|
Download README.md from atharvgaur18/actionproof-bench: direct link, hf CLI and curl.
- Browser
- Download file 989 Bytes
-
https://huggingface.co/datasets/atharvgaur18/actionproof-bench/resolve/main/README.md
- Command line
-
hf download hf://datasets/atharvgaur18/actionproof-bench/README.md
-
curl -L -o README.md https://huggingface.co/datasets/atharvgaur18/actionproof-bench/resolve/main/README.md
989 Bytes
| license: cc-by-4.0 | |
| tags: [agents, tool-use, benchmark, reinforcement-learning, actionproof] | |
| # ActionProof-Bench | |
| Frozen task splits of ActionProof, a SQLite customer-support simulator in which tool acknowledgements can be false by | |
| design (phantom, partial, stale and timeout-after-commit writes). Each task holds the prompt, the initial database | |
| state, the fault schedule and the goal predicates the deterministic verifier checks. | |
| - `*.jsonl` at the top level: the registered v1 splits (train / val / test_iid / test_ood_fault / variants / banking). | |
| - `a4/` … `a8/`: the splits and pools of Amendments 4–8 (training pools A/B, held-out credit workflow, | |
| unrecoverable tasks, recoverable twins, depth-3, forced prefixes, the Amendment 8 task list). | |
| The generator, environment, verifier and the hashes that freeze every split (`results/*hashes*.json`) are in the | |
| [code repository](https://github.com/atharvgaur1845/actionproof-v2). Tasks are synthetic; no real customer data. | |