I read the paper, and it seems to me that it boils down to choosing consistency over availability. That's a tradeoff, which results in the majority of the network getting consistency and high throughput, while the minority getting nothing in case they are cut off or experience significant delays. Very interesting.

That would mean that the minority chain (since they don't call it a fork), would have to acknowledge that they are indeed a minority, and cease all block production. Or wait, they used a DAG, didn't they? In that case, wouldn't each actor have to individually evaluate whether they are cut off or not? That seems like a problematic situation to me.

I'm disappointed that such papers never mention DPOS which apparently doesn't tend to fork. According to Larimer, DPOS doesn't fork because witnesses collaborate instead of competing. What that means at a deeper level I don't understand, but seems like something to dig into. However, I'm sure this Red Belly Blockchain isn't the final word on the subject.