This reminds me of build-time optimizations increasing compile times for faster performance at runtime. A tradeoff I could make without a sweat in prod, but not so much during development, at a certain scale. It feels like a dichotomy almost unavoidable in life.
I think Anthropic made the guardrails overly restrictive on purpose, considering it was unprecedented how powerful the model was. I think it was also intended to help their PR, as they've always marketed themselves as the "ethical" AI lab, so they wish to go overboard so people think they're just working hard on their filters (which I don't doubt they are), especially when we look at the money behind it.
I think scrabble might be a nice one. It is presented in a format that (to me, from just looking at it) could be nicely implemented in it, and would provide for some emotion/traces that would be represented well.