This belongs to the model review category of posts. It starts with “While I’m known for my model comparisons/tests focusing on chat and roleplay…” which is similar to how the Unsloth developers made a reputation for themselves in the subreddit through their work.
OP regularly completes data protection training for work, and was inspired by that process to use as a test case for AI models in a “serious work” setting.
The testing methodology is very narrow, and very specific. Even the setup is—it uses a character card that roleplays a character they made, and it is on the specific training data protection training information and exam questions OP came across.