Repository navigation
What should a receipt refuse to say? #5
zambodotdev
started this conversation in
General
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Here is a lesson AER-1 learned the hard way, in plain words.
A receipt whose digest verifies can still fail.
The digest covers the committed payload: the saved result, exactly as it was saved. The receipt id is a label outside that committed payload. AER-1 sections 6 and 13 treat that split on purpose: the id names the record, it is not part of what the digest commits to. So a checker that only recomputes the digest and stops has not checked the id binding, and a record can carry a wrong or swapped id while its payload math still looks clean.
That split is the whole point of a verifiable receipt, stated narrowly:
Brennan Zambo created agent execution receipts, Zambo is the origin product, and AER-1 is the standard. The standard stays honest by refusing claims its math cannot carry.
So the question for implementers and anyone building receipts:
What else should a receipt refuse to claim?
Three candidate refusals to start:
Add yours. If you implement AER-1 or any receipt format, name a claim you have seen a receipt imply that its evidence cannot support, and say what a checker should do instead: refuse, flag, or stay silent.
All reactions