SCITT S. Mih Internet-Draft Action State Group, Inc. Intended status: Standards Track A. Sokolov Expires: 25 February 2027 Tyche Institute 24 August 2026 Canonical Payload Binding: A Signed Statement Construction Profile draft-mih-sokolov-scitt-payload-binding-02 Abstract Independently written systems that anchor records to a SCITT Transparency Service repeatedly re-derive the same construction: a canonical form of structured content, a content-addressed identifier derived from that form, a receipt placed in the unprotected header of the Signed Statement, and a typed reference mechanism that lets one record cite another by digest across profile boundaries. This document defines that construction as a reusable profile — the Canonical Payload Binding — so that each payload class declares its canonicalization algorithm and exclusion set once, obtains an interoperable derived identifier, and inherits statement-to-receipt binding and typed digest reference semantics without restating the mechanics in every profile. It complements the COSE Hash Envelope mechanism defined in RFC 9995: where that mechanism signals that a Signed Statement's payload is a digest standing in for content held elsewhere, this document defines how that digest is computed from structured content so that independently written implementations converge on the same bytes. An IANA registry governs the canonicalization algorithms; entries are immutable. This document defines no payload content formats and registers no artifact types; the artifact types that a typed reference may cite, and their meaning, are registered in a single shared Artifact Type Registry, governed separately from this document, that payload profiles register into. Note to Readers This document is an individual submission. The intended venue is the SCITT Working Group (scitt@ietf.org). Named attributions and acknowledgments in this document were individually confirmed in writing by the named parties. The short name "Canonical Payload Binding" and the document title are expected to be settled by the adopting working group. The source of this document and the companion interop record are maintained at: https://github.com/action-state-group/scitt-payload- binding Mih & Sokolov Expires 25 February 2027 [Page 1] Internet-Draft Canonical Payload Binding August 2026 Status of This Memo This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at https://datatracker.ietf.org/drafts/current/. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." This Internet-Draft will expire on 25 February 2027. Copyright Notice Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved. This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/ license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License. Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 1.1. Out of Scope . . . . . . . . . . . . . . . . . . . . . . 4 2. Changes from -01 . . . . . . . . . . . . . . . . . . . . . . 5 3. Conventions and Definitions . . . . . . . . . . . . . . . . . 8 4. Payload Canonicalization Algorithms . . . . . . . . . . . . . 10 4.1. Algorithm jcs . . . . . . . . . . . . . . . . . . . . . . 11 4.2. Algorithm jcs-n (Withdrawn) . . . . . . . . . . . . . . . 11 4.3. Algorithm cde-n (Withdrawn) . . . . . . . . . . . . . . . 12 4.4. Algorithm as-transmitted . . . . . . . . . . . . . . . . 13 5. The Derived Identifier . . . . . . . . . . . . . . . . . . . 14 5.1. Representation . . . . . . . . . . . . . . . . . . . . . 14 6. Envelope Conventions . . . . . . . . . . . . . . . . . . . . 15 7. Statement-to-Receipt Binding . . . . . . . . . . . . . . . . 15 7.1. Leaf Construction . . . . . . . . . . . . . . . . . . . . 16 8. Typed Digest References . . . . . . . . . . . . . . . . . . . 16 Mih & Sokolov Expires 25 February 2027 [Page 2] Internet-Draft Canonical Payload Binding August 2026 8.1. Cross-Profile Comparability . . . . . . . . . . . . . . . 18 8.2. Verification Scope . . . . . . . . . . . . . . . . . . . 20 9. Profile Independence . . . . . . . . . . . . . . . . . . . . 20 10. Discovery Mirror . . . . . . . . . . . . . . . . . . . . . . 20 11. Extensibility and Cross-Cutting Facilities . . . . . . . . . 21 12. Security Considerations . . . . . . . . . . . . . . . . . . . 21 12.1. Preimages Are Bytes, Not Renderings . . . . . . . . . . 21 12.2. Low-Entropy Fields . . . . . . . . . . . . . . . . . . . 22 12.3. Float Values and Digest Reproducibility . . . . . . . . 22 12.4. Immutable Coordinates . . . . . . . . . . . . . . . . . 22 12.5. Tamper Evidence and Runtime Honesty . . . . . . . . . . 23 12.6. Long-Term Verifiability Considerations . . . . . . . . . 23 13. Privacy Considerations . . . . . . . . . . . . . . . . . . . 23 14. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 24 14.1. Canonicalization Algorithm Registry . . . . . . . . . . 24 15. Related Work . . . . . . . . . . . . . . . . . . . . . . . . 27 16. References . . . . . . . . . . . . . . . . . . . . . . . . . 28 16.1. Normative References . . . . . . . . . . . . . . . . . . 28 16.2. Informative References . . . . . . . . . . . . . . . . . 28 Appendix A. Synthetic Registration Walkthrough . . . . . . . . . 30 Appendix B. Synthetic Two-Slot Composition . . . . . . . . . . . 31 Appendix C. Field-Verified Instances . . . . . . . . . . . . . . 32 C.1. Deep Mechanism Instances . . . . . . . . . . . . . . . . 33 C.1.1. Glyphzero Byte-Agreement — Algorithm Determinism . . 33 C.1.2. GAR Session Block — Leaf Construction Rule . . . . . 33 C.1.3. A2A Boundary Seal — Derived Identifier as Protocol Gate . . . . . . . . . . . . . . . . . . . . . . . . 34 C.2. Field Table — IETF 126 Participants . . . . . . . . . . . 34 C.3. Agreed and Scheduled . . . . . . . . . . . . . . . . . . 36 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . 36 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 37 1. Introduction Systems that anchor structured content to a SCITT Transparency Service [RFC9943] face a common sub-problem: how does a producer turn a JSON or CBOR object into a content-addressed Signed Statement whose identifier survives serialization, and how does a verifier check that the identifier in hand matches the bytes in hand? Each answer involves the same four moves — canonicalize, derive an identifier, bind a receipt, cite externals by digest — but they have been restated independently in every profile that needed them, with small variations that defeat interoperability. This document extracts those four moves into a single reusable profile called the Canonical Payload Binding (CPB). CPB is the missing piece the COSE Hash Envelope mechanism [RFC9995] deliberately leaves open: RFC 9995 defines how a Signed Statement signals that its Mih & Sokolov Expires 25 February 2027 [Page 3] Internet-Draft Canonical Payload Binding August 2026 payload field carries a hash rather than the content itself, but it does not say how that hash is computed from structured content so that two independently written implementations arrive at the same bytes. CPB fills that gap and stops there — it defines the canonicalization algorithm, the derived identifier it produces, the binding of that identifier to a Signed Statement and its Receipt, and a typed reference mechanism for citing other digests, and it defines nothing about what the hashed content means. CPB is derived from [I-D.mih-scitt-agent-action-capsule] (§Conventions, §envelope, §registration, §identity), which first stated the construction in a SCITT context, and generalized at the IETF 126 hackathon in Vienna, where seven parties participated in the public interop program. The public record reports four codebases demonstrating byte agreement in specific shared, declared contexts. Other frozen artifacts retained separately declared digest contexts. ORPRG retained its CP-JSON-2 context and was represented in the interop design through a typed reference rather than through an assertion of cross-profile digest equality. Digests remain governed by their original contexts; CPB does not relabel an ORPRG CP-JSON-2 commitment as a CPB canonicalization algorithm's output. The provenance is stated here once and not repeated in subsequent sections. For generic citation-binding verification, a CPB verifier can process a typed reference to any artifact type whose digest context it can resolve. Whether a particular citation slot permits that artifact type is determined by the consuming profile. Artifact-specific appraisal, authorization semantics, and application integration remain separate. Supporting a new artifact type requires no change to this document's citation-binding algorithm. Declaring the type, its digest context, and its meaning is a matter for the payload profile that defines it; it may also require consuming-profile integration and artifact- specific appraisal. 1.1. Out of Scope This document does not define: * Payload semantics — what fields a payload contains, what their values mean, or what verdicts or decisions are carried. Those belong to payload profiles that use CPB as their binding layer. * Artifact types and their digest contexts — which named categories of structured content exist, what fields and exclusion sets each declares, and which purpose labels its digest contexts use. Artifact types are registered in the shared Artifact Type Registry, governed separately from this document; CPB defines only Mih & Sokolov Expires 25 February 2027 [Page 4] Internet-Draft Canonical Payload Binding August 2026 the algorithms and the typed-reference container they use. (See [I-D.mih-scitt-agent-action-capsule] for an example payload profile that registers artifact types there.) * Application meaning — the real-world interpretation of any record anchored via this construction. * Transparency Service registration policy — which records a Transparency Service will or must accept. Registration policy is a Transparency Service concern, not a statement profile concern. * Transports — how registration requests or retrieval queries travel between producers, Transparency Services, or verifiers. 2. Changes from -01 The most consequential correction since -01 is registry-level: the registry was re-derived from what the field actually built, not from what -01 originally specified. jcs-n, live and Registered in -01, is withdrawn; jcs — the construction every independent implementation actually converged on — is registered in its place. The rest of this revision consolidates registry, canonicalization, and conformance- checker work landed since -01 was posted, and rescopes the document to its charter. Mih & Sokolov Expires 25 February 2027 [Page 5] Internet-Draft Canonical Payload Binding August 2026 *Charter rescope.* This document no longer normatively defines the Artifact Type Registry or any artifact-type-specific payload-shape rule. What changes is governance ownership, not location: REGISTRY.md does not move, and stays in this repository as the shared home for both registries this document's ecosystem uses. The Canonicalization Algorithm Registry (Section 14.1) remains CPB- normative. The Artifact Type Registry — its registration template, the purpose-label vocabulary, and both live entries (agent-action- capsule, machine-mandate) — is governed separately, by its own Designated Expert checklist and registration rungs already stated in REGISTRY.md, and this document references that registry rather than defining it. It is a single shared registry, not a per-profile one: [I-D.mih-scitt-agent-action-capsule] registers artifact types there alongside any other payload profile that wants to, each citing a CPB algorithm for its canonicalization; no one profile owns the registry. The worked walkthrough of Artifact-Type-Registry governance (Specification Required / Designated Expert / third-party registration) that -01 carried as an appendix is removed from this document, not moved — it belongs beside the registry it documents, in REGISTRY.md, where it already lives. This document now anchors [RFC9995] and keeps only the canonicalization algorithm(s), the derived identifier, Signed-Statement and Receipt binding, and the typed digest-reference container; the Abstract's former claim that this document governs "the artifact types" is corrected. *Registry.* * A machine-readable registry.json is now generated by CI from REGISTRY.md; releases pin a snapshot. A lookup against an identifier absent from the pinned snapshot but potentially valid in a later snapshot now returns a distinct verdict, id-unknown-to- snapshot, rather than being indistinguishable from a genuinely unknown identifier. * REGISTRY.md gained an onboarding ladder and a controlled entry- status vocabulary (owner-confirmed, third-party-documented, provisional, standards-referenced), a three-rung registration path (owner-authored / third-party-documented / provisional) with a template, lifecycle, and removal/correction path, and Designated Expert review stated explicitly as a precondition of merging an entry rather than a status a merged entry can still assert. This infrastructure is shared by every registry this file hosts, including the shared Artifact Type Registry, which [I-D.mih-scitt-agent-action-capsule] registers into alongside every other payload profile that does — no single profile owns it. Mih & Sokolov Expires 25 February 2027 [Page 6] Internet-Draft Canonical Payload Binding August 2026 * The registry generator and validator now source legal status values from REGISTRY.md itself instead of a hardcoded list, and reject a malformed table row (mismatched cell/header count) closed instead of silently mis-assigning columns; a small number of rows that predate the controlled vocabulary are named explicitly as the only ones permitted a legacy status spelling. * A Reserved placeholder token no longer reads as Verified: registry lookups previously returned a "verified" verdict on entry presence alone; a distinct VERDICT_RESERVED verdict now applies to any entry present but not in Registered status. * The required-fields table, the "immutable" language, and the upgrade-acknowledgment gate were corrected: the required-fields table is scoped to new entries; "immutable" is stated consistently as "immutable in behavior"; and the acknowledgment gate now requires the same consuming-profile acknowledgment for an upgrade to owner-confirmed that it already required for initial admission, closing a path where an owner could file without an acknowledgment and self-acknowledge afterward. *Canonicalization algorithms.* jcs — plain RFC 8785 JCS, no normalization pass — is registered (Section 4.1), with a named consuming profile and a discriminating vector against jcs-n: one payload carrying a null member and an empty array (jcs preserves both, jcs-n stripped them) plus a float member (jcs admits it, jcs-n rejected it), failing loudly in both directions. jcs-n is withdrawn (Section 4.2) — the same terminal-marking disposition cde-n already carried in -01 — following an implementer census (the reference implementation was the only implementer of the normalization step it added), a byte audit showing 191 of 203 evaluated records were byte- identical under plain jcs without that step, the 12 divergent records being proof-of-concept artefacts retained by vintage, and the admission bar this revision applies to every entry: a named consuming profile. Separately, a cross-language conformance harness (vectors/ CANONICALIZATION_DECLARATION.md) versions jcs-n's construction precisely enough for an independent implementation to conform against without reading the reference library; it stands as part of the permanent historical record for the now-withdrawn algorithm. The lowercase-\u string-escaping rule and the corresponding control- character sort order — properties of RFC 8785 JCS itself, and therefore shared by jcs and the withdrawn jcs-n alike — are now stated in prose and cross-linked from REGISTRY.md. The shared JCS serialization helper also now rejects non-finite numeric values (Infinity, -Infinity, NaN) before serialization, consistent with RFC 8785 Section 3.2.2.3 admitting finite values only. Mih & Sokolov Expires 25 February 2027 [Page 7] Internet-Draft Canonical Payload Binding August 2026 *Digest determinism and typed references.* Two paragraphs now state explicitly what -01 only implied: each algorithm entry and each artifact type's digest-context declaration names exactly one hash algorithm, so digest_alg is fully determined by type (together with purpose where needed) for any registered reference, and a verifier encountering a digest_alg inconsistent with the resolved context MUST treat it as a failure and MUST NOT attempt to reconcile it (Section 8.1). A MUST-FAIL/PASS vector pair pins that an assembled pre-image — one built from selected source fields rather than the payload minus an exclusion set — is under-determined by algorithm and field set alone; producer-chosen member naming and nesting are part of the bytes. Two conformance-checker categories exercise this: recomputing both pinned pre-images and asserting they diverge for exactly the demonstrated reason, and applying a declared member_mapping to assert it reproduces the vector's own input. Contributed by Rul1an as an external submission, reproduced independently against the reference canonicalizer. *Conformance checker.* A grammar/wire-layer conformance checker (cpb- check) validates a record against its declared profile grammar — a presence-and-number-form walk and duplicate-key rejection — built around a duplicate-preserving raw-bytes lexer, since a standard JSON parser silently drops duplicate keys before any rule can see them; digest recomputation and canonicalization_id resolution remain out of scope pending a later gate. Vector-harness fixes landed alongside it: the lexer now rejects trailing bytes after a JSON document ends and NFC-normalizes before duplicate-key detection, and an inverted must-fail assertion and a -0/duplicate-key gap that could previously let the harness certify a vector as passing for the wrong reason are both closed. 3. Conventions and Definitions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here. Payload Class: A named category of structured content that has declared a canonicalization algorithm (from the registry in Section 14.1) and an exclusion set of fields that are omitted from the canonical form before the derived identifier is computed. A payload class is declared by the payload profile that defines it; this document does not maintain a registry of payload classes or artifact types. Derived Identifier: The content-address of a payload: the output of Mih & Sokolov Expires 25 February 2027 [Page 8] Internet-Draft Canonical Payload Binding August 2026 CANONICAL-DIGEST applied to the canonical form of the payload with the exclusion set removed. Verifiers MUST recompute the derived identifier from the payload bytes; a carried derived-identifier value is advisory only and a mismatch is a defect. Digest Context: The complete set of parameters that determine how a digest was computed: the field set selected, the exclusion set applied, the canonicalization algorithm applied, any domain separation, the encoding of the pre-image, and the representation of the output. Two digest values are comparable only when their full digest contexts are established as compatible. A payload class or artifact type MAY declare more than one digest context over the same payload, each serving a distinct purpose declared by the payload profile that defines the class or type; the contexts are independent and MUST NOT be conflated. CANONICAL-DIGEST: A function parameterized by a canonicalization algorithm A: for any such algorithm A and payload v, CANONICAL- DIGEST(A, v) = ENCODE_A(H_A(A(v))), where H_A is the digest function and ENCODE_A the output encoding declared by A's entry in the Canonicalization Algorithm Registry (Section 14.1). Every entry registered by this document declares SHA-256 and 64-character lowercase hexadecimal; an entry registered by a later document MAY declare another digest function or encoding, and a verifier MUST read both from the entry rather than assuming them. A(v) is the octet string produced by the algorithm applied to v; the specific pre-image construction — field selection, normalization, and encoding — is part of A's definition and is registered per Section 14.1. Signed Statement: A COSE_Sign1 object [RFC9052] that carries a payload, a protected header, and an optional unprotected header; defined in [RFC9943]. Receipt: A COSE structure produced by a Transparency Service that provides verifiable evidence that a Signed Statement was registered; defined in [RFC9943] and format-governed by the Verifiable Data Structure of the service. Transparent Statement: A Signed Statement to whose unprotected header one or more Receipts have been attached. Verifier: Any party that validates a record from its bytes, without trusting the producer. Mih & Sokolov Expires 25 February 2027 [Page 9] Internet-Draft Canonical Payload Binding August 2026 4. Payload Canonicalization Algorithms A canonicalization algorithm specifies how to produce a canonical octet string from a structured value. The canonical octet string is the pre-image to CANONICAL-DIGEST. A payload class declares exactly one canonicalization algorithm; verifiers MUST NOT guess the algorithm from the payload shape. The algorithms defined in this document and registered in the Canonicalization Algorithm Registry (Section 14.1) are: +================+================================+=============+ | Name | Summary | Reference | +================+================================+=============+ | jcs | Plain RFC 8785 JCS, no | Section 4.1 | | | normalization pass; SHA-256; | | | | lowercase hex output | | +----------------+--------------------------------+-------------+ | jcs-n | Withdrawn -- JCS + absent- | Section 4.2 | | | field normalization; never | (withdrawn) | | | carried to IANA | | +----------------+--------------------------------+-------------+ | cde-n | Withdrawn -- token reserved, | Section 4.3 | | | never assigned a definition | (withdrawn) | +----------------+--------------------------------+-------------+ | as-transmitted | No canonicalization; digest | Section 4.4 | | | over a byte sequence fixed by | | | | a cited named production in | | | | the container format; SHA-256; | | | | 64-character lowercase hex | | +----------------+--------------------------------+-------------+ Table 1 Entries in the Canonicalization Algorithm Registry are immutable: new behavior requires a new entry, never a retroactive edit to an existing one. A reserved entry binds its token only; its summary is provisional until the entry is defined, at which point the full entry becomes immutable. A reserved entry may instead be withdrawn (Section 4.3, Section 4.2), which is terminal: the token stays bound, no definition is ever assigned (or, for an entry that was already defined, no further definition ever attaches to it), and the name is not reassigned. The hash function is part of each algorithm's definition; migration to a different hash (for example, a future post-quantum function) is performed by registering a new algorithm entry, never by reinterpreting an existing one. Mih & Sokolov Expires 25 February 2027 [Page 10] Internet-Draft Canonical Payload Binding August 2026 4.1. Algorithm jcs Algorithm jcs is the JSON Canonicalization Scheme [RFC8785] applied directly to the payload, with no normalization pass: no member is removed because its value is JSON null, an empty array, or an empty object. Pre-image construction: 1. Apply JCS [RFC8785] to the octets supplied to the algorithm, to produce the canonical UTF-8 octet string. Exclusion-set removal is not part of this algorithm: the derived identifier construction (Section 5) removes the payload class's declared exclusion set before invoking the algorithm. 2. Compute SHA-256 over those octets. 3. Encode the digest as lowercase hexadecimal. The output is a 64-character ASCII string. The CANONICAL-DIGEST of a payload P using jcs is therefore: CANONICAL-DIGEST(jcs, P) = lowercase_hex(SHA-256(JCS(P))) The exclusion set is matched against the top-level member names of P only; a member of the same name nested inside a member's value is not removed. jcs places no additional restriction on JSON numbers beyond RFC 8785 itself: a JSON floating-point number is permitted and is serialized per the canonical ECMAScript-based number-to-string procedure RFC 8785 [RFC8785] Section 3.2.2.3 defines for IEEE 754 double-precision values. Two conforming implementations that parse the same numeric literal into the same double-precision value therefore produce byte- identical output; see Section 12.3. A payload profile MAY still declare its own stricter constraint (for example, requiring monetary fields to be exact decimal strings) — such a constraint is a payload- profile decision, not a requirement of this algorithm. 4.2. Algorithm jcs-n (Withdrawn) Algorithm jcs-n is withdrawn (2026-08-18) -- terminal marking, never deletion: the token stays bound, the definition it once carried is not reassigned, and it is never carried forward to IANA. That is a terminal marking that cde-n (Section 4.3) also carries, though on different facts: cde-n never acquired a definition, while jcs-n did and its records remain verifiable by vintage. jcs-n applied JCS Mih & Sokolov Expires 25 February 2027 [Page 11] Internet-Draft Canonical Payload Binding August 2026 [RFC8785] to an absent-field-normalized JSON object -- the normalization step removed, bottom-up and recursively, every member whose value was JSON null, an empty array, or an empty object, before JCS serialization. The full original construction is the permanent record in draft-mih-sokolov-scitt-payload-binding-00, Section 3.1, and is not restated here. The withdrawal followed from an implementer census (the reference implementation was the only implementer of the normalization step), a byte audit showing 191 of 203 evaluated records were byte-identical under plain jcs without it, the 12 divergent records being proof-of- concept artefacts retained by vintage, and the admission bar this document now applies to every entry: a named consuming profile. jcs (Section 4.1) is the entry that replaces it going forward; a payload class or typed digest reference that named jcs-n used the withdrawn construction described above, and a party citing that historical construction going forward registers a new entry rather than resuming use of this token. Withdrawal forecloses new declarations of jcs-n; it does not retroactively invalidate records already sealed under it. A payload class or typed digest reference that names jcs-n MUST NOT be newly declared. A verifier encountering jcs-n in a record committed on or after 2026-08-18 MUST fail closed — MUST NOT report the payload class or typed digest reference as verified. A verifier encountering jcs-n in a record committed before 2026-08-18 MAY verify it against the withdrawn construction as that construction is permanently recorded in draft-mih-sokolov-scitt-payload-binding-00, Section 3.1; such a record is a historical record, not a live conformance case, and a verifier that declines to implement the withdrawn construction MUST report the reference as unverified rather than as failed. A historical identifier MUST NOT be relabelled to another algorithm token or recomputed under another algorithm. 4.3. Algorithm cde-n (Withdrawn) Algorithm cde-n is withdrawn. It is a recorded terminal state, not a deletion: the token was reserved for a deterministic CBOR canonicalization profile, but it was never assigned a definition, and it will not be. The entry remains in the Canonicalization Algorithm Registry (Section 14.1) as withdrawn -- the reserved entry bound the token, so the token stays bound, never assigned, never reassigned. A future deterministic CBOR canonicalization profile, if one is specified, is registered under a new token rather than by assigning a definition to cde-n. Mih & Sokolov Expires 25 February 2027 [Page 12] Internet-Draft Canonical Payload Binding August 2026 A payload class or typed digest reference that names cde-n cannot be verified: the token names no defined algorithm and never will, so a verifier encountering it MUST fail closed — MUST NOT report the payload class or typed digest reference as verified. 4.4. Algorithm as-transmitted Algorithm as-transmitted applies no canonicalization. The digest pre-image is the exact octet sequence already fixed by the container format or cryptographic envelope carrying the payload -- for example, the signing input over which a signature was computed. The signature (or other format-defined byte-fixing) is what makes those bytes authoritative; re-canonicalizing them would be redundant at best and would break the very binding that makes the bytes authoritative at worst. Because there is no canonicalization step, as-transmitted has no field set and no exclusion set. An artifact type entry that declares as-transmitted as its canonicalization algorithm MUST instead state a byte-boundary selector in place of a field set: a normative reference plus the name that referenced specification gives to the exact byte sequence in question. Two examples of a valid selector: * RFC 7515 §5.1, JWS Signing Input -- the octets a JWS signature is computed over. * RFC 9052 §4.4, ToBeSigned -- the octets a COSE_Sign1 signature is computed over. A selector that is not a cited named production is prose, not a selector, and this registry exists to eliminate exactly that kind of ambiguity: an artifact type MUST NOT register as-transmitted on the strength of an uncited description such as "the payload bytes." If the container specification carrying the artifact does not itself name the exact byte sequence as a discrete production, the artifact type MUST NOT use as-transmitted -- it registers a canonicalization algorithm instead, one that defines the pre-image construction from first principles. The CANONICAL-DIGEST of a byte sequence B identified by the declared byte-boundary selector is: CANONICAL-DIGEST(as-transmitted, B) = lowercase_hex(SHA-256(B)) Digest: SHA-256, 64-character lowercase hex, matching jcs. These are stated explicitly here as part of this entry, not inherited silently from the generic CANONICAL-DIGEST definition (Section 3). Mih & Sokolov Expires 25 February 2027 [Page 13] Internet-Draft Canonical Payload Binding August 2026 5. The Derived Identifier The derived identifier of a record is computed as: id = CANONICAL-DIGEST(A, payload minus exclusion_set) where A is the canonicalization algorithm declared by the payload class and the exclusion set is the set of fields declared by the payload class as self-referential or chain-linkage fields. The derived identifier is a 64-character lowercase hex string for every algorithm this document registers; for an algorithm registered elsewhere, its representation is the one that algorithm's registry entry declares. The exclusion set MUST be declared by the payload class in its specification. Fields excluded are those that either contain the derived identifier itself (they cannot be inside the pre-image they help compute) or that reference other records in a chain (to keep the content-address stable regardless of what later chains to this record). The exclusion set is normative for the payload class; a verifier MUST apply the same exclusion set as the producer. A producer MAY carry the derived identifier as a field in the payload. A verifier MUST recompute the identifier from the payload bytes and the declared exclusion set. If the recomputed value does not match the carried value, the verifier MUST treat this as a defect in the record. When selective disclosure is in use, the derived identifier MUST be computed over the SD-encoded form of the payload, not the plaintext payload. A payload profile MUST declare non-eligible for selective disclosure any field that the profile's own verifier requires in order to evaluate the binding. 5.1. Representation Representation is normative and MUST be declared by the payload class. The following representations are distinct and are not implicitly interchangeable: * bare 64-character lowercase hexadecimal text; * prefixed textual representation; and * raw 32-byte octet sequence. Mih & Sokolov Expires 25 February 2027 [Page 14] Internet-Draft Canonical Payload Binding August 2026 A payload class MUST specify which representation it uses for each field containing or referencing a derived identifier. A verifier MUST NOT silently coerce among representations. A deterministic conversion MAY be applied only where this specification or the applicable payload profile expressly defines both the conversion and the resulting comparison representation. Such a conversion is an explicit protocol operation and does not make the original representations byte-identical. 6. Envelope Conventions A Signed Statement carrying a CPB-bound payload MUST be a COSE_Sign1 [RFC9052] structure. The protected header MUST carry: * alg: the signing algorithm. * kid or x5chain: the signing key identifier or certificate chain. * content_type: the media type of the payload, as application/ CLASS+json or application/CLASS+cbor according to the serialization the payload class declares, where CLASS is the payload class name as declared by the payload class's own defining specification. A field belongs in the protected header only if a SCITT-generic party — a Transparency Service registration policy or a profile-unaware verifier — must act on it without understanding the payload class. Everything semantically specific to the payload class stays in the payload. Protected-header claims are a closed set per payload class: extensions are payload-only. A Transparency Service that does not understand a protected-header extension MUST be able to register the Signed Statement and verify the envelope without it. The closed-claim principle does not prevent payload-class-specific protected-header fields from existing; it requires that such fields be defined by the payload class specification, not added ad-hoc by producers. 7. Statement-to-Receipt Binding A producer makes a record transparent by registering its Signed Statement with a SCITT Transparency Service per [RFC9943] and attaching the returned Receipt to the unprotected header, forming a Transparent Statement. Mih & Sokolov Expires 25 February 2027 [Page 15] Internet-Draft Canonical Payload Binding August 2026 This profile is VDS-agnostic at the statement layer. Receipt format and proof verification are governed by the Verifiable Data Structure (VDS) of the Transparency Service; this profile imposes no VDS requirement. A verifier MUST NOT report receipt-backed status without having verified a Receipt from a Transparency Service under a key the verifier trusts. A verifier determining which VDS to apply when verifying a Receipt MUST read the VDS identifier from the protected header of the Receipt. The verifier MUST NOT infer the VDS from the COSE structure of the receipt alone. Unknown VDS identifiers MUST be rejected. 7.1. Leaf Construction This profile imposes no leaf construction on a Verifiable Data Structure. Where a Transparency Service's VDS keys its log on the derived identifier, the derived identifier is a 32-byte value and its hexadecimal form is a representation of that value (Section 5.1); a VDS or profile that keys on it therefore states which of the two it uses, and producer and verifier MUST use the same one. The following is the failure this requirement exists to prevent. That is, for a derived identifier whose string value is a 64-character hex string D, the log leaf input MUST be the raw 32-byte value: leaf_input = bytes.fromhex(D) -- correct: 32 raw bytes The following is incorrect and MUST NOT be used: leaf_input = D.encode("utf-8") -- WRONG: 64 ASCII bytes A verifier constructing the leaf for proof verification MUST apply the same rule. Failure to distinguish the byte sequence from its hex encoding produces a silently wrong leaf hash that fails inclusion verification against any correct log. 8. Typed Digest References A typed digest reference is the mechanism by which one record cites an external artifact — another record, an authorization document, a configuration object, or any other verifiable item — by its content- address without embedding it. A typed digest reference is a JSON object with the following fields: Mih & Sokolov Expires 25 February 2027 [Page 16] Internet-Draft Canonical Payload Binding August 2026 +==========+======+===========+====================================+ |Field |Type |Req |Meaning | +==========+======+===========+====================================+ |type |string|REQUIRED |The artifact type identifier. This | | | | |document defines the reference | | | | |container and its verification | | | | |algorithm; it does not itself | | | | |register artifact types or resolve | | | | |type values to digest contexts. | | | | |That resolution is provided by the | | | | |shared Artifact Type Registry, into | | | | |which the payload profile that | | | | |declares the cited artifact type | | | | |registers it (see | | | | |[I-D.mih-scitt-agent-action-capsule]| | | | |for an example). | +----------+------+-----------+------------------------------------+ |purpose |string|CONDITIONAL|The purpose label selecting which of| | | | |the artifact type's digest contexts | | | | |this reference targets, drawn from | | | | |the vocabulary the type's entry in | | | | |the shared Artifact Type Registry | | | | |defines. REQUIRED whenever the | | | | |resolved artifact type declares more| | | | |than one digest context. MAY be | | | | |omitted only when the resolved | | | | |artifact type declares exactly one | | | | |digest context, in which case that | | | | |single context applies; a verifier | | | | |MUST NOT infer a default when more | | | | |than one context is declared. | +----------+------+-----------+------------------------------------+ |digest_alg|string|REQUIRED |The hash algorithm of the digest | | | | |value (e.g., "SHA-256"). The | | | | |canonicalization context of the | | | | |cited artifact is resolved from the | | | | |digest context selected by type and | | | | |purpose, not from this field. | +----------+------+-----------+------------------------------------+ |digest |string|REQUIRED |The digest of the cited artifact, in| | | | |the representation declared by the | | | | |selected digest context. | +----------+------+-----------+------------------------------------+ Table 2 Additional fields MAY be present and MUST be ignored by verifiers that do not understand them. Mih & Sokolov Expires 25 February 2027 [Page 17] Internet-Draft Canonical Payload Binding August 2026 8.1. Cross-Profile Comparability Within typed-reference verification, the digest carried by the reference and the digest recomputed over the referenced artifact are comparable only when both are interpreted under the same established referenced-artifact digest context and comparison representation. If the verifier cannot resolve a digest context for the value of type, it MUST NOT report the typed reference as verified; the reference is present but not verified. Two situations produce that outcome and a verifier MUST distinguish them in what it reports, because they call for different responses: * The type is absent from every registry the verifier consults. No payload profile has declared a digest context under that name, and the citation becomes verifiable only once one does. * The type is declared somewhere, but absent from the particular registry snapshot the verifier holds, which may predate an entry that does exist. The remedy is to obtain a current snapshot, not to seek a new registration. A verifier that reports these as one condition sends an implementer to fix the wrong thing. A verifier that cannot tell them apart -- because it holds no snapshot version -- MUST report the weaker of the two, that its snapshot may be stale. The consuming profile determines the disposition, and a profile MUST state what it does with a present-but-not-verified reference. A citation carrying an unresolvable type is not an error in the citing record. It is also not evidence: Section 12.4 requires that citations pin content by CANONICAL-DIGEST precisely so that an unverified reference cannot be relied on, so a profile MUST NOT treat "not an error" as permission to proceed as though the reference had verified. To verify the reference, the verifier MUST use the type field, together with the purpose field when the resolved artifact type declares more than one digest context, to resolve exactly one of the referenced artifact's declared digest contexts. If type resolves to more than one digest context and purpose is absent, ambiguous (matching no purpose label the resolved artifact type declares), or names a purpose label the resolved artifact type does not declare, the reference is unresolvable: the verifier MUST NOT guess a context and MUST NOT report the typed reference as verified. It MUST confirm that digest_alg identifies a hash algorithm consistent with the resolved context. Mih & Sokolov Expires 25 February 2027 [Page 18] Internet-Draft Canonical Payload Binding August 2026 digest_alg is REQUIRED even though every algorithm registered in Section 14.1 today names the same hash, SHA-256: it is the field that lets a future Canonicalization Algorithm Registry entry using a different hash land as a new token without a breaking change to this wire format, rather than being decorative because only one value is legal now. The hash algorithm is not chosen per-reference: each entry in the Canonicalization Algorithm Registry names its hash function as an immutable part of its definition (Section 4), and each artifact type's own digest-context declaration names exactly one such algorithm. digest_alg is therefore fully determined by type (together with purpose where needed): a conforming reference can only carry the hash algorithm the resolved digest context mandates. It is a redundant consistency declaration by design — hash-in-algorithm is what makes Canonicalization Algorithm Registry entries immutable and enables long-term algorithm migration by registering a new entry rather than reinterpreting an existing one. It MUST then recompute the referenced artifact's digest under that context and compare the recomputed value with the value carried in the digest field. *Comparison is byte-for-byte.* A verifier compares digest_alg against the name the resolved digest context mandates as an exact octet sequence: no case folding, no alias table, no whitespace trimming. sha-256 does not match SHA-256. The two IANA registries an implementer is likely to reach for disagree on spelling for the same function, so a case-insensitive or alias-tolerant comparison silently accepts a reference that names a different registry's token — and once one implementation tolerates it, the field stops being a consistency declaration and becomes decoration. The registered name is the one the Canonicalization Algorithm Registry entry states. A digest_alg value that does not name the hash algorithm mandated by the resolved digest context is a defect in the reference. The verifier MUST treat this as a failure and MUST NOT attempt to reconcile the inconsistency — for example, by silently proceeding with the algorithm the registry mandates and ignoring the mislabeled field. More generally, if the context established from the type and digest_alg fields cannot be reconciled with the context used to recompute the referenced artifact, or if a required deterministic conversion to a common comparison representation is not expressly defined, the verifier MUST NOT report the typed reference as verified. The failure verdict is mandatory at the verifier layer; the consuming profile determines the resulting error disposition, but not the verdict itself. Mih & Sokolov Expires 25 February 2027 [Page 19] Internet-Draft Canonical Payload Binding August 2026 The citing record's own derived-identifier context need NOT be compatible with the referenced artifact's digest context; those contexts govern different computations. The two values actually being compared must share an established comparison context. Bare hexadecimal equality alone is not a join. 8.2. Verification Scope Successful verification of a typed digest reference establishes content binding to the referenced artifact under the declared digest context. CPB verification alone MUST NOT be interpreted as establishing issuer authority, artifact validity, scope, freshness, revocation status, policy compliance, semantic acceptance, or application authorization. Any appraisal required by the referenced artifact type or consuming application profile remains a separate verification step. Missing, indeterminate, or failed required appraisal MUST NOT be treated as authorization success. The interchangeability property of typed digest references -- that any artifact type whose digest context can be resolved may fill a citation slot -- applies to citation-binding interoperability only and does not extend to any appraisal or authorization semantics defined by the artifact type or consuming profile. 9. Profile Independence A payload profile MUST NOT impose requirements on the internal structure or field values of another payload profile. Relationships between artifacts of different types are expressed solely through typed references (Section 8) that resolve against each artifact type's own digest-context declaration. This constraint keeps verification of a multi-artifact chain decomposable: a verifier evaluates each binding under each profile's own semantics independently and never needs to evaluate a pair of profiles jointly. Implementations therefore need not implement, or be aware of, profiles they neither produce nor consume, and a new profile may declare its own artifact types without revalidating existing profiles or implementations. 10. Discovery Mirror This section is informative. Mih & Sokolov Expires 25 February 2027 [Page 20] Internet-Draft Canonical Payload Binding August 2026 A producer MAY place an unprotected COSE header parameter that mirrors the derived identifier of the record. This parameter is advisory only: it allows log tooling, registration policies, and cross-grain citation to locate a record's content-address without parsing the payload, but it carries no binding guarantee. A verifier MUST recompute the derived identifier from the payload. A mismatch between the advisory mirror value and the recomputed value is a defect in the record and MUST be reported. The discovery mirror parameter is aligned with the trace-metadata convention in draft-birkholz-verifiable-agent-conversations §7.4 [I-D.birkholz-verifiable-agent-conversations], which defines a similar unprotected-header mechanism for conversation-grain records. A record using CPB at the action grain and a conversation container using that convention can share one discovery layer. 11. Extensibility and Cross-Cutting Facilities This section is informative. Several concerns are common to all payload profiles and, if defined independently per profile, would undermine decomposable verification or fragment the interoperability surface: selective disclosure, countersignature and multi-party attestation, record relations (supersedes, confirms, corrects), erasure tombstones, producer timestamps and validity periods, batch aggregation, and profile versioning. This specification does not define these facilities in this document. Each will be addressed in a companion document that payload profiles MUST reference rather than developing an incompatible per-profile variant. Defining any of these facilities per-profile would violate the constraint established in Section 9. 12. Security Considerations 12.1. Preimages Are Bytes, Not Renderings The pre-image of a CANONICAL-DIGEST is the octet string produced by the canonicalization algorithm — not a rendered form, not a console output, and not a string with added whitespace, trailing newlines, or encoding differences. A producer that serializes then re-reads the payload before computing the digest MUST ensure the byte sequence entering SHA-256 is identical to what the canonicalization algorithm produces, not what a deserializer happens to emit. Diagnosing divergence requires comparing the exact octets, not visual representations. Mih & Sokolov Expires 25 February 2027 [Page 21] Internet-Draft Canonical Payload Binding August 2026 12.2. Low-Entropy Fields A digest hides its pre-image only to the degree the pre-image space is large and unguessable. When a committed value is drawn from a small enumeration, a short identifier, or a bounded numeric range, an adversary can reconstruct it by enumerating candidates and matching digests. A payload class SHOULD commit low-entropy fields under a per-issuer salt or via a selective-disclosure mechanism (see the SD- JWT commitment pattern in [RFC9901]) rather than digesting the bare value. Bare digests of low-entropy fields are not confidential. 12.3. Float Values and Digest Reproducibility Different JSON implementations can serialize the same numeric quantity ([RFC8259] number values that are not integers) as 1.0, 1e0, or 1.00; a canonicalization algorithm's number-serialization rule determines whether that variation survives into the digest pre-image. Algorithm jcs (Section 4.1) inherits RFC 8785's canonical ECMAScript- based number-to-string procedure ([RFC8785] Section 3.2.2.3), which fixes one serialization per IEEE 754 double-precision value; two conforming implementations that parse the same numeric literal into the same double-precision value therefore produce byte-identical output under jcs. That guarantee is bounded by parsing, not by canonicalization: a JSON parser that rounds a numeric literal to a different double-precision value than another parser produces a different pre-image under any algorithm, jcs included. A payload profile for which this residual risk is unacceptable — for example, one carrying monetary or quantity values — MAY declare its own stricter constraint, such as requiring exact decimal strings instead of JSON numbers, in the fields it selects for digesting; such a constraint is a payload-profile decision, not a requirement this document imposes on every payload class. 12.4. Immutable Coordinates A mutable reference — a branch name, a tag that can be moved, a content URL that is not a content-addressed URL — is not evidence. The moment a record is amended at its referent, any citation to the mutable reference silently refers to the new content. All citations to external artifacts MUST use typed digest references (Section 8) that pin the content by its CANONICAL-DIGEST. Names, labels, and human-readable identifiers MAY appear alongside a typed reference for display purposes but carry no evidentiary weight. Mih & Sokolov Expires 25 February 2027 [Page 22] Internet-Draft Canonical Payload Binding August 2026 When an artifact type cited in an immutable coordinate has no resolvable digest-context declaration at verification time, the citation is present but not verified; the consuming profile determines the disposition (Section 8.1). This is not a defect in the citing record: the citation becomes verifiable once a conforming declaration exists. 12.5. Tamper Evidence and Runtime Honesty The envelope signature and the registration Receipt provide tamper evidence for the record's bytes and bound its timing. They do not prove the recording runtime was honest at the moment of recording. A producer that seals a false record produces a structurally valid record of a fiction. A Transparency Service's append-only property bounds the timing of such a record and makes its omission or substitution detectable; it does not make its content true. 12.6. Long-Term Verifiability Considerations Artifacts bound under this specification may need to remain verifiable over periods considerably longer than the lifetime of any particular digest or signature algorithm. Because a binding is expressed in terms of a registered algorithm identifier rather than a fixed algorithm, artifacts bound under different algorithms are each well-formed and independently verifiable. Preserving verifiability across an algorithm transition requires that evidence be re-established under a stronger algorithm _before_ the original is considered weak; this cannot be done retroactively. Deployments with long retention requirements SHOULD adopt an evidence-renewal scheme. [RFC4998] specifies one such scheme and distinguishes timestamp renewal, which operates on archived evidence alone, from hash-tree renewal, which requires access to the original data objects. This specification does not mandate a particular scheme. 13. Privacy Considerations A record bound under this profile carries digests of content rather than the content itself. The derived identifier and any typed digest references commit to the content without disclosing it; the record is therefore payload-blind to any verifier that does not independently possess the referenced artifacts. Mih & Sokolov Expires 25 February 2027 [Page 23] Internet-Draft Canonical Payload Binding August 2026 Payload privacy is the responsibility of the payload class. A payload class that includes fields identifying persons, sessions, or request content SHOULD document the privacy properties of those fields, including whether they can be inferred from their digests given knowledge of the value space. Low-entropy fields are not confidential even when digested (Section 12). An anchored record cannot be retracted: a Transparency Service's log is append-only and a registered record persists. Payload classes SHOULD specify which fields, if any, must not be present in a record that is intended to be anchored. 14. IANA Considerations This document requests the creation of one new IANA registry, the Canonicalization Algorithm Registry (Section 14.1), under a "Canonical Payload Binding" heading. The registry uses the Specification Required policy ([RFC8126], Section 4.6); a Designated Expert is required for each registration. This document does not define an Artifact Type registry: artifact types are registered in the shared Artifact Type Registry in REGISTRY.md, governed separately from this document under its own Designated Expert checklist and registration rungs; this document references that registry (see Section 1.1) but does not define it. This revision's Canonicalization Algorithm Registry entries reflect a deliberate correction over -01's: the registry was re-derived from what the field actually built, rather than restated from what -01 originally specified (Section 2). Registry entries are immutable. A registered entry defines a specific algorithm. If a behavior change is needed, a new entry MUST be registered; existing entries MUST NOT be modified retroactively. Maintainer is IANA per standard process; no other governance body is defined. Until this registry comes into existence at RFC publication, the table below serves as the provisional living registry, maintained in this document's source repository. If the document is adopted, the provisional registry moves with the document to a repository of the working group's choosing. 14.1. Canonicalization Algorithm Registry This registry records the canonicalization algorithms that may be used to compute CANONICAL-DIGEST values. Mih & Sokolov Expires 25 February 2027 [Page 24] Internet-Draft Canonical Payload Binding August 2026 Each entry pins its canonicalization steps, its hash function, and its output representation together as a single immutable triple, so that changing any one of the three requires registering a new token rather than reinterpreting an existing one — otherwise a token such as jcs would silently come to mean more than its name states. Registration template: * Name: A short ASCII identifier suitable for use in protocol fields. * Description: A normative prose description sufficient to implement the algorithm deterministically. * Reference: The document that specifies the algorithm. Initial contents: Mih & Sokolov Expires 25 February 2027 [Page 25] Internet-Draft Canonical Payload Binding August 2026 +=============+======================================+=============+ | Name | Description | Reference | +=============+======================================+=============+ | jcs | RFC 8785 JCS over the octets | This | | | supplied to the algorithm, no | document | | | normalization pass; SHA-256; | | | | 64-character lowercase hex | | +-------------+--------------------------------------+-------------+ | jcs-n | Withdrawn (2026-08-18) -- never | This | | | carried to IANA. The token was | document | | | reserved and defined a JCS-plus- | (withdrawn) | | | absent-field-normalization | | | | construction, but that construction | | | | is not carried forward; the | | | | permanent record of the construction | | | | is draft-mih-sokolov-scitt-payload- | | | | binding-00, Section 3.1 | | +-------------+--------------------------------------+-------------+ | cde-n | Withdrawn (2026-08-18) -- never | This | | | carried to IANA. The token was | document | | | reserved and stays bound; it was | (withdrawn) | | | never assigned a definition and | | | | never will be | | +-------------+--------------------------------------+-------------+ | as- | No canonicalization: the pre-image | This | | transmitted | is the exact octet sequence | document | | | identified by a cited named | | | | production in the container format | | | | (e.g., a signature's signing input); | | | | an artifact type using this | | | | algorithm states a byte-boundary | | | | selector in place of a field set; | | | | SHA-256; 64-character lowercase hex | | +-------------+--------------------------------------+-------------+ Table 3 A payload class or typed digest reference naming cde-n MUST NOT be treated as verifiable under any vintage: the token was bound by a reserved entry but never assigned a definition, so no construction exists to verify against, and a verifier encountering it MUST fail closed. A payload class or typed digest reference naming jcs-n MUST NOT be newly declared; records committed under it before 2026-08-18 are governed by the vintage rule in Section 4.2. Both withdrawals are recorded terminal states, not deletions: the tokens stay bound and are never assigned or reassigned. See Section 4.3 and Section 4.2. Mih & Sokolov Expires 25 February 2027 [Page 26] Internet-Draft Canonical Payload Binding August 2026 An artifact type MUST NOT declare as-transmitted without a byte- boundary selector that cites a named production in the container specification (Section 4.4). Without that selector, an as- transmitted declaration states nothing: there is no field set, no exclusion set, and no canonicalization to fall back on for the pre- image construction. 15. Related Work COSE Hash Envelope ([RFC9995]) is the hash-side sibling: it defines how to carry a content-addressed reference to an opaque payload in a COSE structure. CPB is the statement-side complement: it defines how the payload content is canonicalized and identified so that the content-address is reproducible across implementations. The CCF Receipt Profile ([I-D.ietf-scitt-receipts-ccf-profile]) and COSE Receipts ([RFC9942]) are the receipt-side twins: they define the Verifiable Data Structure formats that may appear in the unprotected headers of Transparent Statements whose binding layer is defined here. In-toto and DSSE represent an industry two-layer precedent: a content-addressed artifact layer combined with an attestation layer over the artifact's identifier. CPB formalizes the same pattern for the SCITT statement context. [I-D.hillier-scitt-arp] independently derives a similar canonical claim construction in its §2. Its Canonical Claim defines its own key-sort, NFC, number-rendering, and undefined-stripping rules, plus a Claim Hash join key. The construction is near-jcs but not byte- compatible. The independent re-derivation is evidence that this layer is consistently re-invented when it is not standardized; CPB exists to stop the re-invention. Implementations must not assume byte compatibility; ARP's Canonical Claim carries an explicit construction identifier by which a consumer can determine compatibility. [I-D.birkholz-verifiable-agent-conversations] defines trace-metadata conventions at the conversation grain (§7.4). The discovery mirror in Section 10 is designed to be compatible with that convention so that action-grain records and conversation-grain containers share one discovery layer. The alignment is informative; CPB does not normatively depend on that document. [I-D.sokolov-rats-aep-composition] addresses the complementary problem in the RATS domain: composing application-layer action evidence with remote attestation. [I-D.mih-sato-agent-accountability-composition] defines composition Mih & Sokolov Expires 25 February 2027 [Page 27] Internet-Draft Canonical Payload Binding August 2026 and conformance rules for multi-agent accountability chains. Together these documents demonstrate that the canonicalize-and- derive-identifier construction is a recurring primitive across independent use cases — one shared binding layer serving SCITT- anchored agent records, RATS attestation composition, and multi-agent accountability chains. 16. References 16.1. Normative References [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, . [RFC8126] Cotton, M., Leiba, B., and T. Narten, "Guidelines for Writing an IANA Considerations Section in RFCs", BCP 26, RFC 8126, DOI 10.17487/RFC8126, June 2017, . [RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017, . [RFC8259] Bray, T., Ed., "The JavaScript Object Notation (JSON) Data Interchange Format", STD 90, RFC 8259, DOI 10.17487/RFC8259, December 2017, . [RFC8785] Rundgren, A., Jordan, B., and S. Erdtman, "JSON Canonicalization Scheme (JCS)", RFC 8785, DOI 10.17487/RFC8785, June 2020, . [RFC9052] Schaad, J., "CBOR Object Signing and Encryption (COSE): Structures and Process", STD 96, RFC 9052, DOI 10.17487/RFC9052, August 2022, . [RFC9943] Birkholz, H., Delignat-Lavaud, A., Fournet, C., Deshpande, Y., and S. Lasker, "An Architecture for Trustworthy and Transparent Digital Supply Chains", RFC 9943, DOI 10.17487/RFC9943, June 2026, . 16.2. Informative References Mih & Sokolov Expires 25 February 2027 [Page 28] Internet-Draft Canonical Payload Binding August 2026 [I-D.birkholz-verifiable-agent-conversations] Birkholz, H., "Verifiable Agent Conversations", Work in Progress, Internet-Draft, draft-birkholz-verifiable-agent- conversations-00, n.d., . [I-D.hillier-scitt-arp] Hillier, J., "Attestation Reconciliation Protocol", Work in Progress, Internet-Draft, draft-hillier-scitt-arp-01, n.d., . [I-D.ietf-scitt-receipts-ccf-profile] Birkholz, H., Delignat-Lavaud, A., Fournet, C., and A. Chamayou, "CCF Profile for COSE Receipts", Work in Progress, Internet-Draft, draft-ietf-scitt-receipts-ccf- profile-04, 24 June 2026, . [I-D.lee-orprg-permit-receipts] Lee, Y., "Permit Receipts for Permit-Before-Commit Authorization of AI-Agent and Workload External Effects", Work in Progress, Internet-Draft, draft-lee-orprg-permit- receipts-00, n.d., . [I-D.mih-sato-agent-accountability-composition] Mih, S. and T. Sato, "Agent Accountability: Composition and Conformance", Work in Progress, Internet-Draft, draft- mih-sato-agent-accountability-composition-00, n.d., . [I-D.mih-scitt-agent-action-capsule] Mih, S., "An Agent Action Capsule Profile for SCITT", Work in Progress, Internet-Draft, draft-mih-scitt-agent-action- capsule-02, n.d., . [I-D.rampalli-pedigree] Rampalli, K., "PEDIGREE: Provenance and Delegation Records for Digital Artifacts", Work in Progress, Internet-Draft, draft-rampalli-pedigree-00, n.d., . Mih & Sokolov Expires 25 February 2027 [Page 29] Internet-Draft Canonical Payload Binding August 2026 [I-D.sokolov-rats-aep-composition] Sokolov, A., "Composing Application-Layer Action Evidence with Remote Attestation Procedures", Work in Progress, Internet-Draft, draft-sokolov-rats-aep-composition-03, n.d., . [RFC4998] Gondrom, T., Brandner, R., and U. Pordesch, "Evidence Record Syntax (ERS)", RFC 4998, DOI 10.17487/RFC4998, August 2007, . [RFC9901] Fett, D., Yasuda, K., and B. Campbell, "Selective Disclosure for JSON Web Tokens", RFC 9901, DOI 10.17487/RFC9901, November 2025, . [RFC9942] Steele, O., Birkholz, H., Delignat-Lavaud, A., and C. Fournet, "CBOR Object Signing and Encryption (COSE) Receipts", RFC 9942, DOI 10.17487/RFC9942, June 2026, . [RFC9995] Steele, O., Lasker, S., and H. Birkholz, "CBOR Object Signing and Encryption (COSE) Hash Envelope", RFC 9995, DOI 10.17487/RFC9995, July 2026, . Appendix A. Synthetic Registration Walkthrough This appendix illustrates the mechanics of Section 5, Section 6, and Section 7 using a non-domain-specific payload class. No domain vocabulary from any specific profile is used. *Payload class:* temperature-record. Fields: station_id (string), timestamp (string), celsius (exact decimal string), record_id (string). Exclusion set: {record_id}. Algorithm: jcs. Representation: bare 64-char lowercase hex. *Step 1 — Construct the payload:* { "station_id": "WS-42", "timestamp": "2026-07-24T00:00:00Z", "celsius": "21.3", "record_id": null } *Step 2 — Apply the exclusion set:* Mih & Sokolov Expires 25 February 2027 [Page 30] Internet-Draft Canonical Payload Binding August 2026 Remove record_id (it is in the exclusion set). The resulting object is: { "station_id": "WS-42", "timestamp": "2026-07-24T00:00:00Z", "celsius": "21.3" } *Step 3 — Compute the derived identifier:* Apply JCS [RFC8785] to produce the canonical octet string. Compute SHA-256 and encode as lowercase hex. The result is the record_id value to be placed back into the payload for transport. *Step 4 — Construct the Signed Statement:* Wrap the complete payload (including the now-populated record_id) in a COSE_Sign1 with: * content_type: application/temperature-record+json * alg and kid: producer's signing algorithm and key identifier *Step 5 — Register and receive a Receipt:* Submit the Signed Statement to a SCITT Transparency Service. Attach the returned Receipt to the unprotected header. The Transparent Statement is now suitable for distribution to verifiers. *Step 6 — Verify:* A verifier extracts the payload, strips record_id, applies JCS, recomputes SHA-256, and compares to the carried record_id. The verifier then verifies the envelope signature and, if present, the Receipt under a trusted service key. All three checks must pass for the record to be considered fully verified. Appendix B. Synthetic Two-Slot Composition This appendix illustrates Section 8 using two cooperating payload classes. No domain vocabulary is used. *Scenario:* a decision-record payload class cites an authorization- doc using a typed digest reference. *Authorization doc* (payload class authorization-doc; algorithm jcs): Mih & Sokolov Expires 25 February 2027 [Page 31] Internet-Draft Canonical Payload Binding August 2026 { "doc_id": "...", "subject": "WS-42", "scope": "temperature-write", "issued_at": "2026-07-24T00:00:00Z" } Its derived identifier is computed with doc_id in the exclusion set. Suppose the result is "ab12cd34...". *Decision record* (payload class decision-record; algorithm jcs): { "record_id": null, "action": "write", "authorization": { "type": "authorization-doc", "digest_alg": "SHA-256", "digest": "ab12cd34..." } } The typed reference authorization cites the authorization doc by its artifact type and derived identifier. A verifier can confirm the doc was cited by resolving the authorization-doc artifact type's digest context from its governing specification, recomputing "ab12cd34..." from the doc's bytes, and matching. *Composability:* the verifier needs only the authorization-doc digest context — it does not need to understand the decision-record format to verify the citation binding. For generic citation-binding verification, a CPB verifier can process a typed reference to any artifact type whose digest context it can resolve. Whether a particular citation slot permits that artifact type is determined by the consuming profile. Artifact-specific appraisal, authorization semantics, and application integration remain separate. Appendix C. Field-Verified Instances The instances in this appendix were chosen to illustrate the mechanisms of Section 4, Section 7, and Section 8. They are not a ranking. Two parties appear in every instance: the implementing system and the verification counterparty. The common counterparty in each case is the AAC reference implementation, which is present as a verifier, not as the subject. This is a historical record and is not edited retroactively: the instances below report what ran at the time, under algorithm jcs-n, which is withdrawn as of this revision (Section 4.2). The byte-agreement result each instance reports is a Mih & Sokolov Expires 25 February 2027 [Page 32] Internet-Draft Canonical Payload Binding August 2026 property of applying RFC 8785 JCS consistently, which jcs (Section 4.1) also provides going forward. *Owner consent status:* Anton Sokolov (Tyche Institute) — confirmed 2026-07-24. Tom Sato (GAR/SOOS) — confirmed 2026-07-25. Tymofii Pidlisnyi (Agent Passport System) — confirmed 2026-07-24 (on-issue). C.1. Deep Mechanism Instances C.1.1. Glyphzero Byte-Agreement — Algorithm Determinism Public record: Glyphzero PEDIGREE delegation record, IETF 126 hackathon. *What ran:* Two independently written RFC 8785 JCS implementations — Glyphzero's (Rampalli), used to produce its PEDIGREE delegation records [I-D.rampalli-pedigree], and the AAC reference implementation — computed a digest over the same delegation record and both produced subject_digest 0b4da06b... without any coordination on byte ordering beyond RFC 8785 itself. The record carried no null, empty-array or empty-object member, so the absent-field normalization pass jcs-n added to JCS did not apply to it; the agreement is an agreement about RFC 8785 JCS, which is the part jcs (Section 4.1) carries forward. *Mechanism illustrated:* Section 4.1. RFC 8785 JCS is reproducible across separately written implementations. The agreement was not premeditated; it emerged from two systems applying the same algorithm independently. This instance does not evidence an independent implementation of the withdrawn normalization pass, and the implementer census (Section 4.2) records that there was none. *Consent:* Karthik Rampalli (Glyphzero) confirmed 2026-07-25 (email, with corrections). C.1.2. GAR Session Block — Leaf Construction Rule Public record: GAR Session Block anchor, IETF 126 hackathon; gar- core.ts commit fe18f24; CT leaf 166. *What ran:* A GAR Session Block record was registered in a SCITT Transparency Service (RFC9162_SHA256 VDS). The log leaf was constructed as SHA-256 of the raw bytes of the derived identifier — bytes.fromhex(id), not id.encode("utf-8"). The inclusion proof verified correctly against the anchored Merkle root only when the leaf used the raw bytes. Mih & Sokolov Expires 25 February 2027 [Page 33] Internet-Draft Canonical Payload Binding August 2026 *Mechanism illustrated:* Section 7.1. The leaf-bytes-not-hex rule was discovered during live anchoring when a leaf constructed from the hex string failed to verify; switching to raw bytes produced the correct root. *Consent:* Tom Sato (GAR/SOOS) — confirmed 2026-07-25. C.1.3. A2A Boundary Seal — Derived Identifier as Protocol Gate Public record: capsule-emit issue #29, verified offline at https://github.com/action-state-group/capsule-emit/issues/29. *What ran:* An A2A-protocol boundary producer submitted a record to a SCITT Transparency Service and used the derived identifier as a protocol-layer gate (capsule.digest / capsule.resolve). The receipt was verified offline using a conforming SCITT verifier (scitt-cose verify_receipt → ok=True), and the Merkle inclusion proof (verify_inclusion) folded to the anchored root. A DENY negative case was also demonstrated: a fabricated derived identifier not present in the log returned 404 on the resolve step and DENY on the gate. *Classification (exact):* single-machine loopback rehearsal, independently reproduced. The read-only resolve path (/anchor/ inclusion-proof-ct) is live at anchor.agentactioncapsule.org; a networked cross-machine close is pending counterparty schedule. *Mechanism illustrated:* Section 5 and Section 7 applied at a protocol boundary: the derived identifier is stable across network hops and usable as a verifiable join key without payload disclosure. *Consent:* Anton Sokolov (Tyche Institute) — confirmed 2026-07-24. C.2. Field Table — IETF 126 Participants The following table lists all parties that ran verifiable instances at the IETF 126 hackathon. Rows appear in alphabetical order by party name; the order carries no ranking. Mih & Sokolov Expires 25 February 2027 [Page 34] Internet-Draft Canonical Payload Binding August 2026 +===========+=============+=================+=====================+ |Party |Record type |What ran | Public record | +===========+=============+=================+=====================+ |Agent |Decision |Content-derived | draft-pidlisnyi-aps | |Passport |record |action reference;| + hackathon | |System | |NFC + code-point | coordinates | |(Pidlisnyi)| |sort + JCS; | | | | |bidirectional | | | | |cross-runs 6/6 + | | | | |24/24 | | +-----------+-------------+-----------------+---------------------+ |EP |Named-human |Three independent| EMILIA/EP hackathon | |(Schrock) |approval |codebases | record | | | |produced | | | | |8cf0c36e...; | | | | |three-computation| | | | |single-digest | | +-----------+-------------+-----------------+---------------------+ |GAR (Sato) |Kernel |Sealed as record;| gar-core.ts commit | | |session block|CT leaf = SHA- | fe18f24 | | | |256(raw bytes of | | | | |id); leaf 166 | | | | |verified | | +-----------+-------------+-----------------+---------------------+ |Glyphzero |Delegation |Two independent | Glyphzero PEDIGREE | |(Rampalli) |record |JCS | hackathon record | | | |implementations; | | | | |subject_digest | | | | |0b4da06b... | | +-----------+-------------+-----------------+---------------------+ |Microsoft |Two-TS |One payload, two | scitt-ccf-ledger PR | |(Chamayou) |statement |receipt profiles | #424 | | | |(ccf.v1 + | | | | |RFC9162_SHA256) | | | | |in conjunction | | +-----------+-------------+-----------------+---------------------+ |Sokolov |Boundary-seal|A2A gate; | capsule-emit issue | |(Tyche) | |derived-id as | #29 | | | |resolve key; DENY| | | | |negative; offline| | | | |Receipt verify | | +-----------+-------------+-----------------+---------------------+ Table 4 Mih & Sokolov Expires 25 February 2027 [Page 35] Internet-Draft Canonical Payload Binding August 2026 C.3. Agreed and Scheduled The following cross-verifications are agreed and scheduled but have not produced field-verified instances at time of writing: * VTO/libp2p (M.S. Gupta) — content-addressed telemetry objects citing action records across grains. * VSO/VeritasChain (Kamimura) — verifiable service objects under jcs. Field-verified instances are expected to be added in future revisions as cross-verifications complete. The PermitReceipt × MachineMandate composition is excluded from this appendix. It is recorded in the AAC interop registry (INTEROP.md). Acknowledgments The following individuals contributed findings from the IETF 126 hackathon in Vienna that directly shaped the rules in this document. All attributions cite public artifacts. *Contributors* [all named attributions and contributor acknowledgments individually confirmed: Anton Sokolov (confirmed 2026-07-24), Iman Schrock (confirmed 2026-07-24), Tom Sato (confirmed 2026-07-25), Yong Bok Lee (Scott Lee) (contributor attribution confirmed 2026-07-27), Tymofii Pidlisnyi (Agent Passport System, confirmed 2026-07-24, on-issue), Karthik Rampalli (Glyphzero, confirmed 2026-07-25, email, with corrections)]: * Anton Sokolov (Tyche Institute) — assurance-boundary discipline; the A2A boundary-seal instance in Appendix C. * Yong Bok Lee (Scott Lee), Meridian Verity Group — ORPRG-derived cross-profile digest-context discipline: equal-looking digest text alone is not a valid join; a typed reference is verified by recomputing the referenced artifact under its established digest context and comparing that result with the digest carried in the reference, not with the citing record's own derived identifier. Also contributed the representation-boundary distinction among raw digest bytes, bare lowercase hexadecimal text, and prefixed text, and the verification-scope boundary separating typed-reference content binding from artifact-specific appraisal and authorization. See [I-D.lee-orprg-permit-receipts]. Mih & Sokolov Expires 25 February 2027 [Page 36] Internet-Draft Canonical Payload Binding August 2026 * Tymofii Pidlisnyi (Agent Passport System) — the content-derived action reference pattern (NFC + code-point sort + JCS) demonstrating that jcs-n generalizes across canonicalization styles; bidirectional cross-runs with confirmed byte-agreement. * Tom Sato (GAR/SOOS) — the leaf-bytes-not-hex finding documented in Section 7.1: the log leaf hashes the raw bytes of the derived identifier, not the hex-string encoding. * Karthik Rampalli (Glyphzero) — independent JCS implementation byte-agreement on subject_digest 0b4da06b..., demonstrating that RFC 8785 JCS is reproducible across separately written implementations. * Iman Schrock (EMILIA/EP) — confirmed 2026-07-24 — the three- computation single-digest instance (8cf0c36e...) demonstrating byte-agreement across three independent codebases. *Acknowledged* [Amaury Chamayou confirmed 2026-07-24 (email)]: * Amaury Chamayou (Microsoft) — two-TS single-statement demonstration; the vds-from-protected-header finding subsequently mirrored in microsoft/scitt-ccf-ledger #424. Authors' Addresses Steven Mih Action State Group, Inc. Email: spec@actionstate.ai Anton Sokolov Tyche Institute Email: anton.sokolov@tyche.institute Mih & Sokolov Expires 25 February 2027 [Page 37]