Datasetpublic

jaweiss2305/support-agent-prod-2026-08-25

support-agent-prod / 2026-08-25 Production support-agent traces, graded. Analyzer regime TRAIN. Agent rollouts exported from ZeroProof for support-agent. Each row is one rollout: the prompt, the tool trace, the final answer, and the reward the judge gave it. Split ZeroProof dataset Rows Prompts Pass rate Support Mixed prompts Pushed holdout ds_550a2b3ba46aac2b 32 8 72% 0.78 8 2026-09-14 Load from datasets import load_dataset holdout =… See the full description on the dataset page: https://huggingface.co/datasets/jaweiss2305/support-agent-prod-2026-08-25.

sourceHugging Faceupdated 13d agoView on Hugging Face

0likes82downloads

Dataset Card

support-agent-prod / 2026-08-25

Production support-agent traces, graded. Analyzer regime TRAIN.

Agent rollouts exported from ZeroProof for support-agent. Each row is one rollout: the prompt, the tool trace, the final answer, and the reward the judge gave it.

Split	ZeroProof dataset	Rows	Prompts	Pass rate	Support	Mixed prompts	Pushed
`holdout`	ds_550a2b3ba46aac2b	32	8	72%	0.78	8	2026-09-14

Load

python

from datasets import load_dataset
holdout = load_dataset("jaweiss2305/support-agent-prod-2026-08-25", split="holdout")

Pin a version with revision="zp-<dataset id>": every push is tagged with the ZeroProof dataset it came from, and zeroproof.json in this repo maps each split to its dataset, numbers and history.