YUREI RESEARCHdefault-layout geometry on a real holdout
Can the deterministic default-layout proposal serve as a UI-region detector, or is it only suitable for conservative crop routing?
Labnote 001: default-layout geometry on a real holdout
Question
Can the deterministic default-layout proposal serve as a UI-region detector, or is it only suitable for conservative crop routing?
Method
One operator-approved, structurally masked 1920x1080 Guild Wars 2 frame was admitted as a private, non-release test holdout. The operator separately approved tight boxes for three visible regions: objectives, skill bar, and minimap. Chat and party/squad panels remained masked and unlabeled.
The checked-in default-layout proposals were compared with these verified boxes using kind-matched intersection-over-union. No private pixels, recognized text, character names, or paths are recorded here.
Result
| Metric | Result |
|---|---|
| Verified regions | 3 |
| Proposals | 5 |
| Mean IoU | 0.418298 |
| Recall at IoU 0.50 | 0.000000 |
The default geometry matched zero of three verified regions at the preexisting 0.50 threshold.
Decision
The deterministic geometry remains a low-confidence crop heuristic. It is rejected as detector ground truth and must not issue semantic observations or privacy reveal decisions.
One frame is not enough to select or train a learned detector. The next gate is a small, varied, separately consented real holdout collected under the same structural masking policy. Model dependencies remain deferred until that corpus demonstrates sufficient layout variation and reviewed labels.