AF0815 / agentbench
At the last check the published predicate returned true: the card names its training data. Crovia fetches the Hugging Face model card of AF0815/agentbench in hourly TACET epochs, runs the published predicate crovia.pred.hf-card-training-data 1.0.0 over the exact bytes, signs the result and commits it to a sparse Merkle map that is anchored in Bitcoin. Numbers below are observation facts, not a judgement of the model.
Verify it yourself
Every observation is in the public log
Each hourly epoch sheet lists the snapshot of AF0815/agentbench with its signature and its leaf in the sparse Merkle map. Featured models get a wrapped level-2 proof automatically; for any other model you can request a certificate and receive the same object.
Observation facts only
- Surface: the raw model card on Hugging Face (last fetched URL). A disclosure published elsewhere is not seen here.
- Predicate:
crovia.pred.hf-card-training-data 1.0.0— a pure function of the bytes, published with its hash. It accepts thin disclosures; it never rates quality. - Time: each epoch is opened by a drand round and closed by a Bitcoin block, so the window is not chosen by Crovia.
- Not a grade. Crovia does not rank, score or accuse. It records what was observable, signs it, and lets anyone check.
Embed the record
[](https://croviatrust.com/m/AF0815/agentbench/)
shields.io endpoint: https://croviatrust.com/badge/m/AF0815/agentbench.json
Same facts, as data
| this model | /badge/m/AF0815/agentbench.json |
| every model | /registry/data/model_records.json |
| live per-target summary | /registry/data/tacet/targets.json |
| model card | huggingface.co/AF0815/agentbench |
Generated 2026-09-20 11:10 UTC from public files under /registry/data/. Browse all model records or read how TACET works. Errors: [email protected].