World Observation Metrics · WOMPar standalone demo
One object
Split + eye corrector
0 matches the reference. Negative beats it; positive trails it. Range: −100 to +100. The candidate score uses both A/B orders from one model reviewing this source view.
Both A/B comparisons are required for a candidate score. Offline and AI3D generation modes do not review the candidate.
Global WOMPar: Not ranked. Single-source-view demonstration; no complete-course evidence or independent judging panel.
Mode: offline. Review requests attempted: 0; completed comparisons: 0. Offline: no network requests.
{
"provenance": "Saved measurements of the public object-split case; not recomputed by this demo.",
"figureColorError255": {
"reference": 26.17,
"candidate": 18.47
},
"rightEyeShiftPx": {
"reference": 38.629524336486334,
"candidate": 0.038470196932005755
},
"leftEyeShiftPx": {
"reference": 25.631011102658025,
"candidate": 0.041600138374738374
},
"sideGapPixels": {
"angle": 40,
"one": 926,
"split": 2055,
"of": 327680
}
}
No live comparison recorded.
The public saved case demonstrates reconstruction tradeoffs. Its whole-person reference and split candidate are from the scripted object-split experiment, not the frozen global pilot course. Side gaps can increase while source-view detail improves.
WOMPar 0.1 defines Par 0 on its named full course; negative beats that reference. A valid complete-course score requires frozen walkthrough evidence and the full independent judging panel.