Dataset Viewer
Auto-converted to Parquet Duplicate
text
stringclasses
330 values
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
"g_fit": 1.0,
"2d_fit": 1.0,
"id_ig": 1.0,
"reward": 1.0
}
{
"d_fit": 1.0,
End of preview. Expand in Data Studio

raman-lab runs

Raw Harbor trial folders from the multi-model run of 2026-11-18, behind the Raman-fitting example in Public Benchmarks won't save you! How to build better agentic evals.

Six open-weight models ran with the Terminus-2 agent harness on two task sets: Terminal-Bench 2's raman-fitting (3 rollouts per model) and raman-lab, eight adapted synthetic spectra (3 rollouts per task). One folder per rollout holds:

  • agent/trajectory.json: the trajectory in ATIF format
  • result.json: Harbor's trial result
  • verifier/reward.json and verifier/test-stdout.txt: the verifier's score and output
  • config.json, lock.json and trial.log: run configuration and logs

Analysis code, figures and per-rollout tables are in burtenshaw/raman-lab-evals.

Downloads last month
42