After looking at some fairly exotic possible transaction types, I ran into the current policy limit requiring transactions to be 85 non-witness serialized bytes. This was introduced as a covert fix to policy fix for CVE-2017-12842. Later the real motivation was revealed, but the "reasonable" constant chosen was not.

I'd like to propose relaxing this to effectively the value BlueMatt proposed in the Great Consensus Cleanup: 65 non-witness bytes. This would allow a single input, single output transaction with 4 bytes of OP_RETURN padding, rather than padding out 21 bytes to get to p2wpkh size.

The alternative would be to also allow anything below 64 non-witness bytes, but this seems fraught with footguns for a few bytes gain.

The PR is here with more relevant background and alternatives included in the thread:

Please let us know if there's a fundamental issue with this approach, or any other feedback.

Best,

Greg