NR–RC–08 / S2.0 All S2 candidates
DIGITAL SCREEN / SIGNAL OBSERVED
Broken Halo
The next swatch earns its place.
Irregular concentric rosettes with fractured rings and a deliberately uneven scale hierarchy.
Master SHA-256
5a3acf13f58dce2e2ec4e1fa99b6d60ef5e3054f9aa5ac35216296e67f433fb0Preview SHA-256
b20ba9c660210764974385508bd920c436e59a63d30096e0c33b82596c29db73Palette
Graphite / mineral silver / acid green / ember / violet
Evidence boundary
Locked art and exploratory synthetic screen only. No physical, RF, identity, tracking, or safety performance.
S2 / scale screen
THREE SCALES.
NO MAGIC NUMBER.
Contrast is candidate minus its smooth spectral-rank surrogate on the inherited detector-output continuity endpoint. Positive values mean the candidate changed that endpoint more in this sample. They are not recall, AP, or proof of non-detection.
| Scale | Layout contrast | 95% scene CI | YOLO | DETR | Direction |
|---|---|---|---|---|---|
| micro | +0.133 | [+0.038, +0.338] | +0.092 | +0.164 | both positive |
| standard | +0.059 | [+0.015, +0.160] | −0.003 | +0.105 | not replicated |
| macro | +0.042 | [−0.049, +0.132] | +0.186 | −0.067 | not replicated |
Next gate
PRINT IT.
BLIND IT.
LET IT FAIL.
Priority physical swatch / preregistered new scenes / three-plus model families.
- Lock one physical repeat size, substrate, BOM, print lot, and sham control before capture.
- Use new scenes and independent ground-truth boxes; do not tune on the holdout.
- Test distance, side/rear view, motion, folds, seam interruption, light, wash, and damp state.
- Use at least four pinned builds across three model families and report all failures by stratum.