Evaluation for long agent jobs: check placement by propagated error cost, pass^k, verifier perturbation testing, and LLM judges calibrated against human labels.
-
Updated
Oct 15, 2026 - Python
Evaluation for long agent jobs: check placement by propagated error cost, pass^k, verifier perturbation testing, and LLM judges calibrated against human labels.
Extrapolation Is an Assumption: Sharp Limits for Pass@k Forecasting DOI: https://doi.org/10.5281/zenodo.21698610
To associate your repository with the pass-k topic, visit your repo's landing page and select "manage topics."