In theory, a miner could try to manipulate the hash of the future block to change the outcome. It is much more difficult with Bitcoin, where valid blocks are more rare and produce a high income, but with Ethereum blocks are so frequent that it might be possible to collude with the mining pools is done on a large scale.

Additionally, how does this deal with Proof-of-Stake algorithms? Ethereum is heading there, and as far as I am aware the blocks would no longer publish hashes.

A sybill attack would greatly increase the probability. And has cheatingly other gains.

No need for a future block. Recipient commits to the hash of a value. Payer signs a value. Hash the preimage (revealed in claim) and the signed value together. If less than a threshold, pass.

As long as neither party has any way to manipulate the hash. Should probably add a 0th step where the payer commits to use a specific signature.

The signature doesn't matter. Allowing the payer to change it only increases (potentially) the odds of them having to pay, which they won't do. In some constructions--in others it has no effect.

Ok it looks better after looking at it again. Thanks.