That didn’t work out so well.

Exactly.

The general approach is to use interpretable models where you can understand how the model works and what features it uses to discriminate, but that doesn’t work for all ML approaches (and even when it does our understanding is incomplete.)

Maybe not the hardest, but still challenging. Unknown biases in training data are a challenge in any experimental design. Opaque ML frequently makes them more challenging to discover.

I’ve had great luck with Upton teas for the past decade or more in the US. Great selection, teas at multiple price points from broken leaf to first flush single estate.