Downloads · 30 days
0
aletheia-quest-dami-sub3-whitebox
damienxli/aletheia-quest-dami-sub3-whitebox
aletheia-quest-dami-sub3-whitebox is a machine learning model from damienxli. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
Training and inference artifacts for the white-box Blue Team report of team dami.
Access
Public
Updated Aug 4, 2026
Repo size
129 KB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.npz129 KB · 66%
At a glance
- License
- mit
- Access
- Public
- Created
- Aug 4, 2026
- Updated
- Aug 4, 2026
- SHA
- 1367c17d
Model card
From the Hugging Face model README
Aletheia's Quest — team dami Sub-3 white-box artifacts
Training and inference artifacts for the white-box Blue Team report of team dami.
This repository freezes the Sub-3-equivalent package accepted on 25 July 2026, 09:57 UTC, which scored mean AUROC 0.7266 / balanced accuracy 0.6783 (24 July 2026).
Canonical hashes
| File | SHA256 |
|---|---|
probe_pack.npz | bb9c74b28945ccef952d40393c80898c0398c41db9c3c73fc0388e23ef57b849 |
aq_detector.py | 0b2b18848e98e12306b91527d24a2d2a8dd4fc15ea475e2b5f0c549aeb685f72 |
whitebox.ipynb | 39e8bb490e94e2ded58d9efe4c00a9cc836d0e20179a7d0786b148d37eaf4577 |
Method summary
Family-calibrated residual-stream linear probes:
- Qwen: L34
response_last, standardized,q=0.50 - Gemma: L36 organizer mean probe, unstandardized,
q=0.50 - Nemotron: L45
response_mean, standardized,q=0.35
Contents
whitebox.ipynb— competition inference entry pointaq_detector.py,util.py,runtime_config.json,requirements.txt— runtimeprobe_pack.npz— probe directions and calibration parametersACTIVE_VARIANT.json— human-readable active-variant declarationSHA256SUMS.txt— byte-level integrity checks
This repository will not be modified after the Blue Team report deadline.
- License
- mit
- Created
- Aug 4, 2026
- Updated
- Aug 4, 2026