exploratory sampling-domain analysis only. No renderer changes, hit-selection changes, or simulation reruns were performed. Raw validation truth remains separate from coherence previews.
Executive Summary¶
The 2026-04-25 phase/coherence probes support a consistent sampling-domain picture: the visible bands align with real field-geometry structure, but no single global polar center explains them.
Phase-guided hit-selection proxies slightly improve local phase scores but do not strongly reduce band-boundary alignment. Incoherence-centered polar tiling slightly improves bridge/backstep direction fidelity relative to aperture-centered polar tiling. Curvature-center polar tiling improves direction similarity over aperture-centered polar on both tested checkpoints, but loses distance alignment and remains well below adaptive-square directional fidelity. Adaptive-square tiles remain the best local direction-preserving texture.
Inputs¶
phase_guided_first_hit_selection_2026-04-25/band_reduction_metrics.jsonincoherence_centered_polar_2026-04-25/incoherence_polar_summary.jsoncurvature_center_polar_2026-04-25/curvature_center_summary.jsonpolar_edge_comparison_summary.jsonedge_alignment_summary.json
All inputs are existing artifacts under:
output/fixture_runs/fixture_011_wormhole_checkpoint_sequence/2026-04-24T01-44-23_first_hit_baseline_current/
Result Summary¶
The phase-guided first-hit proxy used existing accepted-hit diagnostics rather than true renderer candidate lists. It reduced mean local selection score by 0.0401 on average and reduced visible-band/incoherence correlation by -0.0325, but the visible-band phase-boundary fraction changed only -0.0070. This is useful as a tie-breaker signal, not a standalone band-reduction fix.
Incoherence-centered polar tiling used a centroid estimated from visible-band mask, neighbor-normal-delta, and phase-incoherence signals. It retained the high recall behavior of aperture-centered polar tiling and slightly improved bridge/backstep direction similarity (0.656 vs 0.642), but did not approach adaptive-square direction fidelity.
Curvature-center polar tiling selected centers from Hough circle/arc and contour-fit candidates. It improved direction similarity relative to aperture-centered polar on both checkpoints (mouth: 0.656 vs 0.628; bridge: 0.660 vs 0.642) but reduced symmetric edge-distance score relative to aperture/incoherence polar and adaptive-square baselines.
Direction Fidelity Comparison¶
| method | mouth direction | post-throat direction | key verdict |
|---|---|---|---|
| aperture polar | 0.628 | 0.642 | High recall baseline, weak direction fidelity |
| incoherence polar | 0.624 | 0.656 | Slight bridge direction gain, not global |
| curvature-center polar | 0.656 | 0.660 | Best polar direction result, weaker distance alignment |
| adaptive square | 0.836 | 0.875 | Best local direction-preserving texture |
Current Architecture Doctrine¶
- Raw row pass = scout truth.
- Adaptive-square tiles = best local direction-preserving texture.
- Polar, incoherence-centered polar, and curvature-center polar = high-recall geometric probes.
- Future smart mode should support hybrid and multi-center sampling textures.
- Validation truth must remain separate from coherence previews.
Verdict¶
The field geometry is real in the sampling domain, but the band structure is not explained by a single global polar center. Polar recentering can improve geometric probe alignment slightly, especially for bridge/backstep direction fidelity, but adaptive-square tiles remain the strongest directional match. The next useful sampling texture is likely hybrid: adaptive-square for local direction, plus multi-center annular/curvature probes for high-recall geometry diagnostics.
Artifact References¶
- Phase-guided first-hit proxy:
phase_guided_first_hit_selection_2026-04-25/band_reduction_metrics.json - Incoherence-centered polar:
incoherence_centered_polar_2026-04-25/incoherence_polar_summary.json - Curvature-center polar:
curvature_center_polar_2026-04-25/curvature_center_summary.json - Existing aperture/adaptive comparison:
polar_edge_comparison_summary.json