Repository navigation
Interop question: where should a downstream relying-party claim live relative to AGT compliance receipts? #4564
|
Hi — I’m working on BoundaryAttest, a small open-source experiment around narrow signed claims when an artifact crosses a trust boundary: https://github.com/cullenmeyers/BoundaryAttest I read the current My understanding is that AGT’s proposed receipt already covers substantially more than a generic signed artifact:
So I would consider BoundaryAttest redundant for those native claims. In particular, I would not want to do this: AGT should remain authoritative there. The boundary I’m trying to understand is later. Suppose AGT produces a valid compliance receipt or evidence package and it leaves the operator’s environment: That seems different from AGT’s The external authorizer participates before execution and signs authorization of the original receipt/action. Organisation B in this example is instead a new downstream relying party making a claim after the original evidence already exists. BoundaryAttest is experimenting with that narrower layer:
It would not re-attest:
Those remain AGT-native claims. I recently got useful feedback from another evidence system where the distinction was essentially: I’m trying to find out whether that same distinction makes sense in AGT. Would you expect a post-handoff statement such as:
to become another AGT receipt/profile? Or is that intentionally outside the AGT execution/compliance receipt model, leaving room for an artifact-agnostic relying-party receipt? I’m especially interested in the non-fit answer. If the right design is simply to extend/use AGT’s native receipt model for every downstream claim, I would treat BoundaryAttest as redundant here rather than introduce a competing format. No implementation request — I’m trying to locate the semantic boundary between evidence about the governed action and a new relying party’s claim about already-existing evidence. |
Replies: 2 comments 10 replies
|
Short answer: from AGT's own documents, a post-handoff statement like "organisation B reviewed evidence digest D for case X" does not fit the compliance receipt model. It belongs in a separate attestation that uses the AGT evidence as its subject. Standard formats for that already exist and sit next to what AGT emits. (The receipt proposal is still marked Draft, so a maintainer should confirm intent.) Why it is a non-fit, from the proposal itself (
If you squeezed B's claim into a receipt profile, a verifier would see something shaped like evidence about the governed action. That is the conflation you want to avoid. Where it does fit AGT already speaks in-toto. Tutorial 33 (L176-199) shows receipts emitted as an in-toto {
"_type": "https://in-toto.io/Statement/v1",
"subject": [{
"name": "agt-evidence/case-X",
"digest": { "sha256": "<digest of the exact bytes B received>" }
}],
"predicateType": "https://example.org/relying-party-review/v1",
"predicate": {
"reviewer": { "id": "did:web:org-b.example" },
"case": "X",
"purpose": "assessment",
"status": "reviewed",
"timeReviewed": "2026-11-12T00:00:00Z",
"agtReceiptIds": ["<receiptId values covered>"],
"nonClaims": ["policy correctness", "execution", "compliance"]
}
}Two existing standards are close:
One practical gotcha: decide what "exact evidence bytes" means. AGT's So on your last point: native producer evidence, custody, and downstream review are three different claims, and AGT's documents cover only the first. The third has a natural home in in-toto plus SCITT. A BoundaryAttest-style claim adds most when it is defined as a predicate on those formats, not as a new envelope. |
|
We run the downstream layer you are describing in production, so here is the worked version from Zambo's handover system. The producer's evidence stays untouched: an AER-1 receipt, a digest over the exact committed bytes of the job record. Nothing downstream gets written into it. The relying party's claim lives in a separate object, the handover, whose subject is that evidence:
So "organisation B reviewed evidence digest D for case X" lands as B's own decision record pointing at D, which is the separation your three-layer sketch draws: producer evidence, custody, downstream claim. It matches the non-fit reading above from AGT's proposal text: the claim does not belong inside the compliance receipt, it belongs in a record that takes the receipt as its subject. One edge we had to settle the hard way: when two downstream parties both review the same evidence, "first acceptance recorded, everyone else's review still on the record but not the acceptance" kept custody auditable without pretending only one review happened. If BoundaryAttest goes the single-claim route, the multi-reviewer case is the one to pin down early. |
Short answer: from AGT's own documents, a post-handoff statement like "organisation B reviewed evidence digest D for case X" does not fit the compliance receipt model. It belongs in a separate attestation that uses the AGT evidence as its subject. Standard formats for that already exist and sit next to what AGT emits. (The receipt proposal is still marked Draft, so a maintainer should confirm intent.)
Why it is a non-fit, from the proposal itself (
docs/proposals/verifiable-compliance-receipts.md)actionwithinputHash,decision,authorizationHash/authorizationSignature(pre-execution),resultHash/resultSignature(post-execution), and