The "Direct Push" model can't handle stale witnesses

[adlerjohn]

I think you can adapt the scheme I propose here (i.e. state lookup-less clients).

Specifically, modify the steps below to deal with general witnesses instead of UTXO-specific inclusion proofs:

@adlerjohn

Transactions in a block are validated with the following steps:

1. If the block that includes this transaction is past the timeout D, the transaction is invalid and so is the block.
2. For each input, if the input has an incorrect proof of inclusion, then the transaction is invalid. This can be done by only looking at block headers from at most D blocks ago.
3. For each input, check that the input has not been spent between the proof of inclusion block and the tip. This can be done by only looking at state deltas from at most D blocks ago.
4. The rest of the transaction (sum of outputs <= sum of inputs, etc.) can be validated in isolation, so long as UTXO IDs are constructed using all the necessary information, such as amount and owner .

TL;DR: Witnesses can be updated by applying only the last D blocks, which removes the need to interactively update them. Since D is bounded, we don’t have to worry about runaway DoS attacks.