Disclose the shared band rule behind the finer-scan scores, from Scrolls main 9657b18 (PR #329)

README.md

@@ -26,6 +26,7 @@ On fresh test sites, 91% of snapped PHerc0139 vertices sit within one voxel of t
 1.129 µm scan, against 24% for the published labels.
 On 71 fresh PHerc0814 and PHerc0841 sites, the single-scan snap moved labels 2.15 voxels closer to the recto face of
 a finer scan than the published labels, and closer than seven community snaps run on the same sites.
+Scored against a finer scan's face found with labelsnap's own band rule; with a face rule sharing neither, the lead held at about half the size (+1.04, +1.23 voxels).
 
 - **Viewer**: pick a segment and a z plane; orange is the published label, blue the snapped one, gray strokes join
   matching points; Animate snap moves each vertex from its published position to its snapped one, slowly. Arrow keys

index.html

@@ -86,7 +86,7 @@
 </header>
 
 <section class="stats" aria-label="Measured accuracy">
-  <div class="stat"><b>+2.15 voxels</b><span>closer to the recto face than the published labels on 71 fresh PHerc0814 and PHerc0841 test sites, with the single-scan snap. Seven community snaps, run on the same sites: +0.26 at most.</span></div>
+  <div class="stat"><b>+2.15 voxels</b><span>closer to the recto face than the published labels on 71 fresh PHerc0814 and PHerc0841 test sites, with the single-scan snap. Seven community snaps, run on the same sites: +0.26 at most. Scored against a finer scan's face found with labelsnap's own band rule; with a face rule sharing neither, the lead held at about half the size (+1.04, +1.23 voxels).</span></div>
   <div class="stat"><b>91%</b><span>of scorable vertices within 1 voxel of the recto face in a 1.129 µm scan, on 60 fresh PHerc0139 test sites (sites it could not score left out). Published labels: 24%.</span></div>
   <div class="stat"><b>71%</b><span>within 1 voxel on 60 fresh PHerc0814 test sites with the single-scan snap. Published labels: 24%.</span></div>
   <div class="stat"><b>+0.61 voxels</b><span>closer to the surface: a 3D surface model trained on snapped labels against the same model trained on published ones, scored on a scroll neither saw.</span></div>
@@ -258,7 +258,7 @@
   <p>Truth is the recto face seen in an independent 1.129 µm scan, at test sites drawn after the method was fixed and never used before. PHerc0139, 60 sites: within 1 voxel 23.7% for the published labels, 79.4% for labelsnap, 91.2% for finesnap. PHerc0814, 60 sites: 24.2% for published labels, 71.0% for labelsnap. A 3D surface model trained on snapped PHerc0139 labels placed the surface 0.61 voxels closer on PHerc0814 than the same model trained on published labels (95% interval 0.39 to 0.84; all 9 seed pairings).</p>
   <h2>Against seven community snaps</h2>
   <figure class="fig"><img src="https://huggingface.co/datasets/SharpMeow/vesuvius-snapped-labels/resolve/main/community.png" width="1920" height="1200" alt="Bar chart: distance gained toward the recto face over the published label, 71 fresh PHerc0814 and PHerc0841 sites pooled. labelsnap +2.15 voxels; one fixed offset for every vertex +0.77 (reference); seven community snaps from +0.26 to -0.41, and vc_objrefine with its shipped thresholds -9.72, off the scale." loading="lazy"></figure>
-  <p>On 29 October 2026 the single-scan snap (labelsnap, the method used here for PHerc0172) ran once on fresh test sites of PHerc0814 and PHerc0841 next to seven community snaps, under a rule fixed before it ran. Truth is the recto face in a finer scan of the same scroll (1.129 µm for PHerc0814, 2.403 µm for PHerc0841), registered per site; 71 sites were scored. Pooled, labelsnap moved labels 2.15 voxels closer to that face than the published labels (95% interval 1.82 to 2.51) and 1.89 closer than VC3D's vc_objrefine with thresholds tuned on PHerc0139 (1.48 to 2.35). It was also closer than vc_objrefine as shipped, recto-offset r2 with and without its gate, surface_placement_audit, RectoRefine's midline and a snap onto a surface prediction, on each scroll and pooled. Within 1 voxel: 55% (PHerc0814) and 66% (PHerc0841) for labelsnap, 19% for the published labels on both, 22% and 21% for the best community snap. Every control held: reversing the move pushed labels away, a random sign gained nothing.</p>
+  <p>On 29 October 2026 the single-scan snap (labelsnap, the method used here for PHerc0172) ran once on fresh test sites of PHerc0814 and PHerc0841 next to seven community snaps, under a rule fixed before it ran. Truth is the recto face in a finer scan of the same scroll (1.129 µm for PHerc0814, 2.403 µm for PHerc0841), registered per site; 71 sites were scored. Pooled, labelsnap moved labels 2.15 voxels closer to that face than the published labels (95% interval 1.82 to 2.51) and 1.89 closer than VC3D's vc_objrefine with thresholds tuned on PHerc0139 (1.48 to 2.35). It was also closer than vc_objrefine as shipped, recto-offset r2 with and without its gate, surface_placement_audit, RectoRefine's midline and a snap onto a surface prediction, on each scroll and pooled. Scored against a finer scan's face found with labelsnap's own band rule; with a face rule sharing neither, the lead held at about half the size (+1.04, +1.23 voxels). Within 1 voxel: 55% (PHerc0814) and 66% (PHerc0841) for labelsnap, 19% for the published labels on both, 22% and 21% for the best community snap. Every control held: reversing the move pushed labels away, a random sign gained nothing.</p>
   <p>Limits: PHerc0814 is held out by site, not by scroll. The test shows labels moved closer to the face, not every vertex within 1 voxel. A label displaced toward the verso was not shown to come back. The midline and nearest-peak snaps aim at the sheet center by design, so part of their loss is that target. Where a finer scan exists, finesnap did better than labelsnap on PHerc0814 by 0.93 voxels. The released PHerc0841 meshes were not scored here, because their finesnap read the same 2.403 µm scan this test used as truth.</p>
   <h2>What is in each segment folder</h2>
   <ul>