Internet-Draft AUDIT Decision Receipts November 2026
Gruszka Expires 19 May 2027 [Page]
Workgroup:
Network Working Group
Internet-Draft:
draft-gruszka-audit-decision-receipts-09
Published:
Intended Status:
Informational
Expires:
Author:
K. Gruszka
b7n0de

Decision Receipts for Approvals, Declined Actions and Executed Actions in Agent Audit Records

Abstract

The architecture for auditing agent delegation and interactions proposed for the AUDIT effort names Interaction Records, Action Records and Authorization Transition Records and leaves their encoding open. This document maps an existing pair of signed records, the decision receipt and the outcome receipt of the open-source proofbundle project, onto those record types for three cases, a human approval, a declined action and an executed action with its result, and gives each record exact bytes, a signature and digest bindings that a third party can check offline. Each receipt is created when the decision or the outcome occurs and is checked by an Auditor post hoc, and a declined or refused action is a record of its own. For each case the document states what the receipt adds and what it does not show, and it gives a verification procedure that runs offline. It requires that an approval be honoured at most once, even after the record of consumed approvals is rotated. Test vectors for all three cases, each with a counterexample, are included. The document also describes, informatively, how a receipt can be registered with a Transparency Service under RFC 9943 and what a SCITT Receipt then shows and does not show. It uses existing receipt formats and defines no new predicate or registry. How the issuer of a receipt enforces its decision is out of scope.

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 19 May 2027.

▲

Table of Contents

1. Introduction

The architecture of [AUDIT-ARCH] asks that auditing link user intent, delegation, authorization and execution, and treats authorization as a state that changes over time. Three of its four record types (its Section 5) matter here. Interaction Records capture user-facing events such as approvals and refusals, Action Records capture events at the boundary where an action took effect, and Authorization Transition Records capture changes in permission state. Its field names are placeholders (its Section 7), and a canonical data model is a work item (WI-1).

An Auditor in that architecture does not trust the Agent (its Section 8), so a record helps only when a third party can check it. That needs exact bytes, a signature over them and bindings to inputs and evidence by digest. This document shows that the decision receipt of proofbundle [PB-DECISION], an in-toto Statement [INTOTO-STATEMENT] in a DSSE envelope [DSSE], signed with Ed25519 [RFC8032] over a payload in the serialization of [RFC8785], supplies them together with its companion outcome receipt [PB-OUTCOME]. It does so for three cases that Agent Guard, the gate of the proofbundle plugin for two coding agent hosts, is built for, within the limits of its inspected code that Section 5 states:

  1. a human approval of one action (an Interaction Record of an approval and an Authorization Transition Record with the trigger user_approval);
  2. a declined action, with its reason and without an effect;
  3. an executed action with its result (an Action Record).

The document maps and verifies. It does not propose the decision receipt as the data model of the architecture and does not change the receipts it describes. The mapping of Section 2 follows revision -01 of [AUDIT-ARCH]; a later revision of the architecture can change it.

A decision receipt is an audit record (Section 3.1 of [AUDIT-ARCH]) that is created when the decision is taken, before the action it decides takes effect, and an outcome receipt is created after the action. An Auditor (Section 4.3 of [AUDIT-ARCH]) checks both post hoc, as records, with the procedure of Section 3, which is the post‑hoc use of records that Section 1 of [AUDIT-ARCH] describes. A declined or refused action is a record of its own, a decision receipt with the verdict DENY or REFUSE (Case 2 of Section 3.3). How the issuer of a receipt enforces its decision is not a subject of this document. Section 4 applies to a relying party when it honours an approval, not to an Auditor.

1.1. Requirements Language

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.

In this document, these key words apply to a verifier in Section 2.6 and in the procedure of Section 3, to an Auditor that performs the checks of Section 3.5, to a relying party that lets an action run because of an approval in Section 4, to whoever issues approval records there, and to the use of the test keys of Appendix A.

1.2. Terms

Decision receipt, outcome receipt:
A DSSE envelope whose payload is an in-toto Statement v1 with the predicate type https://b7n0de.com/proofbundle/predicates/decision-receipt/v0.1 or https://b7n0de.com/proofbundle/predicates/action-outcome/v0.1.
Payload bytes:
The bytes obtained by base64 decoding the payload member of the envelope. A DSSE signature covers the pre-authentication encoding of payloadType and these bytes.
Content root:
SHA-256 [RFC6234] over the payload bytes. Other digests commit to the byte strings identified by their respective fields. Two receipts with the same content root have the same payload bytes, assuming the collision resistance of SHA-256; their envelopes, signatures and keys can differ.
Approval record:
The record of one human act that approves one action (Section 2.1).
Relying party:
The component that lets an action run because of a decision.
Consumption state:
The durable record in which a relying party notes each approval it has honoured.
Consumption scope:
The set of decisions of a relying party within which it honours each approval at most once (Section 4), and within which the first check of Section 3.5 counts released ALLOW decisions.
Decision maker, outcome signer:
The holder of the key that signs a decision receipt, named by decisionMaker.id, and the holder of the key that signs an outcome receipt. The outcome signer is the executor or an observer (Section 3.1).
Trust policy:
The configuration of a verifier that says which key it accepts for which decisionMaker.id and for which period and which key may attest which executor.id, and that may name the consumption scope of an Auditor (Section 3.5). This document does not define its format.
Verifier, Auditor:
A verifier runs the procedure of Section 3.2 to Section 3.4 on a receipt. An Auditor is a verifier in the role of Section 4.3 of [AUDIT-ARCH] that can also run the checks of Section 3.5 over the receipts it holds.

2. Mapping the Three Cases onto AUDIT Records

All decision receipts here use the decisionType preActionAuthorization and carry validity with audience, nonce and expiresAt. A decision receipt is the gate's signed view of an event; the human act itself is the approval record, bound by digest. In the tables, IR stands for Interaction Record, ATR for Authorization Transition Record and AR for Action Record, followed by a placeholder field of the architecture. The fields common to all cases map as follows.

event_id
decisionId (outcomeId in an outcome receipt)
trace_id, parent_id
traceContext.traceId, traceContext.parentId
timestamp
decidedAt (performedAt in an outcome receipt)
actor, on_behalf_of
agent.id, principal.id
action
proposedAction (actionType, target, method)
exact action bytes
proposedAction.parametersDigest, where present
authorization.expires_at
validity.expiresAt
delegation_chain
none (delegationRefs is not used here)
record signer
decisionMaker.id, bound to the key by a trust policy

Every decision receipt thus names, in principal.id, the user or other party on whose behalf the agent acted, which step 7 of Section 3.2 requires to be a non-empty string; a delegation through further agents is not recorded here (delegationRefs is not used).

2.1. Case 1, a Human Approval

A gate asks a person before an action and records the answer. An earlier receipt for the same request, such as one with the verdict ESCALATE, is a separate record; this receipt does not reference it. After the answer the gate decides ALLOW for exactly the approved action. The approval record carries approvalId, approver, session, actionDigest (SHA-256 of the approved action bytes), approvedAt and validUntil. The decision receipt names it in inputSnapshot under the name "approval record".

Table 1: Case 1
Record, field Decision receipt
IR, approval approval record, by its digest in inputSnapshot
IR, actor approver in the approval record, equal to principal.id
ATR, previous_state not recorded in this receipt; preceding state must be established separately
ATR, new_state verdict ALLOW for parametersDigest
ATR, trigger reasonCodes contains user_approval
ATR, actor approver in the approval record

The scope of the new state is one action, identified by its bytes. A second action, even one with the same name and target, needs a new approval.

2.2. Case 2, a Declined Action

A gate declines an action because a rule or a person says no. The receipt records the verdict DENY or REFUSE with its reasons. It may say in actionOutcome that the call was blocked, which is the decision maker's own statement. This document treats DENY and REFUSE alike and does not define a difference between them.

Table 2: Case 2
Record, field Decision receipt
IR, refusal verdict DENY or REFUSE
reason reasonCodes
action proposedAction with parametersDigest, where present
no effect actionOutcome.status blocked (self-asserted); no outcome with status executed names this decision in the supplied set
ATR none; the state is unchanged

A receipt cannot prove that something did not happen. An observer at the boundary can separately report that it observed no effect. The check in Section 3.5 detects a conflicting executed outcome in the supplied set; it does not establish absence of an effect.

2.3. Case 3, an Executed Action and Its Result

A gate allows an action, the action runs, and an executor or an observer at the boundary records what happened in an outcome receipt signed with its own key. The outcome receipt names the decision by its content root in decisionRef.

Table 3: Case 3
Record, field Outcome and decision receipt
AR, event_id outcomeId
AR, parent_id traceContext.parentId of the referenced decision
decision binding decisionRef, the content root of the decision
AR, timestamp performedAt
AR, actor executor.id
AR, action requestedActionDigest, equal to parametersDigest
result status executed, effectDigest
authorization the decision (ALLOW, validity.expiresAt)
evidence evidenceRefs of the decision

The architecture asks that Action Records be produced at the boundary where the action took effect. Its WI-1-3 distinguishes the Recorder's signing identity from the recorded Agent's identity where they are operationally separate. Section 3.4 checks that the outcome and decision use different keys and that executor.id differs from decisionMaker.id. Those checks alone do not establish operational separation or that the outcome signer observed the boundary.

2.4. What the Receipt Adds

  • Exact bytes. The payload is its own serialization under [RFC8785]. A verifier uses the received payload bytes for the signature input and content root. Canonical serialization is used to check equality and to derive the predicate's subject digest, never to repair or replace the received payload.
  • A signature. Ed25519 over the DSSE pre-authentication encoding of the payload type and the payload bytes.
  • Binding by digest. The action bytes (parametersDigest, where present), the approval record and other inputs (inputSnapshot), the policy (policyBoundary.policyDigest), evidence (evidenceRefs), the decision itself (decisionRef in the outcome) and the observed effect (effectDigest) are named by SHA-256.
  • Stated limits. The member notChecked lists what the decision maker did not check, and decisionChangeConditions lists what would change the decision.

A record made after the interaction, such as an outcome receipt, a later countersignature or a settlement, can name the decision by its content root. The content root is the same for every copy of the payload and does not change for as long as the payload bytes are kept; a receipt that supersedes it has its own content root and names it in relationships (Section 3.8). decisionId and validity.nonce serve the exchange in which the decision is made, and this document does not require them to be unique beyond it.

2.5. What the Receipt Does Not Do

  • It does not show that a decision was correct, safe or lawful.
  • It does not authenticate the approval. The approval record in this document is not signed by the person it names. A deployment can additionally cite a separately signed approval in inputSnapshot. Validating that signature and establishing its correspondence to the approval record require rules beyond this document. The signed approvals of [IZMAYLOV] are one possible source of such evidence. The proof token of [YOSSIF] is another, a JWS signed by a hardware-protected key that a user-verification event gates for each signature, bound to one named action (Sections 3, 3.7 and 3.13 of [YOSSIF]), which shows that a verified human was present, not which person (Section 6.8.3 of [YOSSIF]). Such a token carries its own signature for a verifier of that format; an entry of inputSnapshot that names it by digest does not carry its signer (Section 3.1). A record of an approval that an authorization server signs has the same limit: its signature does not show that the approver actually approved (Section 9.4 of [LIU]). In cases N17, N18 and N22 of [DPS-RAR], with a test key for the authorization server, such a record passes the checks of Section 7 of [LIU] for an approval the person never gave, for a text other than the one the person was shown, and for a decision recorded before the person signed; only the binding to the person's own signature catches them.
  • It does not show that an effect happened. An outcome receipt shows that its signer stated an effect with a given digest. A digest that cannot be resolved to bytes shows only that a value was written.
  • It does not register anything. Transparency in the sense of [RFC9943] needs a separate registration and a SCITT Receipt; Section 3.7 describes how a receipt of this document can be registered.
  • It does not detect a receipt that was never written. An added observer or a sequence counter can reveal some such receipts, not all (Section 7).
  • It does not show that no action ran while an ALLOW was usable. An ALLOW that expires unused leaves no outcome receipt, and the receipts of this document do not establish an absence over its validity or over any other time window (compare Section 2.2).
  • Its times are the signer's clock.
  • It is not an access token or an authorization grant (Section 6). A verified ALLOW is a record of a decision, and the executing system makes its own authorization check. A pass shows neither that the decision maker or the agent was authorized at the time of the action nor the state of a delegation at that time.

2.6. Evaluation Receipts Are Evidence, Never Authorization

A decision may rest on an evaluation result, such as a test suite that passed. A Signed Evaluation Receipt [SER] records such a result. It appears in a decision receipt only as an entry of evidenceRefs whose digest is the receipt digest, SHA-256 over its payload bytes, as in [SER-MAPPINGS]. A verifier MUST NOT treat a valid Signed Evaluation Receipt, or its presence in evidenceRefs, as permission for an action. In the records of this document only a decision receipt with the verdict ALLOW, from a key the relying party trusts for decisions, records a permission.

2.7. Observations for the Architecture

This section is informative. At the places below, the mapping of this section needed a reading of revision -01 of [AUDIT-ARCH]. They are offered as observations and questions, not as requirements.

  • Evidence for a human approval. Section 3 of [AUDIT-ARCH] associates Interaction Records with the User; Section 3.1 says that user-facing systems produce records of prompts and approvals, and Section 4.2 has the Auditing Service sign records under its own identity. In Case 1, the gate's decision receipt binds an approval record that names a person but is not signed by that person (Section 2.1 and Section 2.5). What evidence is needed to bind such a record to that person's act of approval, and when is the person's own signature required? [YOSSIF] offers one form of such evidence, signed by a device key that user verification gates rather than by the person (Sections 1.2 and 3.7 of [YOSSIF]).
  • Previous state. The Authorization Transition Record of Section 7.4 of [AUDIT-ARCH], and its work item WI-1-4 (Section 6 of [AUDIT-ARCH]), carries a previous state, a new state, a trigger and an actor in one record. A decision receipt records only the new state for one action and leaves the previous state to other records (Section 2.1). Section 5 of [AUDIT-ARCH] reconstructs authorization from the ordered sequence of such records; is a record without the previous state enough for that?
  • Binding an Action Record to its authorization. In Section 7.2 of [AUDIT-ARCH] the parent_id of an Action Record references the Interaction Record that motivated the action, and the authorization appears as scope and expiry. The outcome receipt of Case 3 names the decision receipt by its content root in decisionRef and takes parent_id from that decision (Section 2.3). The examples of [AUDIT-ARCH] have no member that binds an Action Record to the authorizing decision by digest.
  • Refusals before execution. Section 5 of [AUDIT-ARCH] includes model responses and agent-to-agent dialogue among Interaction Records. Its work item WI-1-2 (Section 6 of [AUDIT-ARCH]) also includes system events and does not preclude profiles for non-conversational interactions. In Case 2, a gate can decline a proposed action under a rule before execution (Section 2.2). When such a denial is internal to a workflow and is not part of a conversation, is it an Interaction Record, an Action Record of an attempted request, or a different event type?
  • Who records an action. Section 3 of [AUDIT-ARCH] has Action Records generated by the Agent, while Section 5 of [AUDIT-ARCH] places them at the boundary where the action took effect. Its work item WI-1-3 (Section 6 of [AUDIT-ARCH]) already distinguishes the Recorder's signing identity from the recorded Agent's identity where they are operationally separated. In Case 3 the outcome signer can be an observer that attests the executor's identifier (Section 3.1 and Section 3.4). How should an Action Record and the Auditor's trust policy identify the Recorder separately from the actor whose action is recorded?

3. Offline Verification

The procedure below needs no network access. Each step either passes or ends the procedure with a failure named by the step. A later step never repairs an earlier failure. Where several checks of one step fail, the failure is that of the step; which or how many of the failing members a verifier reports is not part of the result. A requirement that more than one step checks fails at the first of those steps. Wherever this section compares two times, it compares their full value, as if each fraction of a second were padded with zeros to six digits.

3.1. Inputs

The verifier needs the envelope bytes of each receipt and the Ed25519 public keys selected by its trust policy for the decision signer and, for Case 3, the outcome signer. It also needs the expected decisionMaker.id and, for Case 3, the expected executor.id whose assertion that policy permits under the outcome-signing key. An outcome signer may be an observer rather than the executor.

The verifier supplies an evaluation time T in UTC, in the form of the timestamps of Section 3.8; it does not take T from a receipt. It also chooses which case of Section 3.3 it checks; it does not take the case from the receipt. Expected audience and nonce and the action and effect byte strings are optional inputs. Case 1 requires the approval record bytes. Inputs or evidence whose bytes are not checked remain digest references, not verified contents. The members inputSnapshot and evidenceRefs name digests and relations, not the issuer or operator of the evidence they name, so a verifier cannot tell from them who issued each item; separate keys do not show independent operators either (step 5 of Section 3.4).

An Auditor that checks a receipt post hoc sets T to the time that its audit asks about and takes that time from its audit context, for example from a record made at the boundary where the action took effect. With a later T, such as the current time of the Auditor, step 9 of Section 3.2 fails for every receipt whose validity has ended. That failure says that the receipt is not usable at T. It does not say that the receipt was invalid when it was issued. A relying party that lets an action run uses its own current time as T (Section 4).

3.2. Decision Receipt

  1. Envelope. The envelope MUST be valid UTF-8 and parse as one JSON text of [RFC8259] that is an object, with no duplicate member names in any object. The envelope bytes MUST NOT begin with the byte order mark U+FEFF, which Section 8.1 of [RFC8259] forbids adding. It MUST contain payload, payloadType and signatures. payloadType MUST be application/vnd.in-toto+json. An envelope fails this step where signatures is not an array or where an entry of signatures is not an object whose member sig is a string.
  2. Payload bytes. Decode payload as base64 [RFC4648] into the payload bytes P. Here and for sig in step 3, base64 has only the characters of the alphabet of Section 4 of [RFC4648] and its padding, so no line breaks, and pad bits of zero (Section 3.5 of [RFC4648]); a payload in another form fails this step. Padding is required, so the base64 text of payload and of sig has a length that is a multiple of four. A sig in another form is not decoded; step 3 counts it as a signature that does not verify, and it does not fail this step.
  3. Signature. Form PAE = "DSSEv1" SP LEN(type) SP type SP LEN(P) SP P, where SP is one space, LEN is the length in bytes as ASCII decimal without leading zeros, and type is the payloadType. At least one entry of signatures MUST carry a base64 sig that verifies as an Ed25519 signature over PAE under the expected key. The keyid member is not used to select a key. The sig of other entries of signatures is not verified. A sig verifies where its S, read as a little-endian integer, is less than the order L of Section 5.1 of [RFC8032] and the group equation of Section 5.1.7 of [RFC8032] holds in its form without the cofactor 8; a sig whose S is not less than L MUST NOT verify.
  4. Canonical form. P MUST be valid UTF-8 and parse as one JSON value with no duplicate member names. The serialization of the parsed value under [RFC8785] MUST be byte for byte equal to P. A verifier MUST NOT continue with a re-serialized form. Numbers are IEEE 754 double precision values under Section 3.1 of [RFC8785], so a P with the number 9007199254740993, which becomes 9007199254740992, fails this step, as does a P with an unpaired surrogate (Section 3.2.2.2 of [RFC8785]). P MUST NOT contain, anywhere in its parsed value and at any depth, a number whose absolute value is greater than 9007199254740991 and that is written without the frac and exp parts of the grammar in Section 6 of [RFC8259], even where P equals its serialization, because under Section 2.2 of [RFC7493] a receiver cannot be expected to treat such an integer as an exact value. A number written with frac or exp, such as 1e+21, is not covered by that sentence; for such a number this step asks only that P equal its serialization, which a P with 1e+21 does.
  5. Content root. Compute R = SHA-256(P).
  6. Statement. _type MUST be https://in-toto.io/Statement/v1 and predicateType the decision receipt type of Section 1.2. subject MUST contain exactly one entry, whose digest.sha256 is SHA-256 over the [RFC8785] serialization of predicate; it MUST NOT contain other entries. Other members of the Statement, of the subject entry and of its digest are not checked. A Statement without predicate fails this step. A Statement whose predicate is not a JSON object fails this step.
  7. Members. predicate MUST be a JSON object satisfying the field rules of Section 3.8.
  8. Signer. decisionMaker.id MUST be an identifier that the trust policy binds to the key of step 3, and it MUST equal the expected decisionMaker.id of Section 3.1.
  9. Validity. decidedAt MUST NOT be later than T, and T MUST be strictly earlier than expiresAt. Where an expected audience is given, audience MUST contain it; where an expected nonce is given, nonce MUST equal it.
  10. Action bytes. Where the action bytes are given, their SHA-256 MUST equal proposedAction.parametersDigest.sha256.

3.3. Checks per Case

The checks of a case follow step 10 of Section 3.2.

Case 1, approval.

decision.verdict MUST be ALLOW and decision.reasonCodes MUST contain user_approval. inputSnapshot MUST contain exactly one entry whose name is "approval record"; it may contain other entries. The approval record bytes A MUST be supplied, MUST hash to that entry's digest, and MUST be their own [RFC8785] serialization.

A MUST be a JSON object containing approvalId, approver, session, actionDigest, approvedAt and validUntil. Other members of A are not checked. approvalId, approver and session MUST be non-empty strings. actionDigest MUST have the digest form of Section 3.8 and MUST equal proposedAction.parametersDigest. approver MUST equal principal.id. approvedAt and validUntil MUST be [RFC3339] UTC timestamps with a trailing "Z" in the form given in Section 3.8. The bounds approvedAt <= decidedAt <= T < validity.expiresAt <= validUntil MUST hold.

Case 2, declined.
decision.verdict MUST be DENY or REFUSE. Where actionOutcome is present, its status MUST be notAttempted, blocked or refused, and performedAt and outcomeRef MUST be absent or null. An actionOutcome without status fails this case.
Case 3, executed.
The checks run in this order. First, decision.verdict MUST be ALLOW. Second, the outcome receipt MUST pass Section 3.4 against this decision, and an outcome receipt that fails one of its steps fails there. Third, the status of the outcome receipt MUST be executed.

3.4. Outcome Receipt

  1. Apply steps 1 to 5 of Section 3.2 separately to the outcome envelope with the expected outcome-signing key, obtaining payload bytes Q. A failure there is a failure of this step. Retain the decision receipt's content root R for the binding check below.
  2. _type MUST be https://in-toto.io/Statement/v1 and predicateType the outcome receipt type. subject MUST contain exactly one entry, whose digest.sha256 is SHA-256 over the [RFC8785] serialization of predicate; it MUST NOT contain other entries, as in step 6 of Section 3.2. A Statement whose predicate is not a JSON object fails this step.
  3. Members. predicate MUST be a JSON object satisfying the field rules of Section 3.8.
  4. Binding. decisionRef.sha256 MUST equal the content root R of the decision receipt that passed Section 3.2, and requestedActionDigest MUST equal its parametersDigest.
  5. Identity and separation. executor.id MUST equal the expected executor identifier supplied under Section 3.1. The verifier's trust policy MUST permit the outcome-signing key to attest that identifier. That key MUST differ from the decision maker's key, and executor.id MUST differ from decisionMaker.id. These checks do not establish that the two keys are controlled by operationally independent parties. The outcome-signing key is the key under which step 1 verified the outcome envelope. Where the trust policy lists the key of the decision maker for outcome receipts too and step 1 verified the envelope under that key, this step fails.
  6. Result. status executed MUST come with a non-null effectDigest. Where effect bytes are supplied, effectDigest MUST be present, whatever the status, and their SHA-256 MUST equal effectDigest.sha256.
  7. Time. performedAt MUST NOT be earlier than the decision's decidedAt and MUST be strictly earlier than its validity.expiresAt. No step of this section compares performedAt or another member with T.

3.5. Checks over a Set of Receipts

An Auditor that holds the receipts and approval records of a run can check three properties that no single receipt shows. These checks are only as complete as the set of receipts the Auditor holds.

The Auditor supplies the expected signing key for each held receipt under its trust policy and checks every supplied held envelope before the three checks below. A held decision receipt must pass steps 1 to 6 of Section 3.2 under its expected signing key. A held outcome receipt must pass steps 1 to 5 of Section 3.2 under its expected signing key and step 2 of Section 3.4. In addition, the predicate of every supplied held receipt must be a JSON object. Where relationships is present in that predicate, the member and its entries must satisfy the field rules of Section 3.8 for relationships, including the definitions of the types and the notes that those rules reference. These additional requirements are part of the preliminary checks, and failure of either makes the input invalid.

The first and second checks count the receipts that pass these preliminary checks under their expected signing keys; a receipt that has expired at the evaluation time T is not left out of either check for that reason.

The Auditor MUST obtain the relying party's consumption scope from its audit context, or where that names none, from its trust policy; that scope is not inferred from an approvalId alone. First, within that scope, no two distinct released ALLOW decision payloads, identified by different content roots, may cite approval records with the same identity (approvalId, actionDigest and session). Such a pair violates item 1 of Section 4, for an approval identified as in item 2; additional copies or signatures of one payload do not. An ALLOW is released as Section 4 defines it; one that the Auditor can show never left the control of the relying party does not count. The first check does not detect two executions that reuse one ALLOW receipt without producing another decision receipt.

Second, no outcome receipt with status executed may name, in decisionRef, the content root of a receipt whose verdict is not ALLOW. Case 3 of Section 3.3 requires ALLOW for an executed action, so an execution that names a decision with DENY, REFUSE, ESCALATE, DEFER or OBSERVE is the same conflict.

Third, an entry of relationships of a decision or outcome receipt whose relation is supersedes names the predecessor of its receipt, the receipt that the Auditor holds whose content root is the digest of targetReceiptDigest of that entry. A payload is its own [RFC8785] serialization (step 4 of Section 3.2), so that content root is the jcs-sha256-v1 digest of the Statement of the predecessor. The Auditor takes the predecessor as the state before the receipt that its issuer declared, not as a statement that this state is correct (Section 3.8), and derives earlier states from the predecessor of each predecessor in turn. An entry fails this check where the Auditor holds no receipt with that content root, or where the entry has targetSubjectDigest and its digest is not the digest.sha256 of the subject entry of the predecessor; no earlier state is derived from such an entry. Entries with another relation are not part of this check, and no earlier state is derived from them. Where a receipt has more than one entry with relation supersedes, each entry is checked on its own and names one predecessor, and the Auditor derives earlier states from each of them.

For this third check, the lookup of predecessors follows only the entries with relation supersedes of the receipt being verified and, in turn, of each predecessor found; a held receipt that this chain does not reach is subject only to the preliminary checks, and its own entries are not looked up. Either kind of held receipt may be a predecessor of either kind of receipt.

Apart from these preliminary checks, a held receipt used as a predecessor need not pass the remaining checks of Sections 3.2 to 3.4. Failure of one of those remaining checks is not, by itself, a failure of this third check. In particular, a historical predecessor is not rejected by this check merely because it has expired at the evaluation time T or does not pass a case of Section 3.3. This does not waive any check required for a receipt that is also being verified in its own right. An entry of a predecessor, at any depth of the chain, is checked like an entry of the receipt, and where it fails, this third check fails for the receipt.

If any supplied held envelope fails the preliminary checks above, the input is invalid and the Auditor reports no result for this third check. This applies also to supplied held envelopes that no supersedes entry references. An invalid input is distinct from a supersedes entry that fails because its predecessor is absent from a valid set of held receipts. Passing the preliminary checks does not establish that a predecessor's declared state is correct. The third check does not require that a predecessor concerns the same action as its receipt; it matches digests only.

3.6. What a Pass Shows

A pass shows that the holder of the decision maker's key signed exactly these payload bytes. Where action bytes are supplied under Section 3.1, their SHA-256 matches proposedAction.parametersDigest.sha256. In Case 1, the SHA-256 of the required approval record bytes matches the digest.sha256 value of the unique inputSnapshot entry whose name is "approval record". For Case 3, a pass also shows that the holder of a different key, which the trust policy permits to attest the expected executor.id, signed an outcome with status executed bound to that decision. It does not show that the signer's statements are true, that the approval came from the named person, that the outcome signer observed the boundary, that no other receipt exists, or that the keys were not misused. Nor does it show that the decision maker held its authority, or a delegation that delegationRefs names, at the time of the action; a pass binds only what the steps of Section 3 check. A decision maker that did not check its own authority or a delegation at the time of the action can say so in notChecked (Section 3.2, step 7); such an entry is its own statement and not a result of the steps of Section 3. Those questions belong to the trust policy, to signed approvals, to observers and to transparency registration (Section 3.7).

3.7. Registering a Receipt with a Transparency Service

This section is informative. It describes how the receipts of this document can be registered with a Transparency Service under [RFC9943]. It defines no format, predicate type, media type or registry and adds no requirement to those of [RFC9943].

Registration. In SCITT a Signed Statement is a COSE_Sign1 message whose protected header carries CWT Claims with an Issuer Claim and a Subject Claim (Section 6 of [RFC9943]). A decision receipt or an outcome receipt can be registered in two ways. Its envelope bytes can be the payload of a Signed Statement. Or, as Section 6.2 of [RFC9943] allows for Statements too large or too sensitive to send to a Transparency Service, the Signed Statement can be made over a hash, here the content root of the receipt. This document does not define how a Signed Statement indicates the hash, its algorithm or its content type; that belongs to a deployment's profile and to the Registration Policy of the Transparency Service (Section 5.1.1 of [RFC9943]).

The test package of Appendix A.5 gives one example Signed Statement for each of the two ways, with a SCITT Receipt for each.

Issuer. The Issuer of the Signed Statement is the entity identified by its iss Claim, whose key signs the COSE_Sign1 (Sections 3 and 6 of [RFC9943]). It can be the decision maker or the outcome signer, or another party such as the Auditing Service of Section 4.2 of [AUDIT-ARCH]. The Client that registers a Signed Statement need not be its Issuer (Section 7 of [RFC9943]). The Issuer's COSE signature is separate from the DSSE signature of the receipt. The COSE signature authenticates the Signed Statement's protected header and payload. The DSSE signature covers the pre-authentication encoding of the receipt's payloadType and decoded payload bytes, not the DSSE envelope as a whole. Section 3.2 or Section 3.4 checks that DSSE signature and its signer.

What a SCITT Receipt shows. During Registration the Transparency Service verifies the Issuer's signature and applies its Registration Policy (Sections 5.1.1.1 and 6.3 of [RFC9943]). It registers the Signed Statement before it releases a Receipt (Section 6.3 of [RFC9943]), and the Receipt demonstrates inclusion of the Signed Statement in the Verifiable Data Structure of the Transparency Service and can be verified without online access to it (Section 4 of [RFC9943]). The Transparency Service records the Registration time when it adds the Signed Statement (Section 7 of [RFC9943]).

What it does not show. Registering a Statement proves only that it was produced by an Issuer, not that it is accurate (Section 9.2 of [RFC9943]). Issuers can register some Statements and not others (Section 9.3 of [RFC9943]), so a Receipt for one receipt says nothing about a receipt that was never registered. The Subject Claim lets Relying Parties identify the Transparent Statements registered for a Subject (Section 3 of [RFC9943]), not receipts that were never registered. Unless the Registration Policy says otherwise, the order of the Verifiable Data Structure need not be the order of issuance (Section 9.1 of [RFC9943]). A Receipt does not change what the checks of Section 3 show, and it does not make an ALLOW an authorization.

Verification. A verifier checks the SCITT Receipt for the particular Signed Statement, including its inclusion proof, using the verification key of a Receipt Issuer it trusts and the verification rules for the applicable Verifiable Data Structure (Sections 7 and 7.1 of [RFC9943]). It establishes the binding from the registered Signed Statement to the envelope bytes or content root used in the following checks. If the Signed Statement's COSE payload is detached and this binding depends on the Issuer's signature, it verifies that signature. It then applies Section 3 of this document to the receipt. Where the envelope bytes are the Signed Statement's payload, it checks those bytes; where the Signed Statement is made over the content root, it checks an envelope whose computed content root equals the registered value. This latter comparison binds the decoded DSSE payload, not the envelope bytes or the DSSE signature, and does not establish that the DSSE signature existed when the content root was registered. Section 7.1 of [RFC9943] permits verification of a Receipt without rechecking the Issuer's signature on the Signed Statement; this does not dispense with the binding just described. The signatures of the decision maker and of the outcome signer are checked by Section 3 in either case.

3.8. Field Rules

Step 7 of Section 3.2 and step 3 of Section 3.4 apply the rules of this section. Figure 1 lists every member of the predicate of a decision receipt that this document admits, and Figure 2 every member of the predicate of an outcome receipt, each with its full path in predicate, where [] stands for each entry of an array. The rules follow the strict v0.1 form of [PB-DECISION] and [PB-OUTCOME] and their machine schemas, restricted to the members this document admits; where Section 3.2 is stricter, the figures give the stricter rule. This section states these rules itself, and its figures follow those documents as they stand at commit e1fec86351b216af369a2e6ac106dcc41aaf4b4c of [PB-REPO], where the machine schemas are the same as in version 6.4.0.

In the column Type, a digest is an object with exactly one member, sha256, whose value is 64 lowercase hexadecimal digits; a timestamp is an [RFC3339] UTC timestamp with a trailing "Z", with an upper-case "T" and "Z", a date that exists in the calendar, a second from 00 to 59 and, where present, a fraction of a second with one to six digits after the point; a typedDigest is an object whose only members are type, digestAlgorithm, digest and, optionally, purpose, where type is a non-empty string, purpose, where present, a non-empty string, digestAlgorithm jcs-sha256-v1 and digest 64 lowercase hexadecimal digits; a relationDigest is an object whose only members are digestAlgorithm, which is jcs-sha256-v1, and digest, which is 64 lowercase hexadecimal digits; an array of 1 to 64 objects holds at least one and at most 64 entries, each an object; a string is any JSON string, and a non-empty string has at least one character; a boolean is one of the JSON values true or false; any value is any JSON value. The type integer, 0 or more, is a JSON number written without the frac and exp parts of the grammar in Section 6 of [RFC8259], from 0 to 9007199254740991; a number written with frac or exp is not such an integer, even where its value is whole. In an array of objects or strings, each entry is an object or a string, and one array may hold both. The type object admits only the members listed for it, and the type open object admits any members. The type names preActionAuthorization and outcome allow only the JSON strings "preActionAuthorization" and "outcome", respectively. The type "0.1." and one or more digits means a string consisting of "0.1." followed by one or more ASCII decimal digits. The notes (1) to (6) follow the figures.

Required yes means that the member MUST be present whenever the object or array entry that contains it is present. Only decisionMaker.version, actionOutcome.performedAt and actionOutcome.outcomeRef may be null. Everywhere else a null is not allowed, including inside an open object, in an array of any values and inside a non-null member of type any value. A member that the figures do not list for an object that is not open is not allowed, including at the top level of predicate.

An entry of relationships that verifies shows a relation that the issuer of the receipt declared to the receipt or subject that its relationDigests name; it is not a statement that the target is correct.

Member  Type  Required
schemaVersion  "0.1." and one or more digits  yes
decisionId  non-empty string  yes
decisionType  preActionAuthorization  yes
decidedAt  timestamp  yes
recordedAt  timestamp  no
decisionMaker  object  yes
decisionMaker.id  non-empty string  yes
decisionMaker.version  any value  no
agent  object  yes
agent.id  non-empty string  yes
agent.version  string  no
agent.configurationDigest  digest  no
principal  object  yes
principal.id  non-empty string  yes
principal.authnContextRef  object  no
principal.authnContextRef.uri  string  no
principal.authnContextRef.digest  digest  no
delegationRefs  array of any values  no
proposedAction  object  yes
proposedAction.actionType  string  yes
proposedAction.target  object  no
proposedAction.target.name  string  no
proposedAction.target.uri  string  no
proposedAction.target.digest  digest  no
proposedAction.method  string  no
proposedAction.parametersDigest  digest  (1)
proposedAction.parametersRef  open object  (1)
proposedAction.parametersSchemaRef  object  no
proposedAction.parametersSchemaRef.uri  string  no
proposedAction.parametersSchemaRef.digest  digest  no
inputSnapshot  array of objects  yes
inputSnapshot[].name  string  no
inputSnapshot[].uri  string  no
inputSnapshot[].digest  digest  yes
inputSnapshot[].mediaType  string  no
policyBoundary  object  yes
policyBoundary.policyEngine  non-empty string  yes
policyBoundary.policyId  non-empty string  yes
policyBoundary.decisionPath  non-empty string  yes
policyBoundary.policyDigest  digest  yes
policyBoundary.bundleRevision  string  no
policyBoundary.validFrom  timestamp  no
policyBoundary.validUntil  timestamp  no
evidenceRefs  array of objects  yes
evidenceRefs[].relation  string  yes
evidenceRefs[].digest  digest  yes
evidenceRefs[].artifactDigest  digest  no
evidenceRefs[].uri  string  no
evidenceRefs[].predicateType  string  no
evidenceRefs[].typedDigest  typedDigest  no
decision  object  yes
decision.verdict  (2)  yes
decision.reasonCodes  non-empty array of non-empty strings  yes
decision.humanReadableSummary  string  no
decision.obligations  array of any values  no
decision.allowedScope  array of any values  no
notChecked  array of objects or strings  yes
notChecked[].field  string  no
notChecked[].reason  string  no
notChecked[].impact  string  no
decisionChangeConditions  array of objects  yes
decisionChangeConditions[].conditionType  string  no
decisionChangeConditions[].description  string  no
decisionChangeConditions[].requiredEvidenceType  string  no
actionOutcome  object  no
actionOutcome.status  (3)  no
actionOutcome.performedAt  timestamp  no
actionOutcome.outcomeRef  object  no
actionOutcome.outcomeRef.uri  string  no
actionOutcome.outcomeRef.digest  digest  no
traceContext  open object  no
privacy  object  yes
privacy.rawInputsIncluded  boolean  yes
privacy.redactionProfile  string  no
privacy.erased  array of any values  no
privacy.masked  array of any values  no
validity  object  yes
validity.audience  non-empty array of non-empty strings  yes
validity.nonce  non-empty string  yes
validity.expiresAt  timestamp  yes
relationships  array of 1 to 64 objects  no
relationships[].relation  (5)  yes
relationships[].targetReceiptDigest  relationDigest  yes
relationships[].targetSubjectDigest  relationDigest  no
relationships[].reason  string  no
relationships[].reasonCode  (6)  no
relationships[].declaredAt  timestamp  no
Figure 1: Members of the predicate of a decision receipt
Member  Type  Required
schemaVersion  "0.1." and one or more digits  yes
outcomeId  non-empty string  yes
decisionRef  digest  yes
executor  object  yes
executor.id  non-empty string  yes
executor.keyId  string  no
requestedActionDigest  digest  yes
status  (4)  yes
performedAt  timestamp  yes
effectDigest  digest  no
actualActionDigest  digest  no
responseDigest  digest  no
policyPurpose  outcome  no
receiverRefs  array of objects  no
receiverRefs[].relation  non-empty string  yes
receiverRefs[].digest  digest  yes
receiverRefs[].receiverId  string  no
receiverRefs[].receiverKeyId  string  no
receiverRefs[].artifactDigest  digest  no
sequence  object  no
sequence.runId  non-empty string  yes
sequence.seq  integer, 0 or more  yes
recordedAt  timestamp  no
limitations  array of strings  no
traceContext  object  no
traceContext.traceparent  string  no
validity  object  no
validity.audience  array of any values  no
validity.nonce  string  no
relationships  array of 1 to 64 objects  no
relationships[].relation  (5)  yes
relationships[].targetReceiptDigest  relationDigest  yes
relationships[].targetSubjectDigest  relationDigest  no
relationships[].reason  string  no
relationships[].reasonCode  (6)  no
relationships[].declaredAt  timestamp  no
Figure 2: Members of the predicate of an outcome receipt

The notes of the figures are as follows. (1) proposedAction MUST carry parametersDigest, parametersRef or both. This document takes no digest from parametersRef, so with parametersRef alone step 10 of Section 3.2 fails where action bytes are given, and Case 1 of Section 3.3 and step 4 of Section 3.4 fail. (2) decision.verdict is ALLOW, DENY, REFUSE, ESCALATE, DEFER or OBSERVE. (3) actionOutcome.status is notAttempted, blocked, refused, attempted, executed, failed or unknown. (4) status is executed, refused, failed or partial. (5) relationships[].relation is supersedes, revises, corrects, retracts, renews, derivedFrom or amends. (6) relationships[].reasonCode is correction, rerun, data-update, methodology-update, policy-change, withdrawal or other. Section 3.3 adds rules per case, and steps 8 to 10 of Section 3.2 and steps 4 to 7 of Section 3.4 add rules across members.

4. Approvals Are Honoured At Most Once

An approval in Case 1 covers one action and can be used once. Another copy or another signature of the payload of a signed ALLOW that a relying party has released is part of the same honouring and not a second one, as the first check of Section 3.5 counts it. For live use, the relying party MUST apply the time bounds of Section 3.3 with its own current time as T. A historical audit time MUST NOT make an expired approval usable. A relying party that lets an action run because of an approval:

  1. MUST honour each approval at most once;
  2. MUST key its consumption state by the identity of the approval (approvalId, actionDigest and session), not by the bytes of one copy of the approval record. Copies or reissues with other metadata count as the same approval. Among copies received for that identity, the relying party MUST NOT honour the approval unless T is strictly earlier than the earliest of the validUntil values, and it does not replace that bound with a later value. This observed bound is not necessarily the original validUntil: an earlier-issued copy with a shorter validity might not have been received. A relying party can claim to enforce the original bound only if it has a trustworthy basis for that bound, independent of the arrival order of copies; otherwise it claims only to enforce the observed bound. Extending validity requires a new approval act and a new approvalId;
  3. MUST mark the approval as consumed durably before it either allows the action to run or releases a signed ALLOW that cites the approval;
  4. MUST treat an approval recorded as consumed in any current, rotated, archived or compacted representation as consumed for as long as item 6 requires that state to be retained;
  5. MUST NOT honour an approval unless it can consult a complete, durable view of the consumption state required by this section. Rotation, archiving or compaction MAY replace earlier generations if all entries still required by item 6 are preserved. Missing or unreadable state whose required entries have not been preserved elsewhere MUST cause refusal;
  6. MAY drop an entry only if a trustworthy upper bound has been established on the latest time any copy of that approval identity can be honoured, that bound remains enforceable after the entry is dropped, and the bound plus the clock skew the relying party allows has passed. The earliest validUntil among received copies does not by itself establish the bound. Without a trustworthy bound, the consumed identity remains in the durable state required by items 4 and 5.

A relying party that releases a signed ALLOW citing an approval honours that approval in the sense of item 1, as one that lets the action run does. Releasing a signed ALLOW for an approval and then letting exactly the action it names run are one honouring. A relying party releases a signed ALLOW when the signed decision receipt leaves its control, to the executor, to the agent or to any store that another party can read. A signed ALLOW that never leaves its control, for example one discarded after the durable marking of item 3 failed, is not released and honours no approval. A relying party that is shown again the approval record bytes it has honoured refuses them under item 1. One that is shown a copy with other bytes refuses it under item 1 as well, because item 2 makes the copy the same approval. Releasing the same signed ALLOW again is part of the same honouring and not a second one.

Whoever issues approval records MUST NOT give a copy or reissue of an approval a validUntil later than that of the approval act it records; a later validity needs a new approval act and a new approvalId. A relying party cannot check this duty on the records it receives, which is why item 2 has it use the earliest of the validUntil values it has received.

Rotation is the step where this requirement is most easily lost. A record that is full is moved aside so that new approvals can be written; if the relying party then reads only the new generation, approvals of the old one may become usable again. Items 4 and 5 preserve the consumed state without requiring a particular storage format. A decision receipt for Case 1 binds the approval record by digest, so Section 3.5 can expose distinct ALLOW decisions for the same approval identity. Preventing reuse of an approval or of an already issued ALLOW remains the relying party's duty.

5. Implementation Status

This section is given per [RFC7942] and is to be removed by the RFC Editor before publication.

Each entry gives the known items of Section 2 of [RFC7942] with their evidence and date; an entry without evidence is not given. This information does not constitute endorsement.

5.1. proofbundle

Maintained by the author of this document, MIT License, contact through [PB-REPO]. Version 6.4.0 on PyPI [PB-PYPI], built from tag v6.4.0, commit 699545521ec5ef479bb84e61fe7461fc79ed558c, is a Python library and command line interface that produces and verifies the receipts this document describes and produced the seven DSSE vectors of Appendix A.3; its classifier is "Development Status :: 4 - Beta", and [PB-REPO] describes the decision receipt predicate as shipped since version 2.1.0 and the outcome predicate as experimental. Its decision verify and outcome verify commands check receipts of the predicate types of Section 1.2; the exit codes they returned for the vectors of this revision are given below. The commands recorded in Appendix A.3 and in the test package of Appendix A.5 give it no case of Section 3.3, no approval record and no consumption state of Section 4.

In the seven command line runs reported in Appendix A.3, the four passing vectors returned exit code 0, the invalid signature of V1-bad returned 1, and the non-canonical payload of V2-bad and the decision binding mismatch of V3-bad each returned 2. For the 149 vectors classified as failing, refused or input-invalid in the test package of Appendix A.5, the recorded commands returned a non-zero exit code for 88 and 0 for 61; for its 22 passing vectors they returned 0 for 21 and 2 for V2-micro. Sixteen of the 149 vectors have invalid input to the third check of Section 3.5 and therefore no result for that check. These results do not specify the command line interface's complete exit-code contract.

The release information above concerns tag v6.4.0, dated 27 October 2026. The test-package results were recorded on 8 November 2026 and, for the eight vectors added on 13 November 2026, on that date. On 15 November 2026 the PyPI listing of proofbundle, https://pypi.org/pypi/proofbundle/json, named version 6.5.0 as the latest; this section describes version 6.4.0.

5.2. Agent Guard

Maintained by the author of this document, MIT License, contact through [PB-REPO]. Agent Guard is the name used here for the preview in plugins/proofbundle of [PB-REPO], which its public README names the proofbundle plugin for Claude Code and Codex. This entry is based on its source and runbook at commit 8bb289c8b41d32c54734155fc00756231093bc2a of 1 November 2026, not on the release tag v6.4.0 and not on a live execution. A separate signing service signs preActionAuthorization decisions with verdict ALLOW, DENY or ESCALATE for the writes the receipt pilot supports, and an observer, which the runbook assigns to an operator account separate from the agent session, signs outcome receipts for the effects it matches to decisions. The approval path requires an explicit operator configuration, is not enabled under Codex and binds an approval to the command bytes and its session. The pilot does not produce Case 1 receipts as this document describes them; it writes released_by_owner where the vectors of Appendix A.3 write user_approval, and it leaves evidenceRefs empty.

The signing service treats an approval named in its current or a retained record as honoured; the hook treats a copy of a used approval as used and keeps a digest of each removed identity. Neither layer keeps an original validity per identity, so a later copy of an identity not yet used opens a later window, which item 2 of Section 4 does not permit where an earlier copy was received. No conformance of Agent Guard to Section 4 and no enforcement of single use in a running host is claimed from these source-level observations.

5.3. Go Verifier of the Test Package

Written by the author of this document, contact through [PB-REPO]. The prototype in the directory verifier of the test package of Appendix A.5 is written in Go from the text of this document with the Go standard library only and shares no code with proofbundle. It implements checks of Section 3.2 to Section 3.4, Section 3.8, the checks of a relying party in Section 4 that the vectors exercise and the second and third checks of Section 3.5; the test package shows its results for the vectors it holds, not that every condition of those sections is implemented. It agrees with the proofbundle 6.4.0 command line interface on acceptance or rejection of all seven vectors, and all 171 vectors of the test package gave the result recorded for them, on 8 November 2026 and, for the eight vectors added on 13 November 2026, on that date. This does not establish complete equivalence of the verifiers, conformance of Agent Guard to Sections 3 and 4, or enforcement of single use in a running host. No later implementation revision is assessed here.

5.4. Registration at a Local scitt-ccf-ledger

A measurement of the author, not an implementation of this document and not a production service, from tools/scitt_ccf_external/README.md at tag v6.4.0 of [PB-REPO], lines 341 to 398. On 25 September 2026 a Signed Statement over the [RFC8785] root of an example bundle of proofbundle, not a decision receipt, was submitted for registration in eight forms at an instance of [SCITT-CCF-LEDGER] at commit 00101f769d872711356e080fbb089ac48589c60a, run locally with one node in virtual mode and CCF 7.0.17. The service accepted some of the forms and refused others, and [SCITT-VERIFIER] 0.4.0, built at commit bd6fb8ba79dbb521257b7f09682c03c6681dc3d0, ran offline on twelve statements; 58 values compared with proofbundle agreed, none differed and 9 were not comparable. It covers no step of Section 3 and no registration as Section 3.7 describes it.

5.5. Implementation Written from the Text

Toshikatsu Oga (The HORIZONs Co., Ltd.) checked the test vectors of several revisions of this document with an implementation written from the text and reported places that read two ways. By his report, it is one Python 3 file per revision with the Python standard library and the Python package cryptography for Ed25519, implementing checks from Section 3.2 to Section 3.5, Section 3.8 and Section 4 as present in the respective revision; this entry does not establish complete coverage of those sections or independently verified conformance to them. The results below concern revisions -03 to -06; no result for revision -07 is reported here.

By his report, its result and failing step agreed with this document for all 7 vectors of Appendix A of revision -03, which prints their expected results, so that check was not blind; for all 17 vectors of Appendices A.3 and A.5 of revision -04; and for the 21 vectors of Appendices A.3 and A.5 of revision -05 other than F15, which was not run against that revision. For an earlier test package of revision -06 with SHA-256 850195e6288dc1716a77aa92c4732b54caa20aacf7661ef24059e314341f001c, made before that revision was submitted and not published, they agreed for all 96 of its vectors; the verdicts were recorded at 08:48:54 UTC on 4 November 2026, before the expected results of that package were read, and F15 was run against the receipt of vector P1 of [SER].

The evidence for his reported results is the Acknowledgments section of draft‑gruszka‑audit‑decision‑receipts‑06, dated 15 November 2026, and his message to the author of 7 November 2026. His message of 8 November 2026 reports publication of the implementation for revision -06 as dr_check 0.1.0 under the MIT License. The published reader was inspected on 11 November 2026 at https://github.com/ogasurfproject-jpg/horizon-shield/tree/53ea626494f79eec7f935f56406decd7c225ca7d/tools/decision-receipts-reader. On 9 November 2026 the author ran dr_check 0.1.0 at that commit, without network access, on the test package of revision -06 named above; all 96 verdicts agreed with the expected results of that package in both result and failing step, repeating his reported result for that package.

7. Security Considerations

Key binding. A receipt is only as trustworthy as the verifier's choice of keys and its policy for the identities those keys may attest. Section 3.2 checks decisionMaker.id; Section 3.4 checks the expected executor.id under the outcome-signing key. An observer's signature over an executor identifier is not the executor's own signature. A key held in the recorded Agent's process gives an Auditor that does not trust that Agent little assurance. The inspected Agent Guard design uses separate service accounts as described in Section 5.

Key compromise. Whoever holds the signing key of a decision maker or of an outcome signer can produce receipts with any decidedAt or performedAt, since these times come from the signer's clock, as the paragraph on clocks below says. A check under Section 3 does not tell such a receipt from one signed before the key was compromised or replaced. Which keys are accepted for which identity and for which period is set by the verifier's trust policy, not by this document; [RFC9943] likewise leaves revocation strategies for compromised keys out of its scope (Section 9.4.2 of [RFC9943]). A time anchor outside the signer helps, if it was set before the compromise. Where the envelope bytes are the payload of a registered Signed Statement (Section 3.7), a Receipt for it shows that the DSSE signature existed when that Signed Statement was registered, and the Transparency Service records the Registration time (Section 7 of [RFC9943]). Where the Signed Statement is made over the content root, the comparison binds the decoded DSSE payload, not the envelope bytes or the DSSE signature, and does not establish that the DSSE signature existed when the content root was registered (Section 3.7). The rules of Section 4 keep a relying party from honouring one approval more than once; they do not prevent the holder of a compromised decision maker key from signing receipts that cite approvals with new identities.

Exact bytes. A verifier that re-serializes a payload before checking it can be led to accept bytes that were never signed, or to compute another content root. Step 4 refuses such payloads, and vector V2-bad tests one of them.

Algorithms. Section 3 fixes Ed25519 for signatures and SHA-256 for digests. A digest object names its algorithm by its only member, sha256, and a typedDigest or a relationDigest by its digestAlgorithm, which is jcs-sha256-v1. The envelope names no signature algorithm; the verifier uses the Ed25519 public key that its trust policy selects (Section 3.1). A verifier that follows Section 3 accepts no other algorithm, so a change of algorithm needs a revision of this procedure.

Replay. An ALLOW for one action must not be usable for another, and an outcome that names another decision must not count as the result of this one. parametersDigest, where present, binds the action bytes, validity binds audience, nonce and expiry, decisionRef binds the outcome, and Section 4 requires the relying party to prevent reuse of an approval. V1-bad tests rejection of changed payload bytes under an unchanged signature. V3-bad tests rejection of an outcome bound to a different decision. The seven vectors do not test consumption-state retention or rotation; F13 in the test package of Appendix A.5 tests refusal of an approval that is recorded as consumed only in an already rotated generation, not the rotation itself.

Self-assertion. actionOutcome in a decision receipt and an outcome receipt whose effect cannot be resolved are statements of their signers. A digest that is present but cannot be resolved to bytes was chosen by the signer.

Suppression. A decision maker can fail to write a receipt for a decision it does not want seen. Independent observers at the boundary, as the architecture requires for Action Records, reduce this risk. Registration with a Transparency Service helps address selective registration when Relying Parties accept a Signed Statement only if they can discover and verify a Receipt for that Signed Statement from a Transparency Service they trust (Sections 7.1 and 9.3 of [RFC9943]). This document does not ensure that a receipt is produced for every decision. A SCITT Receipt shows inclusion of the registered Signed Statement, not that every receipt was registered (Section 3.7). Completeness presupposes an expectation of which receipts ought to exist, and this document does not define that expectation.

Gaps. For outcome receipts, [PB-OUTCOME] defines an optional member sequence, with a run identifier and a counter that an executor can opt into (its sections 4 and 7); Section 3.4 admits that member. A verifier that holds numbered outcome receipts of a run can report a gap in the numbers it has. Such a gap can reflect withholding or delivery delay; the numbers alone do not establish its cause. Those sequence numbers alone do not show whether receipts above the highest observed number have been withheld, and a run without the member leaves no such trace (compare Section 6.3 of [GILDA]). The profile of Section 3.2 defines no sequence counter and requires no chain of decision receipts; a predecessor named in relationships (Section 3.5) is optional and does not show a receipt that was withheld. Records that a Service or Tool emits at the boundary where an action took effect (Section 4.1 of [AUDIT-ARCH]) can expose an effect for which the Auditor has no matching decision receipt.

Clocks. decidedAt and performedAt come from the signers' clocks. Ordering against outside events needs a timestamp or a transparency registration. A verified time-stamp token under [RFC3161] for a hash of the envelope bytes is, like a registration (Section 3.7), an anchor outside the signer for a verifier that trusts the time-stamping authority. It supports the assertion that those bytes existed no later than genTime plus the applicable accuracy (Section 2.4.2 of [RFC3161]). If the accuracy field is absent, the accuracy may be available through other means, such as the TSA policy; its absence does not establish zero uncertainty. validity.expiresAt, recordedAt and the times of an approval record are statements of their signer as well; the times of an approval record are the gate's statement, not the person's (Section 2.5). The checks of step 9 of Section 3.2 and of Case 1 in Section 3.3 compare decidedAt, validity.expiresAt and the times of an approval record with each other and with the verifier's time T; none of them compares recordedAt. They do not show when a decision was made or when a person approved.

8. Privacy Considerations

The example receipts in Appendix A carry digests of action bytes, approval records and effects rather than those bytes, and the example decision receipts set privacy.rawInputsIncluded to false. The field rules of Section 3.8 allow that member to be true and permit raw content in other admitted members, such as a decision receipt's open traceContext object. A successful check under Section 3 does not establish that raw inputs are absent. A digest of a short or predictable input, such as a command line, can be found by trying candidate inputs; a deployment that must hide such inputs keeps the receipts as private as the inputs, or adds a secret salt to the digested bytes. Identifiers such as principal.id, agent.id and traceContext link a person's actions across services, as Section 9 of [AUDIT-ARCH] describes; pairwise or opaque identifiers reduce that linkage. The approver of an approval record and principal.id can be pseudonymous identifiers or references to a role, and the mapping from them to a person is then kept by the deploying organization. An approval record that names a person identifies that person to everyone who holds it; how long it is held is set by the deployment. Registration shows that the registered Signed Statement existed by the time of registration, not when the decision was taken. Where that Statement is made over the content root, it does not establish that the receipt's DSSE signature existed at that time (Section 3.7). Where the envelope bytes are the payload of a registered Signed Statement (Section 3.7), readers of that payload can read the receipt with its identifiers and digests. Registering a Statement over the content root avoids including the receipt's envelope bytes in that Statement, but the Signed Statement still carries its Issuer and Subject Claims and the Transparency Service records the Registration time (Sections 6 and 7 of [RFC9943]). Access to the registered data and any detached payload depends on the Transparency Service and any Adjacent Services; Issuers and Clients assess their privacy and security posture before registration (Section 8 of [RFC9943]).

9. IANA Considerations

This document has no IANA actions.

10. References

10.1. Normative References

[DSSE]
Secure Systems Lab, "DSSE: Dead Simple Signing Envelope, protocol specification, version 1.0.2", , <https://github.com/secure-systems-lab/dsse/blob/v1.0.2/protocol.md>.
[INTOTO-STATEMENT]
in-toto project, "in-toto Attestation Framework: Statement layer, v1 (release v1.2.0)", , <https://github.com/in-toto/attestation/blob/v1.2.0/spec/v1/statement.md>.
[RFC2119]
Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, , <https://www.rfc-editor.org/info/rfc2119>.
[RFC3339]
Klyne, G. and C. Newman, "Date and Time on the Internet: Timestamps", RFC 3339, DOI 10.17487/RFC3339, , <https://www.rfc-editor.org/info/rfc3339>.
[RFC4648]
Josefsson, S., "The Base16, Base32, and Base64 Data Encodings", RFC 4648, DOI 10.17487/RFC4648, , <https://www.rfc-editor.org/info/rfc4648>.
[RFC6234]
Eastlake 3rd, D. and T. Hansen, "US Secure Hash Algorithms (SHA and SHA-based HMAC and HKDF)", RFC 6234, DOI 10.17487/RFC6234, , <https://www.rfc-editor.org/info/rfc6234>.
[RFC7493]
Bray, T., Ed., "The I-JSON Message Format", RFC 7493, DOI 10.17487/RFC7493, , <https://www.rfc-editor.org/info/rfc7493>.
[RFC8032]
Josefsson, S. and I. Liusvaara, "Edwards-Curve Digital Signature Algorithm (EdDSA)", RFC 8032, DOI 10.17487/RFC8032, , <https://www.rfc-editor.org/info/rfc8032>.
[RFC8174]
Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, , <https://www.rfc-editor.org/info/rfc8174>.
[RFC8259]
Bray, T., Ed., "The JavaScript Object Notation (JSON) Data Interchange Format", STD 90, RFC 8259, DOI 10.17487/RFC8259, , <https://www.rfc-editor.org/info/rfc8259>.
[RFC8785]
Rundgren, A., Jordan, B., and S. Erdtman, "JSON Canonicalization Scheme (JCS)", RFC 8785, DOI 10.17487/RFC8785, , <https://www.rfc-editor.org/info/rfc8785>.

10.2. Informative References

[AUDIT-ARCH]
Kuehlewind, M. and H. Birkholz, "An Architecture for Auditing Agent Delegation and Interactions", Work in Progress, Internet-Draft, draft-kuehlewind-audit-architecture-01, , <https://datatracker.ietf.org/doc/html/draft-kuehlewind-audit-architecture-01>.
[BIRKHOLZ]
Birkholz, H., Petrie, D. G., McCarthy-Howe, T., Rosenberg, J., and O. Steele, "An AI Agent Extension for Virtualized Conversations (vCon)", Work in Progress, Internet-Draft, draft-birkholz-vcon-ai-agent-extension-00, , <https://datatracker.ietf.org/doc/html/draft-birkholz-vcon-ai-agent-extension-00>.
[DPS-RAR]
"dps-scitt-decision-layer, interop/rar, commit c0a7bd2b6e80b294be326f6401d89a25a50da17e", , <https://github.com/Laurent-Lemonnier/dps-scitt-decision-layer/tree/c0a7bd2b6e80b294be326f6401d89a25a50da17e/interop/rar>.
[GILDA]
Gilda, S., "An Audit Record Format for AI Agent Authorization Decisions", Work in Progress, Internet-Draft, draft-gilda-audit-agent-audit-record-03, , <https://datatracker.ietf.org/doc/html/draft-gilda-audit-agent-audit-record-03>.
[IZMAYLOV]
Izmaylov, P., "Signed Permissions and Action Receipts for Automated Agents", Work in Progress, Internet-Draft, draft-izmaylov-agent-permission-receipts-02, , <https://datatracker.ietf.org/doc/html/draft-izmaylov-agent-permission-receipts-02>.
[JCS-JS]
Erdtman, S., "canonicalize 2.1.0, a JavaScript implementation of RFC 8785", , <https://www.npmjs.com/package/canonicalize/v/2.1.0>.
[KRAUSZ]
Krausz, J., "The verification.* Constraint Family: Pre-Action Fail-Closed Gates for AI Agent Decisions", Work in Progress, Internet-Draft, draft-krausz-verification-state-03, , <https://datatracker.ietf.org/doc/html/draft-krausz-verification-state-03>.
[LEMONNIER]
Lemonnier, L., "Decision Provenance for SCITT: binding software releases and agent actions to the decisions behind them", Work in Progress, Internet-Draft, draft-lemonnier-scitt-decision-provenance-01, , <https://datatracker.ietf.org/doc/html/draft-lemonnier-scitt-decision-provenance-01>.
[LIU]
Liu, D., Zhu, H., Krishnan, S., and A. Parecki, "Authorization Evidence and Audit Trail for OAuth 2.0 Access Tokens", Work in Progress, Internet-Draft, draft-liu-oauth-authorization-evidence-02, , <https://datatracker.ietf.org/doc/html/draft-liu-oauth-authorization-evidence-02>.
[PB-DECISION]
proofbundle project, "Decision Receipt predicate decision-receipt/v0.1", proofbundle, tag v6.4.0, docs/predicates/decision-receipt.md, , <https://github.com/b7n0de/proofbundle/blob/v6.4.0/docs/predicates/decision-receipt.md>.
[PB-OUTCOME]
proofbundle project, "Action Outcome predicate action-outcome/v0.1", proofbundle, tag v6.4.0, docs/predicates/action-outcome.md, , <https://github.com/b7n0de/proofbundle/blob/v6.4.0/docs/predicates/action-outcome.md>.
[PB-PYPI]
Python Package Index, "proofbundle 6.4.0", , <https://pypi.org/project/proofbundle/6.4.0/>.
[PB-REPO]
proofbundle project, "proofbundle source repository", , <https://github.com/b7n0de/proofbundle>.
[RFC3161]
Adams, C., Cain, P., Pinkas, D., and R. Zuccherato, "Internet X.509 Public Key Infrastructure Time-Stamp Protocol (TSP)", RFC 3161, DOI 10.17487/RFC3161, , <https://www.rfc-editor.org/info/rfc3161>.
[RFC7942]
Sheffer, Y. and A. Farrel, "Improving Awareness of Running Code: The Implementation Status Section", BCP 205, RFC 7942, DOI 10.17487/RFC7942, , <https://www.rfc-editor.org/info/rfc7942>.
[RFC8949]
Bormann, C. and P. Hoffman, "Concise Binary Object Representation (CBOR)", STD 94, RFC 8949, DOI 10.17487/RFC8949, , <https://www.rfc-editor.org/info/rfc8949>.
[RFC9052]
Schaad, J., "CBOR Object Signing and Encryption (COSE): Structures and Process", STD 96, RFC 9052, DOI 10.17487/RFC9052, , <https://www.rfc-editor.org/info/rfc9052>.
[RFC9053]
Schaad, J., "CBOR Object Signing and Encryption (COSE): Initial Algorithms", RFC 9053, DOI 10.17487/RFC9053, , <https://www.rfc-editor.org/info/rfc9053>.
[RFC9162]
Laurie, B., Messeri, E., and R. Stradling, "Certificate Transparency Version 2.0", RFC 9162, DOI 10.17487/RFC9162, , <https://www.rfc-editor.org/info/rfc9162>.
[RFC9396]
Lodderstedt, T., Richer, J., and B. Campbell, "OAuth 2.0 Rich Authorization Requests", RFC 9396, DOI 10.17487/RFC9396, , <https://www.rfc-editor.org/info/rfc9396>.
[RFC9597]
Looker, T. and M.B. Jones, "CBOR Web Token (CWT) Claims in COSE Headers", RFC 9597, DOI 10.17487/RFC9597, , <https://www.rfc-editor.org/info/rfc9597>.
[RFC9679]
Isobe, K., Tschofenig, H., and O. Steele, "CBOR Object Signing and Encryption (COSE) Key Thumbprint", RFC 9679, DOI 10.17487/RFC9679, , <https://www.rfc-editor.org/info/rfc9679>.
[RFC9942]
Steele, O., Birkholz, H., Delignat-Lavaud, A., and C. Fournet, "CBOR Object Signing and Encryption (COSE) Receipts", RFC 9942, DOI 10.17487/RFC9942, , <https://www.rfc-editor.org/info/rfc9942>.
[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, , <https://www.rfc-editor.org/info/rfc9943>.
[RFC9995]
Steele, O., Lasker, S., and H. Birkholz, "CBOR Object Signing and Encryption (COSE) Hash Envelope", RFC 9995, DOI 10.17487/RFC9995, , <https://www.rfc-editor.org/info/rfc9995>.
[SCITT-CCF-LEDGER]
Microsoft, "scitt-ccf-ledger, commit 00101f769d872711356e080fbb089ac48589c60a", , <https://github.com/microsoft/scitt-ccf-ledger/tree/00101f769d872711356e080fbb089ac48589c60a>.
[SCITT-COSE]
Action State Group, "scitt-cose 0.5.0", , <https://pypi.org/project/scitt-cose/0.5.0/>.
[SCITT-VERIFIER]
Microsoft, "scitt-verifier, commit bd6fb8ba79dbb521257b7f09682c03c6681dc3d0", , <https://github.com/microsoft/scitt-verifier/tree/bd6fb8ba79dbb521257b7f09682c03c6681dc3d0>.
[SER]
Gruszka, K., "Signed Evaluation Receipts", Work in Progress, Internet-Draft, draft-gruszka-signed-evaluation-receipts-04, , <https://datatracker.ietf.org/doc/html/draft-gruszka-signed-evaluation-receipts-04>.
[SER-MAPPINGS]
Gruszka, K., "Citing Signed Evaluation Receipts from Existing Envelopes", Work in Progress, Internet-Draft, draft-gruszka-evaluation-receipt-mappings-04, , <https://datatracker.ietf.org/doc/html/draft-gruszka-evaluation-receipt-mappings-04>.
[WRIGHT]
Wright, D., "Verdict Records: Results on Claims that Cite Receipts as Evidence", Work in Progress, Internet-Draft, draft-wright-verdict-record-02, , <https://datatracker.ietf.org/doc/html/draft-wright-verdict-record-02>.
[YOSSIF]
Yossif, M. K., "PSEA Token Profile: An EAT Profile for Action-Bound, User-Verification-Gated Transaction-Confirmation Evidence", Work in Progress, Internet-Draft, draft-yossif-psea-03, , <https://datatracker.ietf.org/doc/html/draft-yossif-psea-03>.

Appendix A. Test Vectors

The seven DSSE vectors described in Appendix A.3 were generated with proofbundle 6.4.0 from PyPI. All vectors and keys in this appendix are for testing only. Each key seed is SHA-256 over a public ASCII label, so that every byte can be rebuilt; these keys MUST NOT be used for any real receipt. Ed25519 is deterministic, so the signatures below follow from the seeds and the payloads.

A.1. Test Keys and Inputs

Each key is given by its label, Ed25519 seed (hex), public key (base64) and keyid. The key labelled gate-1 signs the decision receipts; the key labelled observer-1 signs the outcome receipts. For these tests, the trust policy accepts decisionMaker.id gate-1 under the gate-1 key and permits the observer-1 key to attest executor.id forge:agent-42-bot. The key label observer-1 does not assert that the observer is the executing account.

gate-1
draft-gruszka-audit-decision-receipts test key gate-1
100fa92fa6464c141f791a96327f555c1c0cfbf3095af986537e50c3c4335f8c
TWfs75IN1In3CNI8TnyIOUHgtwilOPGcRFQLJ670CJc=
SHA256:NjZMAelLq9XON3pxV5tR5hZjX99NkQyw4gGJSRc2gP8

observer-1
draft-gruszka-audit-decision-receipts test key observer-1
f03ad5b668ad225ea15fd9fd272fc8df84be59f467da86b0f2ad0a5ad26f4ddd
4hDsi5TCL+A+9SB7k2l4EkY7XePUheE+zE9mBazt2Fs=
SHA256:q/d0VZCREBObUFi8PWKWXEF4lVNeYPuGn0GuOqd1ij0

Each input is given by a name, its bytes, its size and its SHA-256. "\n" stands for one line feed byte. The bytes of ap, a3 and e3 are the serialization under [RFC8785] of the JSON value shown, on one line and without white space. In ap, SHA-256(a1) stands for the JSON string of the 64 lowercase hexadecimal digits of the SHA-256 of a1. pol is the policy named by every policyDigest. B is the payload of vector P1 of [SER], Appendix A.2.1, cited by V3-decision, F17 and F89 and carried by F15, vectors of the test package of Appendix A.5; its bytes are given in base64 [RFC4648].

a1 "git push origin refs/heads/topic:refs/heads/topic\n"
50 bytes, SHA-256
dec54f771352454097310ae4bd6608c0070ebebb3955c6be013c1b3a8e310bd0
a1x "git push --force origin refs/heads/topic:refs/heads/main\n"
57 bytes, SHA-256
b683418806875ff79c6fa190c50edab0b96577ad8062de5ba7b41dc1d9c363fb
a2 "gh pr merge 7 --repo acme/widgets --merge\n"
42 bytes, SHA-256
5a7549948a1441addb710f5ada59e06dbce25cc5a59accf14f140379f47adb5b
pol "example gate rules, revision 1\n"
31 bytes, SHA-256
43789b41915ae0febcd13263d293101af5ab041eb7c234e1cbe625d6312815f6
ap, 231 bytes, SHA-256
88e5b6d69b3e4bc250a42aeffea0414d406e5f04ba4f991ee432a78231da1170
{
 "actionDigest": {
  "sha256": SHA-256(a1)
 },
 "approvalId": "appr-001",
 "approvedAt": "2026-10-15T09:00:00Z",
 "approver": "user-123",
 "session": "sess-1",
 "validUntil": "2026-10-15T09:15:00Z"
}
a3, 81 bytes, SHA-256
9165c850b7e03cd59578efa1769be74c5386cff5ce96df6b59e25d38ad6f88b4
{
 "body": "All required checks passed for 4f1c2a9.",
 "pull": 7,
 "repo": "acme/widgets"
}
e3, 93 bytes, SHA-256
d3d824cf55168298c0f651244853ec9d9620ea09087abe899243ceb8f9df3a6d
{
 "author": "agent-42-bot",
 "body": "All required checks passed for 4f1c2a9.",
 "id": 9001,
 "pull": 7
}
B, 458 bytes, SHA-256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A.2. Statements

The payload bytes of each vector are the serialization under [RFC8785] of a Statement {"_type": "https://in-toto.io/Statement/v1", "predicateType": P, "predicate": D, "subject": [{"name": N, "digest": {"sha256": H}}]}, where H is SHA-256 over the serialization of D. For a decision receipt P is the decision receipt type and N is "decision:" followed by decisionId; for an outcome receipt P is the outcome receipt type and N is "outcome:" followed by outcomeId.

The predicate of V1-good is given below as a list of assignments that starts from the predicate named in its first line, {} being the empty object; the Statements of the other vectors are in the test package of Appendix A.5. An assignment sets the member at the path to the JSON value shown. A path is a sequence of member names joined by dots, where a decimal index i in square brackets after a name selects element i of an array; missing objects on the path are created, and an assignment to element i of an array with i elements appends it. Members not named keep their value. SHA-256(x) stands for the JSON string of the 64 lowercase hexadecimal digits of the SHA-256 of input x, and R(V) for that of the payload bytes of vector V.

V1-good, from {}
agent.id = "agent-42"
decidedAt = "2026-10-15T09:00:05Z"
decision.reasonCodes = ["user_approval"]
decision.verdict = "ALLOW"
decisionChangeConditions = []
decisionChangeConditions[0].conditionType = "ruleChange"
decisionChangeConditions[0].description =
 "a changed gate, mode or rule of the gate"
decisionId = "dec-0001"
decisionMaker.id = "gate-1"
decisionType = "preActionAuthorization"
evidenceRefs = []
inputSnapshot = []
inputSnapshot[0].digest.sha256 = SHA-256(a1)
inputSnapshot[0].mediaType = "text/plain"
inputSnapshot[0].name = "approved bytes"
inputSnapshot[1].digest.sha256 = SHA-256(ap)
inputSnapshot[1].mediaType = "application/json"
inputSnapshot[1].name = "approval record"
notChecked = []
notChecked[0].field = "content"
notChecked[0].impact = "an approved text can be wrong"
notChecked[0].reason =
 "the gate does not judge whether the text is true"
policyBoundary.bundleRevision = "1"
policyBoundary.decisionPath = "outbound-gate/allow-write"
policyBoundary.policyDigest.sha256 = SHA-256(pol)
policyBoundary.policyEngine = "proofbundle-pilot/github-write"
policyBoundary.policyId = "outbound-gate-rules"
principal.id = "user-123"
privacy.rawInputsIncluded = false
proposedAction.actionType = "git.push"
proposedAction.method = "write"
proposedAction.parametersDigest.sha256 = SHA-256(a1)
proposedAction.target.name =
 "https://git.example/acme/widgets refs/heads/topic"
proposedAction.target.uri = "https://git.example/acme/widgets"
schemaVersion = "0.1.0"
traceContext.parentId = "int-002"
traceContext.traceId = "trace-abc"
validity.audience = ["observer-1"]
validity.expiresAt = "2026-10-15T09:05:05Z"
validity.nonce = "dec-0001"

V1-bad has the payload bytes of V1-good with every occurrence of the hexadecimal SHA-256 of a1 replaced by that of a1x, and keeps the signature of V1-good. V2-bad has as payload bytes the output of Python's json.dumps(s, sort_keys=True) for the Statement s of V2-good, with a space after each comma and colon, and is validly signed over them by gate-1.

A.3. Results

Every envelope is {"payload": base64 of the payload bytes, "payloadType": "application/vnd.in-toto+json", "signatures": [{"keyid": keyid of the signer, "sig": signature}]}. Each of the two blocks below gives the vector, the evaluation time T, the expected result with the exit code of the proofbundle 6.4.0 command line interface, the payload size and signer, the SHA-256 of the payload bytes (the content root where the payload is canonical) and the signature in base64 on two lines. The expected audience is observer-1 and the expected nonce is the decisionId. decision verify ran with --strict, --aud, --nonce and --verification-time T; outcome verify with --strict, --expected-decision-ref R(V3-decision) and --decision-maker-id gate-1.

V1-good, T 2026-10-15T09:01:00Z, pass, exit 0
1829 bytes, signed by gate-1
b9dfd26537542632013d0df166141a788551b4987dfa33b060d5b5ebba78dfb1
GkvzPjklV5ZbtomNRXbS4+O4MIOSw8xGmA0diAEYKEm0
s4Q9qLkL1j7+e5RZnT9l56tfB2utDoNycACsSvSoDw==
V1-bad, T 2026-10-15T09:01:00Z, fail at step 3, exit 1
1829 bytes, signed by gate-1
69eb87fb33a29153182b0e0c89a444023cc00c82fe7c8db1c7e0893363b2c926
GkvzPjklV5ZbtomNRXbS4+O4MIOSw8xGmA0diAEYKEm0
s4Q9qLkL1j7+e5RZnT9l56tfB2utDoNycACsSvSoDw==

The blocks above give V1-good and V1-bad in full. The other five vectors of this appendix are in the test package of Appendix A.5, with their payload bytes, envelopes and Statements; Figure 3 gives for each its name, the procedure it is judged by, its expected result with the evaluation time T and the exit code of the proofbundle 6.4.0 command line interface, and the SHA-256 of its receipt file in the test package.

V2-good: Case 2 of Section 3.3, using a2
  T 2026-10-15T09:06:00Z, pass, exit 0
  3e3b7d575b712ada2c813ee98fd3b0c80d788e2558444a34338510c9978fa47a

V2-bad: V2-good not canonical, see Appendix A.2
  T 2026-10-15T09:06:00Z, fail at step 4, exit 2
  afa39fd7086e4c5af7ef5c7b8cf0ceeaedcb0af26a07ac8082efeb4874b0f424

V3-decision: Case 3, the pair with V3-good
  T 2026-10-15T09:11:00Z, pass, exit 0
  b28dc232b4bb98044a962714c99f1343df3e136246e434bf0f7fdd72dbb925ed

V3-good: Case 3 with V3-decision and e3
  pass, exit 0
  436fb6f6f19cecbd67df9f468556807eab5003ddb8eb60c8b9c2f708e3e9bb66

V3-bad: V3-good with decisionRef R(V1-good)
  fail at step 4 of Section 3.4, exit 2
  2a0fe3ab106095df99e1ee41b938e04678973b7a4fed14c0482f63eca6472d05
Figure 3: Further vectors of Appendix A.3

The payload bytes of V1-good, V2-good, V3-decision, V3-good and V3-bad and the inputs ap, a3, e3 and B are each equal to three serializations of their parsed value: Python's json.dumps with sort_keys=True, separators=(",", ":") and either setting of ensure_ascii; the rfc8785 package 0.1.4 for Python; and the independent JavaScript implementation [JCS-JS]. The values contain no floating-point numbers and no characters outside ASCII, which keeps these serializations equal. The payload of V2-bad differs from all of them, as intended.

Each result is that of one procedure. V1-good is judged by Section 3.2 with Case 1 of Section 3.3, using a1 and ap; V1-bad fails at step 3 of Section 3.2, before any check of Case 1. V2-good is judged by Section 3.2 with Case 2, using a2; V2-bad fails at step 4 of Section 3.2. V3-good is judged as Case 3, by Section 3.2 for V3-decision and Section 3.4 for V3-good, with e3 as effect bytes, no action bytes for step 10 of Section 3.2, and forge:agent-42-bot as expected executor.id; V3-bad fails at step 4 of Section 3.4. V3-decision is judged by the same Case 3 procedure as V3-good and passes; it has no result of its own, and its pass is that of the pair with V3-good. The exit code of proofbundle decision verify for V3-decision reflects Section 3.2 alone. In the test package of Appendix A.5, V3-decision and F31 to F34, which are derived from it, are each checked as the pair with V3-good.

A.4. Signed Statements for Registration

The test package of Appendix A.5 holds, for V2-good, one Signed Statement for each way of registration in Section 3.7, a SCITT Receipt for each and the counterexamples, with the test keys issuer-1 and receipt-issuer-1, built like the keys of Appendix A.1, and the text that gives their bytes and describes how they are built and checked. The Signed Statements are COSE_Sign1 messages of [RFC9052] with an EdDSA signature of [RFC9053], CWT Claims of [RFC9597] and a kid that is a COSE Key Thumbprint of [RFC9679], in the deterministic encoding of [RFC8949]; the second uses the COSE Hash Envelope of [RFC9995]. The Receipts are those of [RFC9942] over a Merkle Tree of [RFC9162], generated for illustration with scitt-cose 0.5.0 [SCITT-COSE], and no Transparency Service issued them.

A.5. Test Package and Further Vectors

Appendix A.3 gives a passing and a failing vector for each case of Section 3.3, V1-good and V1-bad for Case 1, V2-good and V2-bad for Case 2, and V3-decision and V3-good with V3-bad for Case 3 and Section 3.4. For every vector but F38 the expected audience is observer‑1, and the expected nonce is the decisionId of the vector of Appendix A.3 or of F10 that it is derived from; F38 alone is checked with the expected audience observer‑2, which its receipt does not name, and F39 alone with another expected nonce. V1-accept is V1-good checked by a relying party whose consumption state is the empty JSON array; it passes, and no item of Section 4 refuses it.

The other failure vectors, F1 to F9, F11 to F93 and F95 to F125 with F17-outcome, the passing vectors V2-micro, V2-num, V2-rel, V3-rel, V3-seq, F94, V1-accept, V2-ref and V3-late, K1 to K29 of the third check of Section 3.5 and S1 and S2 of its second check are in the test package archive draft‑gruszka‑audit‑decision‑receipts‑08‑test‑vectors.tar.gz, with SHA-256 a76c704a221c63bbceef8baf4c228f81502754a0a56ebe0a68513df373c9140b. The URL https://github.com/b7n0de/proofbundle/raw/32b7fe6a61031026040cb1f0d22a0b156b97c040/drafts/draft-gruszka-audit-decision-receipts/draft-gruszka-audit-decision-receipts-08-test-vectors.tar.gz identifies that exact archive file at a fixed commit of the public proofbundle repository, not merely its unpacked directory. The package holds every vector of Appendix A with the inputs it uses and the result and step expected for it, a description of how each vector is derived and checked with the exit codes of the proofbundle 6.4.0 command line interface, the Signed Statements and SCITT Receipts that Appendix A.4 names with the text that describes them, a verifier of Section 3 and Section 4 written in Go from this document, and a script that runs every vector against that verifier offline. Its file MANIFEST.sha256 gives the SHA-256 of each other file.

F19 to F71 in the test package were added for requirements of this document that F1 to F18 leave without a vector and that can be checked on a receipt and its inputs; the test package records the step at which each fails. The six requirements named below have no vector because a receipt and its inputs do not show whether they are met. They are not the only requirements without a vector; others lack one too, among them the first check of Section 3.5 and the conditions of item 6 of Section 4, and a requirement with a vector need not have one that separates each condition it states. The vectors therefore do not show that every requirement that can be checked on a receipt and its inputs is tested. The consumption scope of Section 3.5, the rule of Appendix A that the test keys are not used for real receipts and the rule of Section 4 for whoever issues approval records cannot be checked on a receipt. The two sentences of the opening paragraph of Section 4 that have the relying party apply its own current time as T and keep a historical audit time from making an expired approval usable, and item 3 of Section 4, which orders durable marking before release, state duties of the relying party outside the procedures of Section 3. The second check of Section 3.5 has the vectors S1 and S2, which the verifier of the test package checks (Section 5.3); its first check has no vector.

Figure 4 names each further vector by group, with the vector it is derived from, its change and its expected result, and gives for each vector the SHA-256 of the receipt file that the recorded command of its test package entry checks; the README of the test package describes them in prose, and its file MANIFEST.sha256 gives the path of each file. A changed payload is signed again by the key that signed its base, with the subject digest taken over the new predicate. For K1 to K29 and S1 and S2, the receipt passes, and Expected gives the result of the third or second check of Section 3.5 over the receipts held, each held under the key that signed it unless the list says otherwise; an invalid input gives no result for that check. F10 is V1-good with decidedAt 2026-10-15T09:20:00Z, decisionId and validity.nonce dec-0004, validity.expiresAt 2026-10-15T09:25:00Z and the approval record ap2, ap with validUntil 2026-10-15T09:30:00Z, in inputSnapshot[1]; c1 is the consumption state of a relying party that has honoured ap. F90 is refused under item 2 of Section 4. The other vectors of the test package, F1 to F9, F11 to F88 and F17-outcome, are described only in the package.

Decision receipts, Section 3.2; Expected names the step of Section
3.2:

F91, F92, F93, from V2-good: member n, 2^53 plus 2, 2^53, -2^53; step
  4
F91  b2b3257777b7bcb262a821602f9a8ec41422c924717cc957f25a4de7b1c9a967
F92  894a24eb9ab0957f2ded0af613770052002d877b297abef38b77bf796d5836b1
F93  5eaeca6ce8746984f645436767f727e808c8d284b92108cd5c2f9dc88753d2b3

F94, from V2-good: member n, 1e+21; passes
F94  6362d8c83b6384c0997ff13695026e9027b8ae16b1bedfd3247ff60a73378c69

V2-num, from V2-good: member n, 1.5e-7; passes
     b0084cb8b2230a7712fcb600e94d6d3fd9ff7ec1539cd23a2cf16d71eef56290

F95, from V2-good: payload without padding; step 2
F95  94d3120c96819760cf93830638304a5ab8fe5828ca9215050a7edc54745b3785

F96, F97, from V2-good: envelope member NaN or Infinity; step 1
F96  0596668f05aa4aa2a7ad990a507917565e3dc3e90a559bfe860b1343d9ff1de5
F97  180ec51cba15d6281971490ebd643b86ce9c9b31136a98a16fe53749a42c953a

F117, F118, F119, from V1-good: second entry of signatures; step 1
F117 0a42b0123c743fee8dec8166ca449a39a3b955ae885e81467141e48e1902f15a
F118 bba56c5960c48800fcf424e0c69caae65f26e81484b6ef8adee4ff4b0d85f378
F119 ea0321a75a2dfe245fde8678cb7e59f7455fb0b06a24547331d7afd6f634fdb4

F123, from V1-good: S replaced by S plus L; step 3
F123 e4685ea2d1c148c3d00b3ef7f8bc849fb3b9538010bb561d31c04207178b1bc7

F120, from V2-good: predicate a JSON string; step 6
F120 3d73fe2aafed0ad10c9fb83187f7c497508bd9a2b2bb0c03665811885411d0a7

V2-rel, from V2-good: entry of relationships; passes
     4f4d24aff9d1ae7f1f116e2075741617a783bb22fc1d4d7a944b3b0836f546b8

F98 to F107, from V2-good: relationships, one rule each; step 7
F98  71d1058d89a9364669ef2502569e5dfaeb359c68ac91aa8a3d303acc59b42385
F99  aca0769bce7ffa77bbc190e49bd8aae0a9ef3d907646299e75439dbad3f41b0c
F100 10631f19d806aaf9cdd6ab8639d21065decfd994e47380ac2b7a199fcffaf3cc
F101 e2ce032c1485ff2f2b2b2f314f7bdb7e0c5d2df2d2049e937b87570405be753f
F102 0201b71fa66c2db76c77e348519126853e7d5c2a2dc8de69942e7fb8f1a25996
F103 37471a238e99e3889dc3fff4436fb8c60a0985af98a13f8450765294733b859b
F104 a3fb77ab21bfdced86aae20519563494ee45e32ffd4e3bb89c7943e715cd03a3
F105 a687b88fc5832b632cc92c6a7d4c1f37d578cde32412e4abd5bd83b0aa6c2220
F106 df1649155b2134d75b3d30c8ed2823535701d9f841c37b6707c0313e4e170006
F107 eeb0e9fb931fc020bd7a89f759d974819f985a3d4f22777d0fe446132bac257f

F124, from V2-good: rawInputsIncluded a string; step 7
F124 06d03339405182e89c441789f73f47ee2aaec623f286bca6ee3f3f98311a5f9c

F122, from V1-good: expected id gate-2; step 8
F122 cce69757bdd3426f505db28b635cb9ab738bc47cd7ec5c497578ebcd062ae7a5

V2-ref, from V2-good: parametersRef alone; passes
     3f8ba601c4109e27fbd26baf89b52bb4aaaa7e86794cfb3effdedd7c34f75fcb

F125, from V2-ref: with action bytes; step 10
F125 3f8ba601c4109e27fbd26baf89b52bb4aaaa7e86794cfb3effdedd7c34f75fcb

Cases, outcome receipts and relying party, Sections 3.3, 3.4 and 4;
Expected names the step of Section 3.4, the case of Section 3.3 or
the refusal under Section 4:

F89, from V1-good: no approval record entry; Case 1
F89  3ad856b6077a153bafe70b197059a7d1e10e413080ae0a891f116bedae88d9ca

V3-rel, from V3-good: entry of relationships; passes
     39a24a14623ee4f9ecd1caa068dfe0d1f49f46d4d965afbc5bce9ea299a7d0e5

F108 to F114, from V3-good: relationships, one rule each; step 3
F108 cbdef86b33a3b04cf6341937b451d22a7466796a939857c2cee386e3af000851
F109 58bf9bebaa6e603df0706ccfddeadbe5c0323e63ee61b6fe2ec6a43d9a76954c
F110 84ed6e1d1f44d6dfa8d67ee9aa31d19249ef4c731e1330fa0b43f6a137efee96
F111 45c0ca3aa613a1403d52098b7e1c6826fae654717cf0ee03a3abe9612aba40a1
F112 687cc56388b83df77934bc34f85c3ed7e8d961962b291dd6b4425e09e87414fe
F113 5f82e7313f2cd702c39f7c62f99b3b06a346feba4148376d4c7211f53ebe7d4f
F114 de78754372b184f4e8458b16ddb350e3973759aca45de1b86fac6548ec1692ec

V3-seq, from V3-good: sequence, seq 1; passes
     83fa5144543213c7923219e78efa65ea11bf6d9a42eba9353c0ba65c7592d715

F115, from V3-good: seq written 1.0; step 1
F115 62469a5a925eb7d92db340b29d5e5c99d16ded5eaaed65fecec79216f6f2d820

F116, from V3-good: seq written 1e+21; step 3
F116 a897b9da00f85da3e73d5d22f14eff902b4b986d7a7d7fbdb2b22fa9ea2393ef

F121, from V3-good: predicate an empty array; step 2
F121 7ddc903e6397cf88516ca0990f82634dd89cb896654051acf785fbd4ac9e2727

V3-late, from V3-good: performedAt after T; passes
     8786418598d415d88673c12ef0f200376036e00b1b4454056b5a7f0ee2c04076

F10, from V1-good: ap2 and c1, see below; refused under item 1
F10  4069c22d741d9f4a08bc41d53664324f345dc58f88d2ff6211171749fe2b1c40
ap2  37ec7f3fa06f086626268a6c818ba5b5af7fd16ba292735ce964a83c42c0357c
c1   f13fce72c54f1872232c0339fa98b33ffe5142f72481cd31e378bc6a680b6f6a

F90, from F10: received ap, honoured nothing; refused under item 2
F90  4069c22d741d9f4a08bc41d53664324f345dc58f88d2ff6211171749fe2b1c40

V1-accept, from V1-good: consumption state the empty JSON array;
  passes, not refused
     cce69757bdd3426f505db28b635cb9ab738bc47cd7ec5c497578ebcd062ae7a5

Third check of Section 3.5:

K1, from V2-rel: V1-good, V2-num held; passes
K1   4f4d24aff9d1ae7f1f116e2075741617a783bb22fc1d4d7a944b3b0836f546b8

K2, from V2-rel: other targetSubjectDigest; fails
K2   cfb435deb29634a4bb182e3ad7a27c4d6ed442ba003846fe6e6767191556ff0d

K3, from V2-rel: V2-good, V2-num held; fails
K3   4f4d24aff9d1ae7f1f116e2075741617a783bb22fc1d4d7a944b3b0836f546b8

K4, from V3-rel: V1-good held; passes
K4   39a24a14623ee4f9ecd1caa068dfe0d1f49f46d4d965afbc5bce9ea299a7d0e5

K5, from V3-rel: V2-good held; fails
K5   39a24a14623ee4f9ecd1caa068dfe0d1f49f46d4d965afbc5bce9ea299a7d0e5

K6, from V2-rel: predecessor's signature changed; invalid
K6   4f4d24aff9d1ae7f1f116e2075741617a783bb22fc1d4d7a944b3b0836f546b8

K7, from V2-rel: names F30, held; passes
K7   b1bdc7f03d1277924592bd4b7a2816e8a84233b44572eb7a796c429dba0aa93e

K8, from V2-rel: other held signature changed; invalid
K8   4f4d24aff9d1ae7f1f116e2075741617a783bb22fc1d4d7a944b3b0836f546b8

K9, from V2-rel: names V3-good, held; passes
K9   06b298e8a8a3d990ef6aa1069540362a058bc542ed17bdc9254f1d421492c14e

K10, from V3-rel: names V3-good, held; passes
K10  66e4aec0fd020ba1d911bbf300d8d6034aea3e43dd658f31a0c445fe3f52e5ee

K11, from V2-rel: V1-good under the observer-1 key; invalid
K11  4f4d24aff9d1ae7f1f116e2075741617a783bb22fc1d4d7a944b3b0836f546b8

K12, from V2-rel: chain of three, signature changed; invalid
K12  d7dab97bace7752ccbe9ed7d99591021a2eece45c125b2d923022ee21ae2c4ab

K13, from V2-rel: chain of three, V1-good absent; fails
K13  d7dab97bace7752ccbe9ed7d99591021a2eece45c125b2d923022ee21ae2c4ab

K14, from V2-rel: chain of three, all held; passes
K14  d7dab97bace7752ccbe9ed7d99591021a2eece45c125b2d923022ee21ae2c4ab

K15, K16, from V2-rel, V3-rel: predicate held not an object; invalid
K15  1094df61875f0dce09fbc40258307d9503132b91421c27c90de3e82630a9b5ce
K16  77ec7b9349e98f9c54c01ffc3981941019d6b59df011a6039e6ed2309f60bc27

K17 to K22, from V2-rel, V3-rel: relationships held malformed;
  invalid
K17  7767687021c5b2e6459e5cf3267b804306fd414987b23fda863961047e96df7e
K18  0f7bd6ab2fd159195f7bf0486d2b712494976baae5a98f3b9dbca83854bd970a
K19  d958e9d21d2100c168107018d50c08d6e0766e162d8769ef24737e4d7bd895ae
K20  da06b9567f39374c89e3f225a946c6c0503cd85b9209cbfe8587a4ae290b5c6e
K21  feaaeee4a0f8737f76ff3bcd0697f778a189572a75e80309e499a315eb486005
K22  151eb1cac4acede5eb26e7e008595997d879c29f1d24216fd5e408a4895f92a4

K23, K24, K25, from V2-rel: the same, named by no entry; invalid
K23  4f4d24aff9d1ae7f1f116e2075741617a783bb22fc1d4d7a944b3b0836f546b8
K24  4f4d24aff9d1ae7f1f116e2075741617a783bb22fc1d4d7a944b3b0836f546b8
K25  4f4d24aff9d1ae7f1f116e2075741617a783bb22fc1d4d7a944b3b0836f546b8

K26, from V2-rel: F120 held, named by no entry; invalid
K26  4f4d24aff9d1ae7f1f116e2075741617a783bb22fc1d4d7a944b3b0836f546b8

K27, from V2-good: V2-rel held, V1-good not; passes
K27  3e3b7d575b712ada2c813ee98fd3b0c80d788e2558444a34338510c9978fa47a

K28, from V2-rel: second supersedes, not held; fails
K28  25579db4807dd8cc2f296a6e2a9aa8afb888fdc1bbf30c468a9a37c3229d131c

K29, from V2-rel: corrects, target not held; passes
K29  d329f5d6d54c1f604010128749dcc9d2a555fa500e616e9e4e75b77d75d5a7a5

Second check of Section 3.5:

S1, from V3-good: ESCALATE decision and its executed outcome held;
  fails
S1   436fb6f6f19cecbd67df9f468556807eab5003ddb8eb60c8b9c2f708e3e9bb66

S2, from V3-good: V3-decision and V3-good held; passes
S2   436fb6f6f19cecbd67df9f468556807eab5003ddb8eb60c8b9c2f708e3e9bb66
Figure 4: Further vectors and the SHA-256 of their files

No vector separates the equation with the cofactor from the one without. F89 illustrates rejection despite the presence of evaluation evidence, but does not establish that a verifier enforces the second half of the rule of Section 2.6. F17 fails at the verdict check of Case 3 whatever its evidenceRefs hold and only illustrates that rule. V1-good and V2-rel decide different actions, and a DENY changes no state (Section 2.2), so K1 to K5 show only how the third check matches digests, not a sequence of states of one action (Section 2.1).

F10 passes receipt verification and is refused by the relying party under Section 4, with c1 under item 1, because item 2 makes ap2, a copy of ap with other metadata, the same approval. The proofbundle 6.4.0 command receives no consumption state, and its acceptance of F10 does not test consumption-state enforcement. Appendix A defines no consumption scope, so Section 3.5 gives no result for F10; within one scope, V1-good and F10 would form a pair that its first check reports.

Appendix B. Changes

This section is to be removed before publication.

B.1. Changes from -08

Each paragraph of this section describes one change from -08.

Appendix A.5 says again, as -07 did, that F38 is checked with the expected audience observer-2, which its receipt does not name. -08 said that every vector is checked with the expected audience observer-1, which does not match the test package.

Section 5.5 says again, as -07 did, that the earlier test package of revision -06 was not published; -08 left this out.

Section 2 says that every decision receipt names, in principal.id, the party on whose behalf the agent acted.

Section 2.4 says that a record made after the interaction can name the decision by its content root, which is the same for every copy of the payload, and that this document does not require decisionId and validity.nonce to be unique beyond the exchange in which the decision is made.

Section 5 is shorter and keeps every implementation it names. Among the details it leaves out are, in Section 5.1, the file that holds the licence of proofbundle; in Section 5.2, the time window of an approval, the order of the signed ALLOW and the hook's record of its use, what the pilot writes into inputSnapshot, traceContext and actionOutcome, how a full record is rotated, the cases in which the service and the hook refuse, and the state at an earlier commit; in Section 5.3, how the Go verifier applies the preliminary checks of the third check of Section 3.5; and in Section 5.4, that the forms were registered twice with fresh keys, that the receipt of each accepted form verified with the key set that the service served, and where the twelve statements came from.

Section 5.1 says that the decision verify and outcome verify commands of proofbundle check receipts of the predicate types of Section 1.2 and gives the exit codes they returned for the vectors, in place of saying that proofbundle covers those checks.

Section 5.3 says that the verifier of the test package implements checks of the sections it names and that the test package shows its results for the vectors it holds, not that every condition of those sections is implemented.

Section 7 says, under Clocks, that validity.expiresAt, recordedAt and the times of an approval record are statements of their signer, which of them the checks of step 9 of Section 3.2 and of Case 1 compare, and that these checks do not show when a decision was made or when a person approved.

Section 7 says that vector V2-bad tests one of the payloads that step 4 refuses.

Appendix A.2 no longer prints the Statements of V2-good, V3-decision, V3-good and V3-bad, and Appendix A.3 gives V1-good and V1-bad in full and names the other five vectors in Figure 3, with their expected results and the SHA-256 of their files in the test package.

Appendix A.5 summarizes F10 and ap2, describes c1 and identifies their files in the test package by SHA-256 in place of printed blocks, leaves out the explanation of items 1, 4 and 5 of Section 4 on the vectors, and lists the further vectors in Figure 4, with the SHA-256 of each file, in place of four tables.

Appendix A.5 says that F19 to F71 were added for requirements that F1 to F18 leave without a vector, and that the requirements without a vector include, besides the six it names, the first check of Section 3.5 and the conditions of item 6 of Section 4.

This appendix lists the changes from -08 before those from -07 and gives each change from -07 as one item of a list.

The test package is the one of -08, unchanged.

B.2. Changes from -07

Each item of this list names one change from -07.

  • Test package: vectors of -07 kept, new ones in expected-08.
  • Section 2.5, Section 3.6: a pass shows no authority at the time of the action.
  • Section 3.1: evidence named without its issuer.
  • Section 1.2, Section 3.2, Section 3.5: three clarifications.
  • Section 3.5: predecessor lookup follows the chain only; K27.
  • Step 8 of Section 3.2: expected decisionMaker.id; F122.
  • Section 3.5: a failing entry of a predecessor at any depth fails the third check; K13.
  • Step 3 of Section 3.2: no cofactor, S below L; F123.
  • Section 3.5: each supersedes entry checked on its own.
  • Section 4: re-release of a signed ALLOW is no new honouring.
  • Section 4: a copy or new signature of a released signed ALLOW is no new honouring.
  • Section 3.5: the third check does not require a predecessor of the same action.
  • Go verifier: names the step of Section 3.4 where a held outcome receipt fails the preliminary checks of Section 3.5.
  • Appendix A.5: adds V1-accept.
  • Package README: forms of consumption state and copies.
  • Package README: F95 to F125 in one range, shorter lines.
  • Section 3.5: first and second checks count receipts that pass the preliminary checks, also after expiry.
  • Section 8: approver and principal.id may be pseudonymous.
  • Section 2.5, Section 2.7: [YOSSIF] as a possible source of evidence.
  • Section 2.5 names [LIU] for the same limit of an approval record that an authorization server signs, and the cases of [DPS-RAR] that show it.
  • Section 6: [RFC9396]; a receipt is no authorization grant.
  • Section 3.6: a decision maker can state unchecked authority in notChecked, as its own statement.
  • Section 2.5: no absence over a time window shown.
  • Section 5.5: version, licence and link of dr_check.
  • Section 3.3: checks of a case follow step 10.
  • Appendix A.5: further requirements lack a vector.
  • Section 1.2: consumption state and further terms.
  • Section 3.1: the time T of a post-hoc Auditor.
  • Section 7: only Ed25519 and SHA-256.
  • Abstract: shorter, same statements.
  • Appendix A.5: vectors without the revision adding them.
  • Section 2.2: verdict DENY or REFUSE.
  • Section 5.3: names the Go verifier.
  • Section 6: paragraph on [KRAUSZ] in three sentences.
  • Section 3.4, step 5: outcome-signing key named, same meaning.
  • Section 3.3, Case 3: checks in the order they run.
  • Section 3.8: other words, same meaning.
  • Section 4: an approval in Case 1 covers one action and can be used once; Section 7: an ALLOW for one action must not be usable for another.
  • Section 3: time comparison stated once; step 9 shorter.
  • Appendix A.5: a vector need not cover each condition.
  • Section 5.2: shorter, same observations.
  • Section 3.5: held receipts and preliminary checks first.
  • Section 3.8: every type defined in one place.
  • Section 3.5, second check: any verdict other than ALLOW.
  • Section 2.2: DENY and REFUSE treated alike.
  • Test package: adds F124, V2-ref, F125, V3-late, K28, K29.
  • Go verifier: second check of Section 3.5; S1 and S2.
  • Appendix A.5: renamed, further vectors by group.
  • Appendix A.3: V3-bad fails at step 4 of Section 3.4.
  • Appendix B of -08: changes from -07, with earlier changes listed in -07.

B.3. Earlier Changes

Appendix B of revision -07 of this document lists the changes made in revisions -01 to -07.

Acknowledgments

Toshikatsu Oga (The HORIZONs Co., Ltd.) checked the test vectors of several revisions with an implementation written from the text and reported places that read two ways.

The author used AI tools to help draft this document and to generate and cross-check its examples. The author reviewed the entire text and is responsible for it.

Citing adjacent work in the References does not imply that its authors endorse this document.

Author's Address

Konrad Gruszka
b7n0de
Germany