Final controlled realism pass
Open the unchanged browser test URL →
One edge-finish adjustment. No colour, geometry, camera, motion, lighting, wall or shadow redesign. The owner-tested pass 04 is preserved for rollback.
Cause and physical construction
The wall is at z = 0. The standoff barrels span 0–10 mm, the backing 10–13 mm, the face 13–16 mm and the caps 16–19 mm. Both acrylic layers are exactly 3 mm; the face and backing meet without an invented inter-layer gap. Mesh normals point outward and have unit length. The backing is genuinely 10 mm from the wall; it has not been moved.
The main light is above and left of the sign (−0.95, 0.9, 0.7). Its shadow projects down and right. The top and left are therefore the lit edges, and their wall gap is partly occluded in shallow views. The uniform matte wall also provides few parallax cues. This is a combination of perspective and lighting, not a missing standoff distance or misaligned contact texture.
The exposed acrylic cut walls inherited the front face’s 0.4 specular multiplier. That weakened the light response on the thin edges. This pass restores a normal dielectric reflection strength only on those existing side surfaces. Oblique views show the two layers; a near-side view within the existing manual orbit limit shows the actual barrel and air gap. No artificial top/left shadow was added.
Exact change
| Surface | Before → after | Reason |
|---|---|---|
| Acrylic outer and cutout side walls | Specular intensity 0.4 → 1.0. Roughness 0.12 → 0.18. | At the existing IOR of 1.49, normal-incidence dielectric reflection rises from approximately 1.55% to 3.87%. The slightly wider lobe avoids needle-like highlights. This is a surface-finish approximation, not a measured manufacturing finish specification. |
| Front acrylic face | Unchanged: specular 0.4, roughness 0.5, calibrated diffuse illumination and RGB. | Protect owner-approved Sage and Anthracite appearance. |
| Gold and Silver backing / hardware | Unchanged shaders, HDR, colour balancing, roughness and palette. | Protect approved metallic response and changing reflections. |
| Wall, light and shadows | Unchanged matte wall, light positions, exposure, cached VSM radius 3 and 25 mm contact maps. | Keep physically consistent shadow direction and prevent broad haze. |
There are no extra vertices, bevels, outlines, normal perturbations, draw calls, texture lookups, render targets or per-frame shadow updates. The change uses the existing side-wall mask and PBR shader. Existing front colour constants, both constructions, every prior asset, motion code, camera angles, limits, calibration, smoothing, controls and interface are unchanged. Only the build identifier and script cache key changed alongside the edge shader.
Construction and visual verification
Inspected both finishes in straight-on, left oblique, right oblique, top/left close-up and angled clearance views. In the side view the barrel spans the gap between the wall and the two-layer assembly. Reflections change with view angle; some directions reflect darker parts of the retained environment. Gold remains warm and Silver keeps its neutral balancing. The effect is intentionally subtle and cannot make every lit edge show an air gap in a frontal view.
Anthracite / Mirror Silver
Straight-on — before and after
Left oblique — before and after
Right oblique — before and after
Top and left edge close-up — before and after
Angled wall clearance — before and after
Sage Green / Mirror Gold
Straight-on — before and after
Left oblique — before and after
Right oblique — before and after
Top and left edge close-up — before and after
Angled wall clearance — before and after
Preserved Unreal comparison
This pair uses Anthracite/Silver, two standoffs, the same canonical geometry hash, the saved camera angles and 774 × 1086 viewport. The archived screenshot and saved state were checked against the preserved reference evidence. Unreal’s brighter fine rim and different lighting remain visible; this pass does not copy an outline or change the approved browser palette to force a match.
Comparison limit: the archive has one exact locked pose. The other available Unreal images are historical and not matched final-V1 captures. All requested browser views were inspected, but a fresh six-view matched Unreal comparison and physical Android performance test were not performed. The Unreal service and capacity investigation were not accessed. Existing import/source triangle-count metadata differences remain unresolved; no current Unreal runtime mesh was re-exported.
Performance and protection
Matched 20-second high and low-quality tests in desktop Chrome on NVIDIA Quadro K2200, 390 × 844 touch emulation at DPR 2, using synthetic 60 Hz orientation events. These are not physical Android battery, thermal or motion-to-photon measurements.
| Metric | Checkpoint 04 | Pass 05 |
|---|---|---|
| high · Average FPS | 59.95 | 60.00 |
| high · p95 frame interval | 16.8 ms | 16.8 ms |
| high · Worst frame interval | 40.1 ms | 17.1 ms |
| high · First visible frame | 1972 ms | 1833 ms |
| high · Resource transfer (local) | 8,276,002 bytes | 8,276,265 bytes |
| high · mean synthetic event → render | 8.55 ms | 8.57 ms |
| low · Average FPS | 60.00 | 60.00 |
| low · p95 frame interval | 16.8 ms | 16.8 ms |
| low · Worst frame interval | 16.9 ms | 16.9 ms |
| low · First visible frame | 1047 ms | 1001 ms |
| low · Resource transfer (local) | 3,373,607 bytes | 3,373,870 bytes |
| low · mean synthetic event → render | 8.40 ms | 8.43 ms |
Both modes stayed approximately 60 FPS. The baseline high-quality run contained one 40.1 ms frame; this is not evidence that the new build is faster. No sustained regression was observed. First-frame times are single local loads, not mobile-network predictions. Loaded resource size increases by just 263 bytes of script text; all asset sizes are unchanged. Both builds use 9 draw calls and 84,258 triangles. High quality retains 10 geometries / 6 textures; battery saver retains 9 / 4.
All 18 automated interaction checks passed, including calibration, sensitivity, camera bounds, monotonic smoothing, recentering, touch drag, pinch, fullscreen, landscape, colour changes, quality switching and both canonical constructions. No page or shader errors were found. Flat patches in matched front images stayed within one RGB level of the checkpoint; mean differences were below 0.000001 of an 8-bit channel level. Changes are confined to the intended edge finish and tiny rasterisation differences.
Verified rollback and approval boundary
Rollback checkpoint: commit 0894dc2, tag browser-owner-tested-pass04-20261016. The full browser ZIP was reopened and every one of its 88 files verified by SHA-256; the complete Git bundle also passed verification. The previous immutable deployment is available here. Restoring that checkpoint to the same isolated Pages project requires no Unreal or production change.
Source and asset protection · Layer / normal audit · Interaction checks · Before performance · After performance
Stopped for owner approval. No Browser V1 approval tag, catalogue work, Unreal modification or production deployment is included.