Agent harness evaluation
TraceTrial-Bench
Task success is only part of the story. 95 mechanism-oriented tasks, four harnesses, nine models, with every verdict traced back to execution.
Summary
Evaluation, data, and training — connected to trace the path toward more capable agents.
BENCHMARKS
Agent harness evaluation
Task success is only part of the story. 95 mechanism-oriented tasks, four harnesses, nine models, with every verdict traced back to execution.
Environment sensitivity
The same task, byte for byte, in a different environment. Measures how much an agent's success depends on the environment it runs in.
Placeholder description text.
Placeholder description text.
Placeholder description text.
Placeholder description text.