Did you read the paper? The paper is arguing the exact same point I was arguing. To quote the paper:

> Finally, assuming that the parties are rational agents and the smart contract language is Turing complete, we argue that it is impossible to implement the basic sales escrow as a smart contract without trusted third-parties or vulnerability to extortion. In other words, any escrow smart contract has one of the following three demerits:

> – Assuming irrational agents who are willing to punish the other side, even if it is not in their own interest; or

> – Relying on a third-party; or

> – Enabling at least one of the two parties to extort the other.

> In summary, we illustrate that the smart contract and Dapp community is wrong in assuming that the current implementations of two-party escrows have a well-designed mechanism that incentivizes rational actors to be truthful. More shockingly, we show that the smart contracts on programmable blockchains have inherent limitations that make it impossible to implement such a contract. In a sense, this can be considered the first incontractability result on programmable blockchains.

----------

This is exactly what I was arguing.

I never claimed that two party escrow is ideal. I was explicitly saying that two party escrow is an intractable problem and that you must formalise your trust assumptions instead and either accept some level of trusted third parties OR without third parties accept some level of risk of exploitation by one party or the other. Even with third parties there is still risk for exploitation but depending how it is implemented that risk is lesser.

Again this is a matter of formalising trust assumptions and explicitly outlining who you are trusting, what you are trusting them to do, and how much you trust them to do it. And in doing so up front both parties can evaluate their risk tolerance based on the agreed upon contract before progressing.