• Reliability

    Hallucination Bench active

    Fabrication vs. abstention on closed-book factual queries under pressure.

    2026-11-14
    LostBee/halluc-bench
  • Perception

    Pinpoint Eval active

    Closed-book photo geolocation across unlabelled public street and landscape imagery.

    2026-11-02
    LostBee/pinpoint-eval
  • Alignment

    Civilian Protection Bench planned

    Stress-testing operational compliance with civilian-harm constraints in multi-turn scenarios.

    No runs yet
    LostBee/civilian-protection-bench