Attempt at evaluation for CoPE-B vs CoPE-A vs gpt-oss-safeguard

[julietshen]

I am NOT an AI engineer or AI researcher, but I tried to do a little evaluation of CoPE-B vs CoPE-A vs gpt-oss-safeguard

https://github.com/julietshen/cope-evaluation/blob/main/RESULTS.md

I'd love to hear how others' evaluations have gone. It was interesting to see the differences across policy length and detail.

I also need to double check if I accounted for the different formats gpt-oss-safeguard requires (harmony) or CoPE-B, but I think vLLM may have handled it automatically.