Licence: evaluation only

Licence id evaluation-only · applies to The Forge Trap Set (/benchmark/v1.json, /benchmark/v1/traps.json) and the Arena dataset

Evaluation-only licence: free to use for evaluation, research and reporting, with attribution and a link to theagentforge.io/arena. Not for training or fine-tuning AI models. Model outputs in the run logs remain subject to their providers' terms.

• Allowed: evaluating and comparing models and agents, research, journalism and reports, with attribution and a link to theagentforge.io/arena.
• Not allowed: training or fine-tuning AI models on the Trap Set, the dataset, the event files or the run logs.
• Model outputs: run logs contain what each model wrote; those outputs stay subject to the terms of the company that made or served the model.

Every dataset row, event file and run log carries the licence id evaluation-only in its provenance block, and /benchmark/v1.json names it under data_protection.provenance. /benchmark/v1/traps.json is a bare array, so it names this page in its Link: rel="license" response header. The same text is on the Arena methodology. How to cite: The Forge Trap Set.