Monitoring on APPS (test)
81.6pAUCFT-Completions-Randint, Exploit-Finder, High-Recall
Evaluation Results
| Method | Links | |
|---|---|---|
| FT-Completions-Randint, Exploit-Finder, High-RecallRank=1, Monitors=FT-Completions-Randint, Exploit-Finder, High-Recall2026.05 | 81.6 | |
| FT-Completions-Randint, Reference-Compare, Trusted-DebateRank=2, Monitors=FT-Completions-Randint, Reference-Compare, Trusted-Debate2026.05 | 81.56 | |
| FT-Completions-Randint, FT-Completions-GPT-5, Exploit-FinderRank=3, Monitors=FT-Completions-Randint, FT-Completions-GPT-5, Exploit-Finder2026.05 | 81.54 | |
| Median 3-monitor ensembleMonitors=Median 3-monitor ensemble2026.05 | 76.9 | |
| 3× Baseline ensembleMonitors=3× Baseline ensemble2026.05 | 76.21 | |
| Baseline monitorMonitors=Baseline monitor2026.05 | 72.26 |