Surgery BenchHecaton multi-arm surgical robotics benchmark
Gallery: what the environments look like
These are the only two media items accepted so far; both are simulated surgical runs recorded straight from the renderer.
Each item plays as video (click to load; nothing downloads until you press play) with still frames, and carries the full honest label written at review time — visible defects included.
Rejected media is never shown on this page. Realism scores are cross-model judged (a vision model verdict plus the lead's frame read); the defect lines in each label are shown in full, never hidden.
i4h chole_tet_med - Slang renderer base reproduction
(lossless original not shipped on the preview site: 257 MiB source exceeds the 25 MiB per-file deploy cap — see the sha256 above; the file is on record in the lane data tree)
SOTA base reproduction -- i4h official Slang renderer (chole_tet_med), scripted driver.
Physics 48.8 Hz with render+capture, 0 dropped frames.
NOT shown: bleeding, smoke. Clip + thermal fire in the engine but are not visible in the frames.
Cross-model review (glm-5.3-flash): claim PARTIAL, realism 5/10, show only with this label.
Defects: gallbladder too small vs tools, 6-7 crossing shafts, GB hidden at 16 s.
i4h artery-cut patch r2 - visible bleeding at cystic pedicle
(lossless original not shipped on the preview site: 287 MiB source exceeds the 25 MiB per-file deploy cap — see the sha256 above; the file is on record in the lane data tree)
unpatched vs r2 side-by-side (t=8.0 s)unpatched vs r2 side-by-side (t=14.0 s)
i4h base + our artery-cut patch (r2), scripted driver, i4h Slang renderer.
Engine: 1 cystic-artery node cut, blood from step 255, 205 tets removed (baseline 128), gallbladder intact.
Blood is VISIBLE at the cystic pedicle from ~8 s but sprite-like: flat decals and disc splatter, no stream or wetness; it barely grows (particle cap 4096).
White/grey ring discs on the floor = irrigation particle sprites (artefact).
Cross-model review (glm-5.3-flash): claim YES, realism 4/10, show only with this label. Lead look agrees.
Next: best blood/fluid render backend (D-S15b).