[SCITT] Re: New I-D: Evidence Requirements for Agent Control Delivery and Outcome Reconciliation (draft-abak-agent-control-delivery-evidence-00)

[Walter Hawkins <[e-mail]>]

Emek,

Congratulations on publishing draft-dogru-cedulon-decision-profile-00.

Naming `effect-against-refusal` as a first-class reconciliation failure
mode is an important contribution. Existing payment and spend profiles
almost universally assume non-execution under denial; formalizing rogue
execution against an authoritative refusal fills a genuine gap in the audit
model.

Regarding your IANA registration question: your reasoning is sound and
consistent with SCITT/COSE architectural practice.

If the Effect Extract is delivered as an application-level detached JSON
payload rather than an encapsulated COSE structure whose MIME type is
parsed and validated inside an authenticated protected header parameter
(e.g., `content type` / label 3), registering a standalone media type adds
registry noise without cryptographic or protocol enforcement. Limiting the
formal media type registration strictly to the COSE_Sign1 Decision Record
keeps the IANA parameter footprint clean and avoids encouraging fragile
MIME-sniffing downstream.

We will run the conformance suite against commit da7bf9b and review the
interaction with receipt-invariant checks.

Best regards,
Walter

On Fri, Sep 4, 2026 at 9:14 AM Emek Can Dogru <[e-mail]> wrote:

> Ali, Iman, all,
>
> Following this thread, I wrote the decision side out as a profile on the
> Cedulon reconciler and posted it today:
>
>   draft-dogru-cedulon-decision-profile-00
>   https://datatracker.ietf.org/doc/draft-dogru-cedulon-decision-profile/
>
> The population is the one this thread has been circling. A Decision
> Record is a COSE_Sign1 signed by the party that decided whether an agent
> may act (allow, deny or defer, with the request hash, the policy hash
> and, for an allow, the reference and the content hash of the effect it
> allowed). An Effect Extract is an authenticated list of what actually
> occurred on a channel. An allow must be matched by exactly one effect
> carrying the content hash it named; a refusal must be matched by none.
> The finding the profile exists for has its own name,
> effect-against-refusal: an effect under a reference a refusal names.
> The spend vocabulary had no word for that case.
>
> What is measured and what is not, in the document's own terms: the
> profile runs on the same reconciler that runs spend, and the spend
> behaviour is held byte for byte by a fifteen-case golden file; eighteen
> conformance cases and four offline fixtures for a direct-message reply
> log; the live log's field names were not read, and no independent
> implementation has run it. Before posting, the draft was read by an
> outside model barred from editing, and the code behind it by two; what
> those readings found is in the companion's review log, the code defects
> each with the test that was red before the fix.
>
> Pinned commit: dogrucanemek-alt/cedulon@da7bf9b. `npm ci && npm run
> test:pre-release`; the cases are tests/decision-profile.test.ts, the
> fixtures interop/mizan-ig/. Same terms as before: run it, try to break
> it, say what you found.
>
> One registration question for anyone who has been through it. The
> profile registers one media type, for the Decision Record, and none for
> the Effect Extract, which is a JSON body with a detached signature; the
> reasoning is that the core registers names only for objects whose
> content type is checked inside a protected header. If that reasoning is
> wrong I would rather hear it before the six-type request is answered.
>
> Emek
>
> On Fri, Sep 4, 2026 at 6:05 AM Ali Toygar Abak <[e-mail]> wrote:
> > Hi Emek,
> >
> > Thank you. That closes the focused review on my side as well.
> >
> > I also have the pinned reproduction file and exact invocation now. I’ll
> run the 0.8.0 and 0.12.0 cases independently from clean directories and
> record the result separately, without changing or repinning the historical
> 1 September probe.
> >
> > I’ll keep the distinction explicit between your reported measurements
> and my independent reproduction until that run is complete.
> >
> >
> >
> > Best,
> >
> > Ali Toygar Abak
> >
> >
> >
> > Phionyx
> >
> > https://phionyx.ai
>
> --
> SCITT mailing list -- [e-mail]
> To unsubscribe send an email to [e-mail]
>