遇见数据集

AgentPROV corpus: ControlArena persistent-state evaluation logs, metadata only (MIT)

收藏
Zenodo2026-10-30 更新2026-11-05 收录
官方服务:

资源简介:

An execution-provenance corpus built by AgentPROV v0.1.0 from the ControlArena persistent-state Inspect evaluation archive (MIT). This is a metadata-only release: the source profile withholds all 5,742 executions at the seal (D-344), so the corpus releases no execution record (no AER, SEM or PROV); it keeps its manifest, register, datasheet, withholding record and the raw objects the seal does not withhold. This deposit draft holds 1 AgentPROV corpus version(s), prepared offline by agentprov package --deposit for the release date 2026-10-30. Nothing has been uploaded or published. Corpus controlarena-persistent 0.1.0: Control-arena persistent-state evaluation archive: Inspect .eval logs of AI-control runs (HF idacy, 2026-07-30). 0 execution records, of which 0 are outcome-only records. Licence class: MIT. Content digest sha256:15b3e055eb8823324182f0331543d9d904e36c65786e6e8cf0f27a60f92c52c8; root digest sha256:1539bc388f58e6ab2f74455485928928912004e477f8c4645ef803678f9f310c. Counts are computed from each corpus manifest at sealing. Each DATASHEET states what the records do and do not establish; a statement about the data is scoped as it is there. To verify: compare each archive's sha256 or md5 with DEPOSIT-MANIFEST.json, run agentprov verify on the archives (archive mode, no extraction; H.5.5), and compare each root digest with the one published outside the deposit (the corpus tag annotation and the paper).

提供机构:
Zenodo
创建时间:
2026-10-30
二维码
社区交流群
二维码
科研交流群
商业服务