Neighbor project: https://github.com/mohamedzhioua/agent-done-or-not
@mohamedzhioua — I am testing the boundary between runtime proof-of-done receipts and reviewer-side before/after evidence.
Your current review-pr semantics are unusually careful: re-run the PR's testable claims on the current candidate, call them RE-EXECUTED / ASSERTED / UNPARSED, and explicitly avoid upgrading a green rerun into 'VERIFIED'. That is very close to the boundary CounterProof is trying to preserve.
The seam I see is:
agent-done-or-not
Did the claimed command actually run and pass on this candidate?
CounterProof
Does the same submitted evidence distinguish BASE from HEAD?
For example, review-pr can establish:
"tests pass" -> npm test -> exit 0 on HEAD
but a fix claim may still need:
same exact regression command
HEAD -> PASS
BASE -> FAIL
before a reviewer can call it a witnessed fix rather than simply a current green state.
So the narrow question is:
Would you want a machine-readable review-pr receipt to be reusable as upstream evidence for a BASE→HEAD verifier, or do you deliberately want those receipts to stay current-candidate-only?
If reuse is intended, I would expect the minimum handoff to be command/label + candidate identity + exit result + output digest, while the downstream verifier remains responsible for BASE replay and scope semantics.
No adapter proposed yet. I want to establish the ownership boundary before writing glue.
CounterProof: https://github.com/hippoley/CounterProof
Neighbor project: https://github.com/mohamedzhioua/agent-done-or-not
@mohamedzhioua — I am testing the boundary between runtime proof-of-done receipts and reviewer-side before/after evidence.
Your current
review-prsemantics are unusually careful: re-run the PR's testable claims on the current candidate, call them RE-EXECUTED / ASSERTED / UNPARSED, and explicitly avoid upgrading a green rerun into 'VERIFIED'. That is very close to the boundary CounterProof is trying to preserve.The seam I see is:
For example,
review-prcan establish:but a fix claim may still need:
before a reviewer can call it a witnessed fix rather than simply a current green state.
So the narrow question is:
Would you want a machine-readable
review-prreceipt to be reusable as upstream evidence for a BASE→HEAD verifier, or do you deliberately want those receipts to stay current-candidate-only?If reuse is intended, I would expect the minimum handoff to be command/label + candidate identity + exit result + output digest, while the downstream verifier remains responsible for BASE replay and scope semantics.
No adapter proposed yet. I want to establish the ownership boundary before writing glue.
CounterProof: https://github.com/hippoley/CounterProof