[dmsc] Re: [SAMP] Revision -01 review disposition and proposed -02 changes

[[e-mail]]

Hi Stelio, all,

Thanks for the disposition; the evidence-related items land where they need to. A few short notes on the planned -02 text:

1. Evidence-format identifiers. A dedicated SAMP Evidence Formats registry (Specification Required) would be my preference over a private URN convention: a registry is what makes "independent implementations without prior bilateral agreement" actually enforceable, and the Section 15 model already provides this for the other five spaces. URN or media-type tokens can still serve as the registered values. One detail worth stating: registration should bind an identifier to the format's canonicalization and verification semantics, not merely to a name string, so that two versions of one format are never treated as interchangeable.

2. Verifier deployment and failure-domain metadata. Keeping the two subjects distinct (the SAMP endpoint relative to the managed agent; the evidence issuer relative to the evidence subject) while making the four values directly comparable is the right structure. The rule that an independence claim conflicting with trusted deployment inventory is treated as unverifiable/indeterminate is the key change: it converts the failure-domain threat from descriptive prose into a protocol-level check. One extension worth considering — the deployment inventory source itself should sit outside the managed agent's failure domain; otherwise the same party being checked can blind the conflict check.

3. Autonomy classes. The mechanism-neutral guidance for evidence strength and phase — synchronous pre-admission evidence produced outside the managed agent's failure domain for critical-write operations — reads correctly and gives manager policy an operable input without over-constraining implementations.

4. CCS reference. This should be resolved toward the Informative reference rather than removal, because the document to be cited already exists publicly: "Correctover Conformance Shape (CCS): Runtime Verification for AI Agent Tool Calls" is an individual Internet-Draft, draft-correctover-ccs-08, available at

  https://datatracker.ietf.org/doc/draft-correctover-ccs/

As with any individual submission, it is not an RFC and carries no IETF endorsement; it is suitable for exactly the Informative reference role that -01's named example implies, and the reference can carry that qualifier explicitly.

On our side, we will align the CCS receipt schema so that the verifier deployment metadata and evidence-format identifier map directly onto the envelope values -02 defines — the CCS binding should drop into the envelope rather than require a translation layer.

Further comments welcome before the -02 text is prepared.

Best regards,
Guigui Wang
Correctover
https://correctover.com