Hi Pavan, It's great to hear that you are working on an implementation. Receipts are COSE_Sign1 with a detached payload that captures the Verifiable Data Structure, not just the statement (e.g. the root of the Merkle Tree for RFC9162_SHA256: https://www.ietf.org/archive/id/draft-ietf-cose-merkle-tree-proofs-18.html#section-5.2.1), and so the contents of the proof are authenticated by that (despite being in the unprotected header), as well as the statement, because following the verification algorithm (https://www.rfc-editor.org/rfc/rfc9162.html#name-verifying-an-inclusion-proo) amounts to re-constructing the COSE_Sign1 payload before verifying it. There is no race condition with getting the root, since this is not something that needs to be fetched separately.
I hope this helps, but please do feel free to ask if anything is still unclear, Thank you, Amaury ________________________________ From: Pavan Kumar <[email protected]> Sent: 13 April 2026 06:28 To: [email protected] <[email protected]> Cc: [email protected] <[email protected]>; Thomas Howe <[email protected]>; Harinder Bhinder <[email protected]>; Chris Celiberti <[email protected]>; Matt Watson <[email protected]>; Jacob Howe <[email protected]>; Sunaina Khan <[email protected]> Subject: [EXTERNAL] [SCITT] Two concerns with receipt design: non-repudiation and offline verification You don't often get email from [email protected]. Learn why this is important<https://aka.ms/LearnAboutSenderIdentification> Hi SCITT working group, We are implementing a SCRAPI-compatible transparency service (scittles) and integrating it with a conversation recording system (vCon). While testing end-to-end we found two related concerns with the current receipt design. We wanted to raise them with the group to understand whether we are missing something or whether these are known open issues. Background Per the architecture spec, a receipt is a COSE_Sign1 where: * Protected header: alg, kid, VDS type (-111), CWT claims (iss, sub) * Unprotected header: Merkle inclusion proof [tree_size, leaf_index, [siblings]] * Signature covers: protected header + statement hash (detached payload) ________________________________ Concern 1: Non-repudiation failure The inclusion proof lives in the unprotected header and is therefore not covered by the transparency service's signature. This means the transparency service can deny having issued the proof — it can claim "I signed this statement hash, but I never committed it to position N in a tree of size T with those siblings." The recipient cannot prove otherwise. This is a fundamental non-repudiation problem. The entire purpose of a transparency log is to create a binding commitment from the service to the log state. But the proof of that commitment — the inclusion proof — is the one part the service hasn't actually signed. The spec justifies placing the inclusion proof in the unprotected header so that fresh proofs can be attached without re-signing. We understand the intent, but this design trades non-repudiation for proof freshness — and non-repudiation is the more fundamental property. ________________________________ Concern 2: No atomic signed root anchor Even setting non-repudiation aside, offline verification of the inclusion proof requires comparing the computed root hash against a trusted signed root. The receipt contains no signed root hash, so a verifier must call the service for a Signed Tree Head (STH). A separate GET /tree-head endpoint does not solve this because of an unavoidable race condition: between POST /entries (receipt issued at tree_size=N) and GET /tree-head, further entries may be registered, returning an STH for tree_size=N+K. The inclusion proof for tree_size=N cannot be verified against a root for tree_size=N+K. There is no way to call these two endpoints atomically. ________________________________ Proposed fix Both concerns are resolved if the inclusion proof and root hash are included in the protected header of the receipt, signed atomically by the transparency service at registration time. This means: * The service signs: {alg, kid, VDS, claims, tree_size, root_hash, timestamp, leaf_index, siblings} * A verifier walks the siblings → computes root → compares against signed root_hash → fully offline, non-repudiable * "Fresh" proofs require issuing a new signed receipt — which is the correct behaviour: a new receipt is a new commitment from the service We recognise this is a meaningful change to the spec. We raise it in case it reflects a broader concern others have encountered, or in case there is an existing mechanism we have missed. Thank you for any guidance. -- Pavan Kumar Lead Technical Consultant Strolid 📧 [email protected]<mailto:[email protected]> 🌐 www.strolid.com<https://www.strolid.com/>
_______________________________________________ COSE mailing list -- [email protected] To unsubscribe send an email to [email protected]
