Isolated contact-map difference, amplified 16× for visibility
Same locked geometry, camera and materials; only the wall-contact texture changed. This diagnostic is not a customer image.
MME DIRECT · OWNER REVIEW · 16 OCTOBER 2026
Return to isolated catalogue proof
The capture pipeline does not lose the final acrylic-edge shader. The old PNGs were made after approval. We found a small contact-map adaptation difference and mismatched comparison configurations; we have not established that these fully explain the owner's perceived flatness. No catalogue rollout or checkout approval is implied.
Locked release MME-BROWSER-3D-V1-OWNER-APPROVED, commit caeb8856461e6b43421a9dd9ac8f271dc1e0c02e, build browser-quality-20261016-05. Frozen app SHA-256: 8071d2d3b9d2dd09927a021f7b4c5722fc024256f95fde409a644ef680d3d73a. All 16 package files and 121 catalogue-source files remain unchanged.
The earlier PNG receipt timestamps are 08:55:56–08:57:26 UTC, after the 07:35:41 UTC approval-package creation. Their renderer source hash is 1eb0aec62bdb8cd8e7f17368e3fbd50d61e856f8b9ab858ea14b31ad9f88c402. That source contains final pass-05 side-wall specular intensity 1.0 and roughness 0.18, calibrated Sage/Anthracite colours, neutralised face irradiance and Silver lighting balance. The masters are not pre-correction renders; they have not been discarded or regenerated wholesale.
Same locked geometry, camera and materials; only the wall-contact texture changed. This diagnostic is not a customer image.
These two pairs use the same approved 82 Maple Road geometry, standoff mounting, initial camera and 1024 × 826 output. Both captured PNGs match their frozen V1 reference PNG pixels exactly. The pairs below use responsive WebP display copies; full PNGs remain available in the evidence. The reference fixture is kept separate from available catalogue offers; it cannot be used to bypass manufacturing admission.
Open the Sage/Gold V1 fixture and capture locally · Open the Anthracite/Silver fixture · Original approved V1, unchanged
Capture and animation now call shared camera-pose and viewport functions extracted verbatim from the viewer. The capture no longer maintains a separate trigonometry recipe. The initial orbit is yaw −16.60838°, pitch 0°, zoom 1; the base position is [0.2634919882, 0.0291114151, 0.5888126607] metres. Target z is 0.013 m for standoffs or 0.007 m for adhesive. The existing portrait/landscape projection rule, perspective and catalogue family framing factor are retained. Receipts record position, quaternion, target, FOV, projection matrix, framing factor and resolution. Camera state is restored after capture, including when the renderer is paused.
All seven comparisons have identical live/capture camera matrices. The locked standoff fixture retains wall z=0, backing 10–13 mm, face 13–16 mm, caps 16–19 mm. Face and backing are each 3 mm, touch without an invented gap, and retain their source normals. The five family samples retain their validated canonical mesh and mounting evidence.
1024 × 826; native Chrome / Quadro K2200; shared workshop workload, not idle-GPU benchmarking. One scene at a time, paused between captures, with two-second gaps. Timing includes PNG and WebP encoding; production-style WebP-only results from the previous device-tier report remain separate.
| Sample | Changed channels / maximum step live → still PNG | Camera | Capture ms | PNG / WebP bytes |
|---|---|---|---|---|
| locked-bee-SG | 0 / 0 | Exact | 192.7 | 248,818 / 35,026 |
| locked-bee-AN | 2 / 1 | Exact | 165.2 | 230,802 / 33,664 |
| vertical-bee | 0 / 0 | Exact | 155.7 | 138,210 / 17,656 |
| DATUM_LANDSCAPE_V1 | 0 / 0 | Exact | 138.5 | 148,052 / 19,636 |
| MONOLITH_VERTICAL_ADDRESS_V1 | 0 / 0 | Exact | 154.5 | 119,557 / 13,118 |
| tree-of-life | 0 / 0 | Exact | 154.9 | 247,376 / 38,466 |
| coastal-lighthouse | 0 / 0 | Exact | 169.2 | 139,657 / 17,982 |
Six pairs are pixel-identical. The successive live/capture Anthracite pair differs in two channels by one level each out of 2,537,472 RGB channels; its captured PNG is pixel-identical to the independently loaded frozen V1 reference. WebP is lossy, so exact pixel equality applies to PNG masters. Cross-backend SwiftShader/Quadro differences are shown below, not attributed to a shader correction.
The existing 384 × 310, 768 × 620 and 1024 × 826 responsive tiers remain; 1536 × 1239 is for explicit inspection. The new Bee reference WebPs reuse its verified 1024 PNG master. No additional maximum-size catalogue matrix was generated. New personalised cache keys include the canonical configuration, geometry/policy, renderer/assets/capture version, explicit approved-opening camera matrices, environment resolution and output tier. Old saved designs still revalidate, but their old capture-version images cannot satisfy the new cache match.
Only affected capture-version entries miss cache. Existing catalogue masters retain their original evidence because no obsolete shader was found. Exact locked-fixture contact textures are used for the control views. The broader catalogue's geometry-dependent contact adaptation remains a documented approval boundary; no universal V1 equivalence claim is made.
Before customer integration: review these controlled images, qualify the generalised contact-map output for all admitted families, then verify physical phones and the trusted matching-preview checkout route. No checkout, manufacturing validation, running catalogue rollout, production, approved V1, Unreal, Salad or Vast service was modified. No paid GPU was created. The cinematic sequence and all prior shared-workload measurements are preserved.
Camera, timing, configuration and protection evidence. Stopped for owner review.