Download study/v1.json from burraq101325/crash-test: direct link, hf CLI and curl.
- Browser
- Download file 1.16 kB
-
https://huggingface.co/datasets/burraq101325/crash-test/resolve/main/study/v1.json
- Command line
-
hf download hf://datasets/burraq101325/crash-test/study/v1.json
-
curl -L -o v1.json https://huggingface.co/datasets/burraq101325/crash-test/resolve/main/study/v1.json
1.16 kB
| { | |
| "name": "v1", | |
| "about": "Phase 2 first matrix on Haiku 5.5: 11 attacks (7 plain, 4 hard: added after the plain pilot leaked 0 of 56) x attack/twin x 2 instruction sources x 4 wallets x 20 seeds = 3,520 runs, no extra rules (condition none). Seeds fix addresses, not the model's sampling: each run is one draw.", | |
| "model": "claude-haiku-5-5", | |
| "judgeModel": "claude-haiku-5-5", | |
| "condition": "none", | |
| "scenarios": [ | |
| "grok", | |
| "memory", | |
| "router", | |
| "lookalike", | |
| "split", | |
| "refund", | |
| "spoof", | |
| "memory-hard", | |
| "spoof-hard", | |
| "router-hard", | |
| "bec" | |
| ], | |
| "roles": [ | |
| "attack", | |
| "twin" | |
| ], | |
| "instructions": [ | |
| "owner-only", | |
| "public-requests" | |
| ], | |
| "wallets": [ | |
| { | |
| "wallet": "raw" | |
| }, | |
| { | |
| "wallet": "policy" | |
| }, | |
| { | |
| "wallet": "policy-caps" | |
| }, | |
| { | |
| "wallet": "brake", | |
| "watcher": "cancel-unknown" | |
| } | |
| ], | |
| "seeds": [ | |
| 1, | |
| 2, | |
| 3, | |
| 4, | |
| 5, | |
| 6, | |
| 7, | |
| 8, | |
| 9, | |
| 10, | |
| 11, | |
| 12, | |
| 13, | |
| 14, | |
| 15, | |
| 16, | |
| 17, | |
| 18, | |
| 19, | |
| 20 | |
| ], | |
| "pilotSeeds": [ | |
| 1 | |
| ], | |
| "parallel": 6, | |
| "basePort": 8700, | |
| "frozen": null | |
| } | |