Network Working Group K. McGuinness Internet-Draft Independent Intended status: Standards Track 19 November 2026 Expires: 23 May 2027 Mission-Bound Runtime Enforcement: AuthZEN Profile draft-mcguinness-mission-authzen-latest Abstract Mission-Bound Runtime Enforcement defines a substrate-independent decision contract: before each consequential action runs, a Policy Enforcement Point (PEP) obtains a permit from a Policy Decision Point (PDP) that evaluates the action against the established Mission. This specification defines the OpenID AuthZEN profile of that contract. Its Decision Base maps the decision inputs onto the AuthZEN Authorization API request, shapes permit and denial responses, and binds every runtime failure condition to a wire- visible identifier, so two implementations interoperate on decisions with the base alone. Six optional feature profiles compose on the base: Transaction Assurance, Runtime Evidence, Obligations, the Access Request and Approval Profile (ARAP), History, and Batch. About This Document This note is to be removed before publishing as an RFC. The latest revision of this draft can be found at https://mcguinness.github.io/mission-bound-authorization/draft- mcguinness-mission-authzen.html. Status information for this document may be found at https://datatracker.ietf.org/doc/draft- mcguinness-mission-authzen/. Source for this draft and an issue tracker can be found at https://github.com/mcguinness/mission-bound-authorization. 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 23 May 2027. Copyright Notice Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved. This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/ license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License. Table of Contents 1. Introduction 2. Profiles and Capabilities 3. Conventions and Terminology 4. Mission-to-Policy Materialization 5. PDP Request 5.1. AuthZEN Envelope Mapping 5.2. Operation Target 5.3. Resource Contract 5.4. Mission Decision Context 5.5. Actor Decision Context 5.6. Credential Decision Context 5.7. Action Parameters and Parameter Digest 5.7.1. Idempotency Claim 5.8. Action Phase Context 5.9. Compensation Link Context 5.10. Evaluation-Context Binding 5.11. Audience and Mission State Observation 5.11.1. Observation Modes 5.12. External-Communication Context 5.13. Taint Context 5.14. Protected-Data Context 5.15. Work-Item Context 5.16. Approval Context 5.17. History Context 5.18. Canonical Projections 5.19. PDP Request Examples 5.20. PDP-Side Consistency Checks 5.21. Clock Skew 5.22. Batch Evaluations 5.23. Timeouts, Retries, and Overload 6. PDP Response 6.1. Response Decision Context 6.1.1. Decision Conditions 6.2. Obligations 6.2.1. Obligation Processing 6.2.2. Step-Up Obligation 6.2.3. Further Obligation Types and Advice 6.3. Permit Response 6.4. Denial Response 6.5. Runtime Denial Classification 6.6. Requestable Denials 6.7. Choosing the Lane 6.8. Failure-Condition Coverage 6.9. Permit Binding in Split Topologies 6.10. Evaluation Identifier Propagation 7. Runtime Evidence 7.1. Evaluation Request Digest Input 7.2. Execution Evidence Requirement 8. Conformance 8.1. Capability Discovery 9. Security Considerations 9.1. Unbound-Evaluation Downgrade 9.2. Access Request Service in the Trusted Base 9.3. Denial Oracle 9.4. Materialized View Fidelity 9.5. Transport 10. Privacy Considerations 11. IANA Considerations 11.1. AuthZEN Policy Decision Point Capabilities Registration 11.2. HTTP Field Name Registration 12. References 12.1. Normative References 12.2. Informative References Appendix A. Document History Acknowledgments Author's Address 1. Introduction Mission-Bound Runtime Enforcement [I-D.draft-mcguinness-mission-runtime] (the "runtime profile") defines the decision contract for Mission-bound authorization: action classification, PEP placement, parameter binding and the time-of- check to time-of-use gap, the consumption-bound failure posture, failure modes, runtime evidence, and conformance scope. It leaves the decision API wire format to the deployment. This document defines that wire format for the OpenID AuthZEN Authorization API [AUTHZEN]: which AuthZEN members carry each decision input, how permits and denials are shaped, and which identifier surfaces each failure condition. It carries only the AuthZEN deltas. It does not redefine which actions are consequential, where the PEP sits, the semantics of parameter binding, the failure modes, or the runtime conformance scope; the runtime profile defines them normatively. This profile inherits the runtime profile's substrate requirements. The decision contract is defined against the Mission model, not against OAuth 2.0 mechanics: OAuth enters only through the credential-derived decision inputs (the token's sub, client_id, cnf, authorization_details, and mission claim), which the substrate's Mission-bound credential supplies. The examples use the OAuth binding, Mission-Bound Authorization for OAuth 2.0 [I-D.draft-mcguinness-oauth-mission]. A deployment on another Mission substrate maps that substrate's credential to the same inputs and uses this profile unchanged. The Decision Base is sufficient for two implementations to interoperate on decisions alone. It defines: * how the materialized policy view is referenced on the wire through its policy_view_id (Section 4); * how the decision inputs map onto the AuthZEN subject/resource/action/context envelope, with worked requests and the PDP-side consistency checks (Section 5); * the permit and denial responses, and the rules that keep a permit bound to the PEP, channel, and inputs it was issued for (Section 6); * the denial classification and the complete mapping of the runtime profile's failure conditions onto wire-visible identifiers (Section 6.5, Section 6.8). Six optional feature profiles compose on the base (Section 2): * *Transaction Assurance*: the parameter-digest and idempotency-key members, and the high-consequence permit's use_limit condition (Section 5.7, Section 6.1.1); * *Runtime Evidence*: emission of the runtime evidence companion's Decision Evidence, Execution Evidence, and Refusal Record, and the response members this profile echoes into them (Section 7); * *Obligations*: mandatory PEP work under an existing decision (Section 6.2); * *ARAP*: requestable denials through the AuthZEN Access Request and Approval Profile [ARAP] (Section 6.6); * *History*: policy-selected predicates over prior Decision and Execution Evidence (Section 5.17); * *Batch*: evaluations over the AuthZEN evaluations endpoint (Section 5.22). Two companions extend the request without changing the base. Mission Capability Binding ([I-D.draft-mcguinness-mission-capability-binding]) adds capability- source context and establishes the action identity this profile consumes. The experimental metering companion ([I-D.draft-mcguinness-mission-metering]) defines the AuthZEN representation of cumulative consumption metering, including the settlement exchange and duration-lease renewal. The end-to-end flow is: Agent PEP PDP Access Request Service | | | | |- action ->| | | | | validate token | | | |- evaluation -->| | | | request | decide vs Mission | | |<- permit ------| | | | (+ context) | | | | execute | | | |- Execution --->| | | | Evidence | commit / release | |<- result -| | | | | | | | |<- deny --------| | | | (+ access_request) | | |- submit access request ------------->| | |<--------------- approval ------------| | |- re-evaluate ->| | 2. Profiles and Capabilities This profile is a Decision Base plus six feature profiles. Each feature profile depends on the Decision Base alone; none depends on another: Decision Base +-- Transaction Assurance +-- Runtime Evidence +-- Obligations +-- ARAP +-- History +-- Batch Feature profile selection is issuer or deployment policy plus authenticated capability discovery, never a requester's own choice (Section 8.1). A PDP or PEP that does not support a profile a policy or action class requires denies rather than falling back to the base, and a requester cannot omit a required profile to downgrade enforcement. A feature profile never widens what the runtime profile's own rules already require for a class. Where the runtime profile makes a property mandatory for a class, the corresponding feature profile is the wire vehicle for that requirement, not a free election ([I-D.draft-mcguinness-mission-runtime]). Each feature profile states its prerequisites, roles, required members, the behavior when it is absent, and the assurance it enables. Transaction Assurance: * Prerequisites: the Decision Base. * Roles: the PEP supplies parameter_digest and, where the Operation Profile requires one, idempotency_key; the PDP binds and claims them; the executing PEP reverifies. * Required members: action.properties.parameter_digest and, conditionally, action.properties.idempotency_key (Section 5.7); the response conditions.parameter_digest and, for the high- consequence classes, conditions.use_limit: 1 (Section 6.1.1). * Absence: for the parameter-bound and high-consequence classes, the runtime profile makes exact parameter binding and single use mandatory, so a PEP and a PDP MUST support this feature profile for those classes and the PDP MUST advertise its capability URN (Section 8.1). The Decision Base alone does not conform for those classes, and a PEP whose PDP does not advertise the URN fails closed. The Decision Base checklist repeats the conditional enforcement duties these members carry (Section 8); the capability URN declares support for this feature profile's complete applicable contract. Outside those classes, it is optional. * Assurance: exact operation and parameter binding with replay and duplicate-execution protection (the runtime profile's transaction-assurance tier). Runtime Evidence: * Prerequisites: the Decision Base. * Roles: the PDP and PEP are producers of Decision Evidence, Execution Evidence, and Refusal Records; an auditor or relying party is the verifier ([I-D.draft-mcguinness-mission-runtime-evidence]). * Required members: none beyond what the Decision Base carries on the wire; the records are populated from the base's request and response members (Section 7). * Absence: the runtime profile's baseline still requires an evidence record for every decision ([I-D.draft-mcguinness-mission-runtime]). A Decision-Base-only deployment may satisfy that with its own internal record, never with this feature profile's portable schema, and MUST NOT represent an internal record as the portable schema. A deployment composes this feature profile when it needs decision or execution evidence portable to a verifier it does not operate. * Assurance: portable, independently verifiable Decision, Execution, and Refusal evidence (the evidence requirement of the Runtime-Enforced Mission Assurance Level, [I-D.draft-mcguinness-mission-architecture]). Obligations: * Prerequisites: the Decision Base. * Roles: the PDP attaches an obligation; the PEP fulfills it or fails closed (Section 6.2). * Required members: the response obligations array. * Absence: a PDP that never attaches an obligation composes nothing. Fail-closed handling of an obligation the PEP does not recognize or cannot fulfill is a Decision-Base rule regardless. This feature profile adds specific obligation types (the step-up obligation) and the supported_obligations advertisement. * Assurance: mandatory PEP-side work under an existing decision, with no second grant and no governance adjudication. ARAP: * Prerequisites: the Decision Base; an Access Request Service in the trusted base (Section 9.2). * Roles: the PDP marks a denial requestable; the PEP submits the access request and re-evaluates; the Access Request Service adjudicates (Section 6.6). * Required members: the response access_request, next_action, and retry_after; the request context.approval on a fresh evaluation. * Absence: a deployment that does not compose ARAP never marks a denial requestable; the denial is terminal for the attempted action, as under the Decision Base. * Assurance: demand-driven, runtime-initiated authority requests without a standing grant. History: * Prerequisites: the Decision Base; the runtime profile's runtime.history-input capability ([I-D.draft-mcguinness-mission-runtime]). * Roles: the PDP evaluates a policy-selected predicate against its evidence store; the PEP optionally supplies evidence_reference or attribute_values in context.mission_history (Section 5.17). * Required members: the optional request object context.mission_history and the Decision Evidence mission_history extension, which the PDP computes. * Absence: a deployment that does not declare history evaluation in its Enforcement Scope Statement never receives a history- dependent denial. Where deployment or Resource policy requires a history predicate, the PDP still fails closed under the runtime profile's rule when this feature profile is absent. * Assurance: sequence-aware decisions, where an action's authorization depends on where the undertaking stands, not only on the request in isolation. Batch: * Prerequisites: the Decision Base. * Roles: the PEP submits a boxcar; the PDP evaluates every item under execute_all semantics (Section 5.22). * Required members: the AuthZEN evaluations array; this feature profile defines no member of its own. * Absence: a deployment that never batches submits one evaluation per action, at the same enforcement strength. * Assurance: transport amortization only, never a change in per- item enforcement semantics. 3. Conventions and Terminology 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. Two member-presence conventions extend that vocabulary: CONDITIONAL: The member is present when the stated condition holds and absent otherwise. The condition accompanies each member so marked. REQUIRED when known: The sender includes the member when it holds the member's value. A receiver MUST NOT infer meaning from the member's absence. JSON [RFC8259] is the data model for all PDP requests, responses, and evidence objects. Where this document computes an integrity hash, it uses JCS [RFC8785] under the canonicalization rules of [I-D.draft-mcguinness-oauth-mission] and defines no second canonicalization. "SHA-256" refers to [RFC6234]. A digest is encoded in the integrity-anchor encoded form of [I-D.draft-mcguinness-oauth-mission]: sha-256: followed by the base64url, no-padding encoding of the digest. The canonical projections of Section 5.18 are canonical-object digests under the substrate's default commitment construction, which this document imports normatively ([I-D.draft-mcguinness-mission-substrate]). This document uses the following terms as defined in [I-D.draft-mcguinness-mission-runtime]: Policy Enforcement Point (PEP), Policy Decision Point (PDP), consequential action, Resource policy, decision, Mission state source, enforcement scope, high- consequence classes, parameter-bound, Operation Profile, Enforcement Scope Statement, materialized policy view, trusted compiler, and the action-class names (consequential read, consequential write, irreversible action, external commitment, and privileged administration). It also uses: * Mission reference (id, issuer): a binding's Mission Reference identifier and Controller namespace, in the representation that binding defines. The OAuth realization is the Mission claim of [I-D.draft-mcguinness-oauth-mission]. * Authority Set, Approver, and the authority_hash integrity anchor, as defined in [I-D.draft-mcguinness-oauth-mission]. * mission_resource_access: the authorization_details type of the Mission Resource Access Profile ([I-D.draft-mcguinness-oauth-mission-resource-access]). * Effective Authority Set: as defined in [I-D.draft-mcguinness-oauth-mission-status]. * Decision Evidence, Execution Evidence, and Refusal Record: the records the PDP and PEP emit for a decision, an execution outcome, or a pre-decision refusal, as defined in [I-D.draft-mcguinness-mission-runtime-evidence]. HTTP message examples follow [AUTHZEN] for the decision request and response, and [RFC9457] for problem-details error bodies where a deployment carries them outside the AuthZEN envelope. 4. Mission-to-Policy Materialization Where the PDP evaluates a Mission against an action through a materialized policy view, the runtime profile defines the view, its trusted-compiler and reproducibility rules, its bounded-fidelity property, and the content-addressed policy_view_id with its mission- policy-view integrity envelope ([I-D.draft-mcguinness-mission-runtime]). A PDP that evaluates the Mission's recorded authority directly meets that profile's authority input (its Section "Authority") without a view, and its requests omit policy_view_id. The trusted compiler computes the view's envelope, and the authority_hash it binds, from its own held Mission record. The OAuth binding's baseline mission claim does not carry authority_hash ([I-D.draft-mcguinness-oauth-mission]), so the consistency check between a decision request and the loaded view is an equality test on mission_id and issuer alone, with policy_view_id as the optional, stronger content check where the requesting PEP holds it (Section 5.20). This profile carries policy_view_id only in the request, at context.mission.policy_view_id (Section 5.4). The PDP does not echo it on the response. Where the PDP evaluates through a view, it records that view in Decision Evidence ([I-D.draft-mcguinness-mission-runtime-evidence]); evaluation_id (Section 6.1), not a view identifier, correlates a permit or a denial with its evidence. The PDP need not be remote. A PDP embedded in or colocated with its PEP, evaluating against a loaded materialized policy view, is a conforming deployment; the decision's network cost is then paid per freshness window rather than per action. This profile does not define a policy-language wire form for the materialized view. Implementations MAY use canonical input bundles the AuthZEN PDP consumes directly, or an engine-native artifact. Compiling a Mission into an engine-native policy artifact and standardizing a policy-view carriage format are out of scope ([I-D.draft-mcguinness-mission-runtime]). 5. PDP Request The PDP request carries the runtime profile's decision inputs in the OpenID AuthZEN Authorization API envelope [AUTHZEN], per the AuthZEN request model: action-scoped members in action.properties, resource- scoped members in resource.properties, and evaluation-wide members in context. This profile does not change which inputs the PDP evaluates; the runtime profile defines them. This profile is used after ordinary access-token validation under [I-D.draft-mcguinness-mission-runtime]: the PEP MUST NOT ask a PDP to authorize an action from unverified token claims, and the PEP-PDP channel MUST be integrity-protected and mutually authenticated as that profile requires. 5.1. AuthZEN Envelope Mapping +==========+=======================================================+ | AuthZEN | Mission-bound content | | member | | +==========+=======================================================+ | subject | The principal the decision is requested for. | +----------+-------------------------------------------------------+ | resource | type/id carry the fine-grained object identity the | | | action names (for example, a specific journal entry), | | | for Resource-policy evaluation; properties carries | | | resource-scoped members, including operation_target | | | (Section 5.2), audience (Section 5.11), and, where a | | | resource contract applies, resource_contract | | | (Section 5.3). Only operation_target is matched | | | against an entry's resource. | +----------+-------------------------------------------------------+ | action | name is the requested action identifier (for example, | | | journal-entries.write), which the PDP evaluates | | | against the approved actions per | | | [I-D.draft-mcguinness-oauth-mission]; properties | | | carries action-scoped members (Section 5.7). | +----------+-------------------------------------------------------+ | context | Carries the Mission-bound, evaluation-wide members | | | defined below. | +----------+-------------------------------------------------------+ Table 1 The runtime profile's Authority Set check matches the action's resource and action identity against an entry's resource and actions ([I-D.draft-mcguinness-mission-runtime]). For a mission_resource_access entry, the action's resource is the operation target in resource.properties.operation_target (Section 5.2). A PDP MUST match each such entry's resource against that target under [I-D.draft-mcguinness-oauth-mission-resource-access], Section "Resource Matching", in the credential's authority (Section 5.6), in the current effective authority ([I-D.draft-mcguinness-mission-runtime]), and in any other mission_resource_access authority bound it applies. A PDP MUST NOT match an entry against resource.properties.audience, resource.type or resource.id, or substitute any of them for the operation target. Audience validation (Section 5.11) and Resource-policy evaluation of type and id remain separate checks, and the action, constraint and delegation checks remain independent of the target match. The AuthZEN subject is the Subject the Mission's authority is exercised for. It does not change under delegation: the acting agent's client_id and any act delegation chain are carried in context.actor (Section 5.5), never in subject. The PDP binds the permit to subject together with the actor context, and the confused- deputy check ([I-D.draft-mcguinness-mission-runtime]) re-verifies that the action is for the same Subject it was authorized for. subject.type is user unless the deployment profiles another value. subject.id is the credential's authenticated subject identifier (the OAuth realization is the token's sub claim). subject.properties.iss is REQUIRED when known (Section 3), carrying the issuer that authenticated the Subject, so a sub is disambiguated across issuers; a PEP that cannot establish the Subject's issuer omits it. 5.2. Operation Target operation_target: REQUIRED for an action evaluated against mission_resource_access authority. A string, carried at resource.properties.operation_target, holding one absolute URI (Section 4.3 of [RFC3986]): the operation target of [I-D.draft-mcguinness-oauth-mission-resource-access], Section "Resource Matching", which identifies the resource the operation acts on. The authenticated PEP establishes the target from the trusted operation contract of the resource it enforces for and from the operation's validated arguments, as that section requires of a resource server. The PEP MUST NOT derive the target from the credential's entries, the Mission's entries, or a target value the caller asserts. A PEP that cannot establish one unambiguous target MUST refuse the operation before requesting a decision (Section 6.8). The PDP checks the target's syntax (Section 5.20) and passes the received string to the matcher unchanged: an exact match compares that string, never a parsed and reserialized form, and a prefix match applies only the comparison normalization of [I-D.draft-mcguinness-oauth-mission-resource-access]. The operation target is part of the operation identity, the authorization binding and the cache key (Section 5.18). A permit is therefore never reused for another target, even when the audience, the action, resource.type and resource.id are equal, and the PEP executes the operation only on the target the permit was evaluated for (Section 6.9). 5.3. Resource Contract resource_contract: CONDITIONAL. An object, carried at resource.properties.resource_contract: the id, version, and digest of the resource contract under which the PEP will perform the operation. REQUIRED when a resource contract applies to the action ([I-D.draft-mcguinness-mission-runtime], Section "Resource Contract"). The PEP establishes the reference from the applicable entry's reference, resolved against its own trusted contract declarations, and never from a value the caller asserts. The PDP MUST deny the action with resource_contract_drift unless the reference resolves under the PDP's own trusted declarations and equals, member for member, the reference of an applicable entry that authorizes the action. An absent member where a contract applies is denied the same way. The reference is part of the authorization binding and the cache key (Section 5.18), so a permit is never reused under another contract. Because it is a member of the resource object, it is also part of the approved Resource an ARAP approval binds (Section 5.16). 5.4. Mission Decision Context The mission member identifies the governance record and its current materialized view. It carries authority identity only: current Mission state and its freshness are carried separately, in context.mission_state_observation (Section 5.11), so a change in state never mints a new policy_view_id (Section 4). id: REQUIRED. A string. The Mission Reference identifier, in its binding's representation (the OAuth realization is the Mission's id). issuer: REQUIRED. A string containing a URI. The Mission Reference's Controller namespace, in its binding's representation (the OAuth realization is the Mission's issuer). It is not the credential's issuer, which context.credential.issuer carries (Section 5.6). authority_hash: OPTIONAL. A string. The Authority Set integrity anchor, in the integrity-anchor encoded form ([I-D.draft-mcguinness-oauth-mission]). The OAuth binding's baseline mission claim and default introspection projection do not carry it. A PEP includes it only where it holds it under the Local Approved-Set Verification profile ([I-D.draft-mcguinness-oauth-mission-approved-set-verification], Section "Local Approved-Set Verification") or introspection's disclosure privilege ([I-D.draft-mcguinness-oauth-mission]). The mandatory identity check (Section 5.20) does not depend on it; a PDP that receives it and independently verifies it MAY use it as a further audit correlator, never as a substitute for that check. policy_version: REQUIRED when known. A string. The policy_version recorded at the approval event. It is a Mission-record field ([I-D.draft-mcguinness-oauth-mission]) not carried on the mission claim or the introspection projection, so a PEP that is not co- located with the Mission record may not have it; such a PEP omits it and relies on policy_view_id for view correlation. policy_view_id: OPTIONAL. A string. The materialized view identifier (Section 4). The PDP is authoritative for the current view, so a PEP need not supply it; a PEP that has the value supplies it, and the PDP checks it as Section 5.20 specifies. subject: OPTIONAL. A closed object with exactly iss and sub: the Mission's origin principal, value-unchanged from the Mission record, per the cross-domain profile's Origin Principal rules ([I-D.draft-mcguinness-oauth-mission-cross-domain]). REQUIRED for a request claiming that profile. The PEP populates it only from a verified token or delegation chain, or from trusted local token metadata, never from an unverified request value. The PDP relates it to the request subject only through the origin-principal mapping check (Section 5.20), never by direct comparison. The AuthZEN subject (the authenticated local token subject), this member (the origin principal), and context.actor (the acting lineage) remain distinct roles even when two identifiers are equal. id and issuer anchor the Mission's approved Authority Set, resolved against the materialized policy view the PDP loads for that Mission (Section 4); authority_hash, where a PEP carries it, is an audit correlator, not the anchor. Where a Mission runs a narrowing mechanism, the PDP evaluates the request against the Effective Authority Set ([I-D.draft-mcguinness-oauth-mission-status]), not the approved set alone, per the runtime profile's Authority decision input ([I-D.draft-mcguinness-mission-runtime]). The Containment profile is one such mechanism ([I-D.draft-mcguinness-oauth-mission-containment]). This document defines no member carrying the Effective Authority Set's evaluated state. A companion profile MAY extend the mission context with the member it needs, by specification. An extension member name MUST be a collision-resistant name or a name coordinated within this document family, as for denial reasons (Section 6.5), and a PDP that does not recognize an extension member does not apply the subtraction carried only by that member. A denial produced against the Effective Authority Set carries its narrowing mechanism's own denial reason, not out_of_authority. For example, the Containment profile's authority_contained denies capability the Approver granted and the issuer later removed, a distinct history from capability never approved ([I-D.draft-mcguinness-oauth-mission-containment]). 5.5. Actor Decision Context The actor member carries the authenticated actor context when delegation is in effect, reconstructed from the credential's actor- delegation chain and authenticated client identity (the OAuth realization is the access token's act claim and its authenticated client identity, [I-D.draft-mcguinness-oauth-mission]): client_id: REQUIRED when known. A string. The authenticated client identity. client_instance_id: OPTIONAL. A string. A deployment-defined client-instance correlator when the PEP can establish one. act: OPTIONAL. An array of objects. The delegation chain projection, ordered root to leaf. For a single actor, the array has one member. The actor member carries the delegation chain only. Provenance beyond the delegation chain (the tool a request invoked, a named workflow step, a human approver) MUST NOT be encoded inside the act chain; the PDP evaluates the act chain as defined by the runtime profile, and provenance is recorded in dedicated evidence fields where the deployment captures it. Where tokens carry Instance Context ([I-D.draft-mcguinness-oauth-client-instance-id]), the PEP can set client_instance_id from it once it has validated that context as a Context Consumer (Section 7.5 of [I-D.draft-mcguinness-oauth-client-instance-id]) and established its association with the current presenter. The PEP MUST NOT populate context.actor.client_instance_id from context that establishes only upstream instance participation. The correlator represents the pair (iss, id) without collisions across the Instance Context Authorities accepted by the PDP and its evidence consumers, including when separate PEPs each accept only one authority. Attributing the presentation requires the association with the presenter that Section 7.5 of [I-D.draft-mcguinness-oauth-client-instance-id] establishes: a sender-constraint key unique to the instance (Section 7.3 of [I-D.draft-mcguinness-oauth-client-instance-id]) where the token issuer conveys context only from direct Client Attestation validation, and authenticated provenance where it may preserve context from an input token. Decision Evidence then records which instance acted, and Execution Evidence does through the evaluation_id link, without new members. 5.6. Credential Decision Context The credential member carries credential-derived facts the PEP has already validated and that the PDP needs to enforce the runtime decision's time, issuer, sender-constraint, and credential authority checks: issuer: REQUIRED when known. A string containing a URI. The credential issuer. expires_at: REQUIRED when the credential carries an expiry. An RFC 3339 [RFC3339] timestamp corresponding to the credential's expiry (the OAuth realization is the token's exp claim, [I-D.draft-mcguinness-mission-runtime-oauth]). confirmation: OPTIONAL. An object. A sender-constraint confirmation value or digest of that value, included only after the PEP has verified the proof-of-possession check for the presented credential. authority: REQUIRED. An array of authorization_details entries: the authority the verified credential itself carries, the credential authority of [I-D.draft-mcguinness-mission-runtime]. For a Mission-bound token it is the token's own entries; for an ordinary token joined to a Mission it is the authority that token carries as issued, as the join profile establishes it ([I-D.draft-mcguinness-mission-runtime-oauth]). The PEP MUST NOT include unverified credential claims in this member. The PDP evaluates the action against authority independently of the Mission entry it matches: * The action MUST fall within one entry of authority as a whole: its resource, its actions, and every constraint it carries. * The PDP MUST NOT substitute the Mission's Authority Set, or any other record of Mission authority, for authority. * A condition on that entry that the PDP cannot establish from its own state does not permit, and an action-approval requirement on it applies as one on the matched Mission entry does (Section 5.16). * An action that no entry of authority names by resource and action is denied out_of_authority. An action that entries name, but whose parameters violate a constraint on each of them, is denied parameter_violation, as on the matched Mission entry (Section 6.5). Decision Evidence's authority_bound names which bound decided either deny (Section 7). * A request whose authority is absent, or carries an entry the PDP cannot evaluate in full, is denied credential_invalid, and the PDP MUST NOT fall back to the Mission's authority. This profile deliberately requires the credential's entire authority description to be evaluable, so an unevaluable credential entry invalidates the request even when another entry could authorize it under Resource Access's alternative-entry rule ([I-D.draft-mcguinness-oauth-mission-resource-access]). That is this profile's credential validation policy; the prohibition on falling back to the Mission's authority does not by itself require it. An entry the PDP evaluates and finds unsatisfied invalidates nothing: another complete entry of authority still authorizes the action. The policy has an interoperability cost: adding an entry a PDP cannot evaluate makes the credential unusable at that PDP. 5.7. Action Parameters and Parameter Digest When parameter binding is required for the requested action's class under [I-D.draft-mcguinness-mission-runtime], the PEP supplies, in the AuthZEN action object's properties (action.properties), the following action-scoped members: parameters: CONDITIONAL. An object. When present, it MUST be the operation-profile-normalized parameter object ([I-D.draft-mcguinness-mission-runtime]): the same bytes the parameter_digest is computed over, so the PDP's recomputation matches. The shape is action-specific. The PEP MAY omit parameters and supply only parameter_digest where the raw values are sensitive ([I-D.draft-mcguinness-mission-runtime-evidence]), but only when the PDP can still enforce the applicable parameter policy from the digest, the supplied parameter_attributes, or local state. If the PDP needs raw parameter values to evaluate an applicable constraint and they are supplied through neither parameters nor the equivalent privacy-preserving parameter_attributes, it MUST deny with parameter_violation. parameter_digest: REQUIRED for parameter-bound classes. A string. The parameter_digest defined by [I-D.draft-mcguinness-mission-runtime], carried on the wire with no second digest or canonicalization. The PDP recomputes it over any supplied parameters, and the executing PEP recomputes and reverifies it immediately before acting, both as the runtime profile requires. idempotency_key: CONDITIONAL. A string, distinct from parameters and parameter_digest. REQUIRED where the Operation Profile defines one for the action: the idempotency key the Operation Profile requires for a non-idempotent action in the high- consequence classes ([I-D.draft-mcguinness-mission-runtime]). It identifies one intended execution of one normalized request, so a legitimate re-execution mints a new key. The PDP processes it as Section 5.7.1 specifies. parameter_attributes: OPTIONAL. An object. Privacy-preserving attributes derived from the action parameters that the PEP supplies for constraint evaluation when it withholds raw parameters ([I-D.draft-mcguinness-mission-runtime-evidence]). It is the wire carriage of the derived attributes the runtime profile's privacy carve-out relies on ([I-D.draft-mcguinness-mission-runtime]); each member is a deployment-defined attribute the applicable constraints evaluate against. A constraint the PDP cannot evaluate from parameter_digest, parameter_attributes, or local state MUST fail closed. 5.7.1. Idempotency Claim The PDP handles an idempotency_key by action class: * For a key in the irreversible-action, external-commitment, or privileged-administration classes, the PDP atomically claims the pair (idempotency scope, idempotency_key) together with the request's operation identity (Section 5.18) before issuing a permit ([I-D.draft-mcguinness-mission-runtime]). * For a reversible consequential write's key, the PDP makes no claim, and the enforcing PEP or resource deduplicates instead ([I-D.draft-mcguinness-mission-runtime]). A claim resolves per the runtime profile's idempotency claim-state table: * While the claim is unresolved, a matching key and operation identity is a transient duplicate_suppressed (Section 6.5), carrying next_action: retry with the reconciliation window as its retry_after guidance (Section 6.1). * Once the operation completes, a matching key and operation identity is a terminal duplicate_suppressed, carrying next_action: none. The prior result is available from the resource under the Operation Profile, never re-executed and never replayed by a fresh PDP decision. * When the same key instead matches a prior claim under a different operation identity, the reuse is a conflict, never a new execution: the PDP MUST deny with idempotency_conflict (Section 6.5), terminal in every claim state. Reuse of an idempotency_key never yields a second execution within the deployment's declared idempotency horizon, which for the high- consequence classes is backed by a durable consumed-key tombstone ([I-D.draft-mcguinness-mission-runtime]). An intentional re- execution is a new operation under a new key, which an action-bound approval MAY authorize as such, never as re-execution under the consumed key. 5.8. Action Phase Context context.action_phase is CONDITIONAL: REQUIRED when the PEP's trusted Operation Profile identifies the operation as a phase of a compound action ([I-D.draft-mcguinness-mission-runtime]). Its value is a string from the closed set preflight, prepare, commit, or compensate. The PEP MUST derive the value from that profile, never from an agent- supplied argument. The PDP MUST refuse, before evaluation, a request whose action_phase is outside this set, or is absent where its trusted operation declaration requires a phase (Section 6.8), and MUST return the validated phase as conditions.action_phase on a permit. Decision Evidence records the validated phase and omits malformed input ([I-D.draft-mcguinness-mission-runtime-evidence]). The request member communicates the phase being evaluated; it is not a permit binding, and neither it nor Decision Evidence's action_phase substitutes for the permit condition (Section 6.1.1). If the PEP cannot establish the phase before requesting a decision, it MUST refuse locally without requesting or using a permit; Section 6.8 defines the carrier. 5.9. Compensation Link Context context.compensates_evaluation_id is CONDITIONAL: REQUIRED on a request whose action_phase is compensate. Its value is a string with the syntax of evaluation_id (Section 6.1): the evaluation identifier of the Decision whose committed effect the operation offsets ([I-D.draft-mcguinness-mission-runtime]). The PEP MUST establish the value from retained, verified evidence of that committed effect and of the Decision it executed under, taking that Decision's evaluation_id, and MUST NOT take it from an agent- supplied argument. An agent MAY identify the effect it asks to offset; the PEP resolves that identification against its own evidence. A PEP that cannot establish the link MUST refuse before requesting a decision (Section 6.8). The PDP checks the member (Section 5.20) and records the validated value as Decision Evidence's compensates_evaluation_id, on a permit or a denial, omitting malformed input ([I-D.draft-mcguinness-mission-runtime-evidence]). The member is not a permit condition and confers no authority: the compensation is evaluated as its own action, and the compensated action's permit never authorizes it ([I-D.draft-mcguinness-mission-runtime]). 5.10. Evaluation-Context Binding For an operation covered by the runtime profile's optional Evaluation-Context Binding extension, the PEP MUST supply context.evaluation_context_digest and context.evaluation_context_binding. * evaluation_context_digest is the prefixed canonical-object digest of the captured context. It is an observation the PDP echoes, not one it can recompute without the PEP's secret salt and authoritative resolved facts. * evaluation_context_binding is the versioned descriptor reference {id, version, digest} defined by [I-D.draft-mcguinness-mission-runtime], committed inside the context digest. Granularity belongs to the referenced descriptor's individual facts, not a scalar wire mode. These inputs come from the authenticated PEP authorized for the enforcement scope, never from agent arguments. The PDP MUST validate their shape and applicability against the declared operation binding, and MUST NOT issue a covered permit when either is missing, malformed or incompatible with that declaration; it refuses such a request before evaluation (Section 6.8). Policy-evaluable values still travel in the existing parameter or attribute inputs. The extension is claimed per class: neither input becomes mandatory for an operation outside its declared scope, and existing permit lifetime, single-use, phase, and parameter bindings are unchanged. 5.11. Audience and Mission State Observation audience: REQUIRED. A string. Carried at resource.properties.audience (Section 5.1). The boundary identifier of the enforcement boundary the PEP established for the request, byte for byte ([I-D.draft-mcguinness-mission-runtime], Section "Enforcement Boundary Identifier"); on OAuth, the protected resource's identifier, never an aud value or alias ([I-D.draft-mcguinness-mission-runtime-oauth]). Entries are matched against operation_target, never this member (Section 5.1). mission_state_observation: CONDITIONAL. An object, carried at context.mission_state_observation. REQUIRED where the deployment's declared state-source placement has the PEP supply state; absent where state establishment is placed with the PDP, whose obligation the PDP-side consistency checks state (Section 5.20). It consolidates the PEP's Mission-state observation and its freshness into one snapshot, conveying the runtime profile's state and freshness inputs on the wire. It is the carrier for a PEP-supplied observation, never a substitute for a state source the PDP can itself consult (Section 5.20). Members: state: REQUIRED. A string. The current Mission lifecycle state the PEP established from its Mission state source ([I-D.draft-mcguinness-mission-runtime]). termination: OPTIONAL. An object. The Mission's termination ([I-D.draft-mcguinness-oauth-mission], Section "Mission Termination") when state is terminated; absent otherwise. version: OPTIONAL. An integer. The Mission's state version as the status profile defines it ([I-D.draft-mcguinness-oauth-mission-status]), populated where the PEP holds it. The PDP compares it with its own tracked state version (Section 5.20). mode: REQUIRED. A string. One of fresh, cached, or event_driven (Section 5.11.1). freshness_at: REQUIRED in every mode. An RFC 3339 [RFC3339] timestamp. When the PEP's view of the Mission state was current. mission_status_issued_at: REQUIRED for cached and event_driven, OPTIONAL for fresh. An RFC 3339 timestamp. When the relied-on Mission state was issued: for a signed Mission Status Response, the RFC 3339 representation of its iat, a NumericDate ([I-D.draft-mcguinness-oauth-mission-status]). freshness_at can precede this value by no more than the published maximum clock skew (Section 5.21). mission_status_expires_at: REQUIRED for cached and event_driven, OPTIONAL for fresh. An RFC 3339 timestamp. When reliance on the observed Mission state ends: for a signed Mission Status Response, its mission.fresh_until, never its exp, or an earlier end deployment policy sets; for a lease, the lease's end. assertion: OPTIONAL. A string. The signed Mission Status Response [I-D.draft-mcguinness-oauth-mission-status] the PEP obtained state and version from, when the PEP holds one, so the PDP can verify the snapshot instead of trusting an unsigned PEP assertion. 5.11.1. Observation Modes The runtime profile defines the Mission state source, the deployment's maximum staleness bound, and the rule that a consequential action fails closed when the Mission cannot be established as active within that bound ([I-D.draft-mcguinness-mission-runtime]); mission_state_observation is only their wire representation. Its mode member records how the PEP obtained the state: fresh: The PEP consulted the Mission state source synchronously before the action. cached: The PEP used cached Mission state within the deployment's staleness bound. event_driven: The PEP relies on event-channel invalidation, but the cached state remains bounded by the deployment's staleness bound, or the lease the state carries, exactly as for cached; it is not valid indefinitely. A missed or delayed invalidation event does not extend validity, and when the bound elapses without a confirming event the PEP MUST refresh from a Mission state source rather than continue on the cache. When freshness cannot be established within the bound, the PDP fails closed for consequential actions as the runtime profile requires, with a stale_state denial (Section 6.5). 5.12. External-Communication Context external_communication: CONDITIONAL. A boolean, carried at action.properties.external_communication alongside the other action-scoped members (Section 5.7). A registered extension point owned by the harness profile ([I-D.draft-mcguinness-mission-harness]), which defines the predicate and how the PEP computes it. REQUIRED for every action to which a class applies for which the deployment declares PDP- enforced taint; OPTIONAL otherwise. 5.13. Taint Context taint: OPTIONAL, carried at context.taint. A registered extension point owned by the harness profile ([I-D.draft-mcguinness-mission-harness]), which defines the member shape (tainted, granularity, source_class) and the taint-state trust rules. Where the deployment's Enforcement Scope Statement declares PDP-enforced taint for a class that applies to the action ([I-D.draft-mcguinness-mission-runtime], Section "Action Classification"), the fail-closed default applies: the PDP MUST deny with taint_context_missing (Section 6.5) when context.taint is absent from an in-scope decision. 5.14. Protected-Data Context protected_data: OPTIONAL, carried at context.protected_data. A registered extension point owned by the harness profile ([I-D.draft-mcguinness-mission-harness]), which defines the member shape (exposed, established_by), the establishing layer's attachment the PEP populates it from, its reading when absent, and the exposure qualification it can satisfy. It never relaxes the fail-closed default of context.taint (Section 5.13). 5.15. Work-Item Context work_item: OPTIONAL, carried at context.work_item. A registered extension point owned by the harness profile ([I-D.draft-mcguinness-mission-harness]), which defines the qualified work-item identifier and its attacher. An object with exactly two members, both strings: issuer, the URI of the harness that issued the identifier, and id, that harness's identifier for the work item. The PEP populates context.work_item only from the harness's authenticated attachment: the work item a verified Mission reference carries, for the attacher the Enforcement Scope Statement declares for the authenticated client ([I-D.draft-mcguinness-mission-harness]). A reference that verifies does not by itself establish that attachment, and the PEP never takes a work item from tool arguments or another value the agent supplies. The PDP MUST refuse, before evaluation, a request whose context.work_item is present but not of that shape: an unusable input, recorded as the Refusal Record work_item_invalid (Section 6.8), never an evaluated denial. Where the deployment's Enforcement Scope Statement claims work-item attribution for a class that applies to the action, the PDP MUST deny with work_item_context_missing (Section 6.5) a decision from which context.work_item is absent. Elsewhere its absence relaxes nothing, and the deployment claims Mission-level attribution only. The member is part of the authorization binding and the cache key (Section 5.18), so no cached decision or resumed execution crosses work items, and a permit for a decision that carried it binds it as the work_item condition (Section 6.1.1). 5.16. Approval Context The OPTIONAL approval member is carried at context.approval on a fresh evaluation after an access request is approved. The PEP returns here exactly the approval object the Access Request Service issued: ARAP's approval object [ARAP] unchanged, with all of ARAP's members. ARAP makes approved_at OPTIONAL and approved_until REQUIRED; this profile does not subset or rename either. On top of ARAP's own requirements, before treating context.approval as satisfying approval_required, the PDP MUST additionally establish the Mission checks: that the approval binds to the request's parameter_digest (a mismatch means the approval does not cover these parameters), that its grant time is within the deployment's maximum approval age ([I-D.draft-mcguinness-mission-runtime]), and that approved_until has not passed. This profile requires a verifiable completion time for an action- bound approval: approved_at when present, otherwise the PDP MUST establish the grant time from the approval record it resolves by id or from verified signed state. An approval whose completion time cannot be established is treated as unverifiable. An approval that fails a Mission check, or whose completion time cannot be established, does not satisfy the denial. Where a resource contract applies to the action (Section 5.3), the approval binds its reference through the approved Resource, which ARAP requires the approval record or its binding material to contain or let the PDP determine ([ARAP], Section 12.3.1). The Mission checks then also establish that the approved Resource's properties.resource_contract equals the request's, member for member. An approval given under one reference does not cover an action under another, even when both references resolve: the PDP MUST deny that re-evaluation with ARAP's out_of_scope, which ARAP defines for a valid approval whose scope the current evaluation falls outside ([ARAP], Section 8.2). The PDP MUST record the presented approval id in Decision Evidence. The approval is an input attribute, never a bearer grant: the PDP evaluates current policy and current state. When policy requires an action-bound approval and no context.approval is presented, the PDP MUST deny with approval_required (Section 6.5). When a re-evaluation that presented context.approval fails a Mission or ARAP check, the PDP instead returns one of ARAP's five registered re-evaluation reasons in context.reason, with ARAP's default next_action for each: * approval_expired (request); * out_of_scope (request); * grant_pending (retry); * policy_denied (none); * approval_unverifiable (none), which also covers an approval whose completion time cannot be established. These are ARAP-owned values carried under its extensibility rule [ARAP], not values this profile defines. 5.17. History Context Which history predicates a decision requires is a policy decision: the materialized policy view, or the PDP's own policy, names the predicates it evaluates for the action. The request does not select them; a requester able to choose which history a decision turns on could pick predicates favorable to it. This document defines one predicate type that policy MAY name, action_class_completed. It is satisfied when the PDP's evidence store holds, for the established Mission, at least one Decision Evidence record with decision permit to which the named class applied, and every such permit has a completion witness: a linked Execution Evidence record ([I-D.draft-mcguinness-mission-runtime-evidence]) whose outcome is completed and in which each authorized digest present (authorized_parameter_digest, authorized_evaluation_context_digest) has an equal effective counterpart (effective_parameter_digest, effective_evaluation_context_digest). The named class applied to a record when its action_classes member lists that class or, in a record without action_classes, its action_class is that class, so an execution the class applied to counts whichever class was assigned. The predicate carries an action_class: a runtime action class name, from the same value space as the Decision Evidence action_class member ([I-D.draft-mcguinness-mission-runtime-evidence]). For a migration Mission whose copy steps are consequential writes, policy names the delete step's precondition this way: { "predicate": "action_class_completed", "action_class": "consequential_write" } The predicate is per permit: a permit has a completion witness when any of its executions is one, so a failed use of a multi-use permit, or a replay suppressed as permit_consumed, does not remove it. A predicate that requires every use of a permit to complete is a different predicate type. A single-use permit has one authorized completion ([I-D.draft-mcguinness-mission-runtime-evidence]): only the completion that took its single use can be its witness, and any other completed record under it creates none, whatever its digests. The order in which records reach the evidence store does not establish which completion took the use, nor do the records' sequence or timestamps. Where no trusted source, such as the PEP's consumed- identifier store, establishes it, a single-use permit with more than one completed record has a witness only when each of them would be one on its own, and its completion is undetermined when only some would. A completed record whose effective digest differs from its authorized digest records an unauthorized execution ([I-D.draft-mcguinness-mission-runtime-evidence]): it remains valid evidence, and it is not a completion witness. A completed record that carries an authorized digest without its effective counterpart is not a completion witness either, and a permit with such a record and no witness has an undetermined completion. The predicate is not satisfied when some permit has neither a witness nor an undetermined completion, and is otherwise not establishable while any permit's completion is undetermined. The evaluation relies on the evidence store returning the complete, current Mission history; sequence checks detect gaps only within the records returned ([I-D.draft-mcguinness-mission-runtime-evidence], Section "History Completeness"). An extension predicate type MUST be either a collision-resistant name or a name coordinated within this document family, under the same rule as denial reasons (Section 6.5). The specification defining a predicate type states whether it selects an operation and which of its members names it; a mission_history entry for a type the PDP does not recognize carries no action. The harness profile defines the coordinated type work_item_prerequisite_completed ([I-D.draft-mcguinness-mission-harness]). The OPTIONAL context.mission_history request object carries only what the PEP can supply in support of the policy-selected predicates, never a selection of which predicates apply. Members: evidence_reference: OPTIONAL. A string. A reference to the evidence or state source the PEP consulted, when that source is not the PDP's own evidence store, so the PDP can resolve the same history the PEP saw. attribute_values: OPTIONAL. An object. Precomputed values for the policy-selected predicates, supplied where the deployment permits the PEP to compute them directly, mirroring the privacy-preserving carve-out for action parameters (Section 5.7); each member is a deployment-defined attribute the applicable predicate evaluates against. The PDP evaluates each policy-selected predicate against its own evidence store, or, where the deployment's policy permits a non- default evidence source, against the supplied evidence_reference or attribute_values. A caller-supplied context.mission_history value is never itself evidence of a predicate's outcome, and carries no result member; the PDP MUST NOT accept a predicate outcome asserted through any context member. The request object is distinct from the Decision Evidence mission_history member, which the PDP computes (Section 7): request input neither chooses the required predicates nor asserts their outcomes. The PDP MUST refuse the evaluation of a policy-selected predicate it does not recognize, treating the predicate as not establishable, consistent with this profile's unknown-value conventions (Section 6.5). Where deployment or Resource policy requires a history predicate that is unsatisfied or cannot be established, the PDP MUST deny with history_not_satisfied (Section 6.5); the runtime profile fixes the fail-closed posture and the evidence store's freshness discipline ([I-D.draft-mcguinness-mission-runtime]). The PDP MUST record the evaluated predicates and their outcomes in Decision Evidence ([I-D.draft-mcguinness-mission-runtime-evidence]). Where a predicate's type selects an operation, its mission_history entry carries that operation's identifier as action; an action_class_completed entry, which selects by class alone, carries none. 5.18. Canonical Projections The idempotency claim (Section 5.7.1), permit retransmission and reuse, and this profile's permit-binding rules each turn on a different relation over the normalized evaluation request, so this section defines three internal canonical projections. Each is sha- 256: followed by the base64url, no-padding SHA-256 [RFC6234] digest of the JCS [RFC8785] serialization of its defined subset of the normalized request. None is a wire parameter: an implementation computes each internally from the request and response members it already carries. Operation identity: The action, the resource, including its operation target (Section 5.2), and the normalized effectful parameters. This is what an idempotency key identifies one execution of. Authorization binding: The subject, the actor, the audience (Section 5.11), the operation target, the resource contract reference (Section 5.3), credential-derived facts, the Mission reference, the authority and view identity, the approval, and every policy-relevant context input, including taint, protected_data, work_item, the Mission history reference, the capability source, and any extension member a companion profile adds. This is what a permit is bound to. A new companion context member is part of the authorization binding by default; a companion narrows it only by amending this enumeration. It excludes only the request's volatile machinery: the supported_obligations declaration and transport metadata. Cache key: The authorization binding plus the explicit state generation the decision consulted (the PDP's tracked mission_state_version, or an equivalent status-issuance generation marker where the deployment does not track one), excluding observation telemetry: freshness_at, retrieval or issuance timestamps, and other volatile machinery. State generation is cache-relevant; observation telemetry is not. A decision computed against an unchanged authorization binding and an unchanged state generation is the same decision regardless of when either side observed it, so permit retransmission and permit reuse test the cache key for equality (Section 6.9). The idempotency claim identifies one execution rather than one decision, so it tests operation identity instead ([I-D.draft-mcguinness-mission-runtime]). What a permit is bound to, and what a re-presented request must match to rely on that binding, is the authorization binding. Each projection includes the operation target, so two requests that differ only in their target differ in all three, including cacheable reads that carry no parameter_digest. 5.19. PDP Request Examples In this document's ERP examples, the operation contract identifies https://erp.example.com as the operation target ([I-D.draft-mcguinness-oauth-mission-resource-access]); actions and constraints determine the permitted operations within that service. The AuthZEN resource object's type and id name the object for Resource-policy evaluation, not the operation target. The examples' policy_view_id (sha-256:kP3xR9sQ...) differs from the runtime profile's worked value (sha-256:RBI8j_YN...) because the two manifests name different compilers; both are valid views of the same Mission. In these examples, journal-entries.write is a reversible consequential write whose deployment chooses single-use permits; that choice, not a high-consequence class, is why its permits carry use_limit: 1 (Section 6.1.1). The following is an example of a PDP request for the ERP reconciliation Mission: POST /pdp/access/v1/evaluation HTTP/1.1 Host: pdp.example.com Content-Type: application/json Authorization: ... { "subject": { "type": "user", "id": "user_3p2q8mN1a0kV7tR", "properties": { "iss": "https://idp.example.com" } }, "resource": { "type": "journal-entry", "id": "je_2026Q3_inv_8421", "properties": { "audience": "https://erp.example.com", "operation_target": "https://erp.example.com" } }, "action": { "name": "journal-entries.write", "properties": { "parameters": { "amount_usd": "423.50", "source_invoice_id": "inv_2026Q3_842" }, "parameter_digest": "sha-256:WPVi6EnQ7H9Fh-qk9ADxmTg8zruOdVUX1esl-v3TfCI" } }, "context": { "mission": { "id": "msn_8RfX2Lqv9TqMv4z7sA2bN1k0YpEdHc9-", "issuer": "https://as.example.com", "policy_version": "deploy-policy:v17", "policy_view_id": "sha-256:kP3xR9sQ7nM2vL4tY6bD1eF8jC5wH0pV2nR3kQ4mZ7s" }, "mission_state_observation": { "state": "active", "mission_status_issued_at": "2026-11-02T08:14:00Z", "mission_status_expires_at": "2026-11-02T08:15:00Z", "mode": "cached", "freshness_at": "2026-11-02T08:14:00Z" }, "actor": { "client_id": "s6BhdRkqt3", "client_instance_id": "inst_macbook_7f3a", "act": [ { "iss": "https://as.example.com", "sub": "s6BhdRkqt3" } ] }, "credential": { "issuer": "https://as.example.com", "expires_at": "2026-11-02T08:19:00Z", "authority": [ { "type": "mission_resource_access", "resource": "https://erp.example.com", "actions": ["journal-entries.write"], "constraints": { "max_amount": { "amount": "500.00", "currency": "USD" } } } ] } } } For an agent acting for its own workload principal, subject names the workload principal and context.actor names the agent's client. The token carries no act, so that client is the immediate actor ([I-D.draft-mcguinness-mission-runtime]). The subject.type value workload is deployment-profiled. The following example shows the ledger-reconciler agent reading an invoice: POST /pdp/access/v1/evaluation HTTP/1.1 Host: pdp.example.com Content-Type: application/json Authorization: ... { "subject": { "type": "workload", "id": "agt_ledger_reconciler", "properties": { "iss": "https://login.example.com" } }, "resource": { "type": "invoice", "id": "inv_2026Q3_842", "properties": { "audience": "https://erp.example.com", "operation_target": "https://erp.example.com" } }, "action": { "name": "invoices.read" }, "context": { "mission": { "id": "msn_3Jt8Vw2pQ6sN1xR4kL9zB7mD0cF5hYq2", "issuer": "https://as.example.com" }, "mission_state_observation": { "state": "active", "mode": "fresh", "freshness_at": "2026-11-02T08:14:00Z" }, "actor": { "client_id": "ledger-reconciler" }, "credential": { "issuer": "https://as.example.com", "expires_at": "2026-11-02T09:14:00Z", "authority": [ { "type": "mission_resource_access", "resource": "https://erp.example.com", "actions": ["invoices.read"] } ] } } } 5.20. PDP-Side Consistency Checks A PDP that also serves non-Mission AuthZEN traffic MUST NOT downgrade: for an action within a runtime enforcement scope it mediates, a request whose context lacks the mission member (Section 5.4) is malformed under this profile, and the PDP MUST refuse it rather than evaluate it against non-Mission policy alone, and MUST record the refusal as a Refusal Record with denial_reason mission_context_missing ([I-D.draft-mcguinness-mission-runtime-evidence]), populating the members the PDP itself can attest. A permit issued to an in-scope consequential action without Mission evaluation is an enforcement bypass, not a decision ([I-D.draft-mcguinness-mission-runtime]). Likewise, where the runtime Enforcement Scope Statement declares action classes, a mediated request that reaches the PDP with no action class established, neither carried from the PEP's Operation Profile nor assigned by the PDP's own trusted declaration, or carried with a class other than the one that declaration assigns its operation and phase, is malformed: the PDP MUST refuse it before evaluation, MUST NOT evaluate it under a default class or record a class in Decision Evidence, and MUST record the refusal as a Refusal Record with denial_reason action_class_missing ([I-D.draft-mcguinness-mission-runtime-evidence]). In addition to evaluating the decision inputs the runtime profile requires, the PDP MUST verify that the AuthZEN-carried envelope is self-consistent: 1. When present, the Mission state conveyed in context.mission_state_observation.state is exactly active; every other value, recognized or not, is non-active per the OAuth binding's forward-compatibility rule ([I-D.draft-mcguinness-oauth-mission]) and the PDP returns mission_inactive (Section 6.5). Where state establishment is placed with the PDP and the member is absent (Section 5.11), the PDP MUST establish state from its own source or deny with stale_state. A PDP with direct access to a Mission state source MUST prefer its own fresher view over context.mission_state_observation.state, and MUST return mission_inactive when its view disagrees with the PEP-supplied state. PEP-supplied state is a floor, never a substitute for a state source the PDP can itself consult. 2. The id and issuer in context.mission equal the mission_id and issuer committed in the materialized policy view the PDP has loaded for this Mission, or, for a PDP that evaluates the Mission's recorded authority directly, those of the trusted Mission record it evaluates ([I-D.draft-mcguinness-mission-runtime]); the PDP returns view_inconsistent on any inequality. When context.mission.authority_hash is also present, it MUST equal the authority_hash that view or record commits, with the same view_inconsistent result on inequality (its omission is covered by rule 5 below). 3. Where the PDP evaluates through a view and context.mission.policy_view_id is present, it MUST equal the loaded view's policy_view_id, and the PDP returns view_inconsistent on inequality. 4. When context.mission_state_observation.version is present, the PDP compares it against its own tracked Mission state version, recorded in Decision Evidence rather than echoed on the response ([I-D.draft-mcguinness-mission-runtime-evidence]), never against policy_view_id, which commits to authority only (Section 4), and treats a mismatch as staleness (stale_state): one side has missed a committed change. 5. A PDP MUST NOT fail a decision solely because the optional policy_view_id, policy_version, or authority_hash was omitted; the view the PDP loaded, or the Mission record it evaluates directly, is authoritative. 6. When context.credential.expires_at is present, it has not passed; otherwise the PDP returns credential_invalid. 7. The freshness the PEP supplied in context.mission_state_observation is within the deployment's staleness bound; otherwise the PDP returns stale_state, with the freshness-window violation in the denial reason. 8. For an action whose class requires parameter binding ([I-D.draft-mcguinness-mission-runtime]), action.properties.parameter_digest MUST be present; if it is absent the PDP returns parameter_violation. When action.properties.parameters is also present, the PDP-recomputed digest MUST match action.properties.parameter_digest, otherwise parameter_violation. When parameters is omitted under the privacy carve-out, the PDP returns parameter_violation if it cannot evaluate an applicable parameter constraint (Section 5.7). A parameter-bound action MUST NOT be permitted without a verified parameter_digest. 9. A deployment adopting Mission Capability Binding applies that companion's source checks and its capability_drift extension reason ([I-D.draft-mcguinness-mission-capability-binding]); this profile consumes an already established action identity. 10. For a request claiming the cross-domain Origin Principal profile ([I-D.draft-mcguinness-oauth-mission-cross-domain]), the PDP establishes the mapped local principal and requires it to equal the authenticated request subject before entitlement lookup: mapped_subject = map(context.mission.subject, audience, tenant, mapping_policy) require mapped_subject == (subject.properties.iss, subject.id) entitlement = lookup_current_entitlement(mapped_subject) The PDP never compares context.mission.subject with the request subject directly: the two occupy different namespaces, and only the mapping's output is comparable. The PDP is authoritative for the mapping in the default placement. A deployment that instead places mapping at the PEP MUST convey it as a structured, integrity-protected mapping observation carrying its source, the mapping-policy identifier and version, the issuer- qualified input, the local output, the observation time, and a validity bound; the PDP verifies it as it verifies context.mission_state_observation. A failed, missing, ambiguous, or stale result at either step denies with the profile's principal_mapping_failed extension reason ([I-D.draft-mcguinness-oauth-mission-cross-domain], Section 6.5); entitlement staleness beyond the declared bound denies likewise. The permit and its evidence bind to both the origin and the mapped local identity through the principal_mapping evidence object ([I-D.draft-mcguinness-mission-runtime-evidence]). 11. For an action evaluated against mission_resource_access authority, resource.properties.operation_target is present, is a string holding one absolute URI, and names a resource within the declared enforcement scope for which the requesting PEP is authorized ([I-D.draft-mcguinness-mission-runtime], Section "Remote Decision Channel"). Otherwise the PDP MUST refuse the request before evaluation, with no fallback to resource.properties.audience, resource.type or resource.id, and MUST record the refusal as a Refusal Record with denial_reason operation_target_invalid ([I-D.draft-mcguinness-mission-runtime-evidence]). 12. context.compensates_evaluation_id, when present, is a string with the syntax of evaluation_id (Section 6.1), and it is present when context.action_phase is compensate (Section 5.9). Otherwise the PDP MUST refuse the request before evaluation and MUST record the refusal as a Refusal Record with denial_reason compensation_link_invalid ([I-D.draft-mcguinness-mission-runtime-evidence]). 5.21. Clock Skew The following time comparisons MAY allow a small leeway to absorb clock skew between the components: the permit's valid_until at the PEP (Section 6.1.1); the context.mission_state_observation freshness window and context.credential.expires_at at the PDP; and the approval maximum age (Section 5.16). Any leeway MUST NOT exceed the deployment's maximum clock skew, and the deployment MUST publish that maximum-skew assumption in its Enforcement Scope Statement ([I-D.draft-mcguinness-mission-runtime]). A value outside its window by more than the published maximum skew MUST fail closed: an expired permit, a stale freshness window, or an expired credential is refused, and the leeway never extends a permit lease or freshness lease beyond its bound plus the published skew. 5.22. Batch Evaluations The AuthZEN evaluations (boxcar) endpoint MAY be used to submit several Mission-bound decisions in one request. Batching is a transport optimization and changes none of the per-item enforcement semantics: * Each item yields its own Decision Evidence Object ([I-D.draft-mcguinness-mission-runtime-evidence]) with its own evidence_id and sequence, assigned in request order. * Any metered bounds apply per item in request order ([I-D.draft-mcguinness-mission-metering]). * Permits are per item, so a boxcar MAY return a mix of permits and denials. * This profile requires execute_all semantics: every item MUST be evaluated independently, on the same terms as a single request, regardless of an earlier item's decision. A PDP MUST reject, as an AuthZEN request error rather than a silent downgrade, a request whose recognized options ask for deny_on_first_deny, permit_on_first_permit, or any other recognized short-circuit or fail-fast semantics. An options member the PDP does not recognize follows AuthZEN's own rules for unrecognized options. The following example is a batch request for two journal-entry writes under the ERP reconciliation Mission, where the second exceeds the entry's max_amount ceiling of 500.00 USD. The shared subject is hoisted to the request's default members per [AUTHZEN]; each item carries its complete context: POST /pdp/access/v1/evaluations HTTP/1.1 Host: pdp.example.com Content-Type: application/json Authorization: ... { "subject": { "type": "user", "id": "user_3p2q8mN1a0kV7tR", "properties": { "iss": "https://idp.example.com" } }, "evaluations": [ { "resource": { "type": "journal-entry", "id": "je_2026Q3_inv_8421", "properties": { "audience": "https://erp.example.com", "operation_target": "https://erp.example.com" } }, "action": { "name": "journal-entries.write", "properties": { "parameters": { "amount_usd": "423.50", "source_invoice_id": "inv_2026Q3_842" }, "parameter_digest": "sha-256:WPVi6EnQ7H9Fh-qk9ADxmTg8zruOdVUX1esl-v3TfCI" } }, "context": { "mission": { "id": "msn_8RfX2Lqv9TqMv4z7sA2bN1k0YpEdHc9-", "issuer": "https://as.example.com" }, "mission_state_observation": { "state": "active", "mode": "fresh", "freshness_at": "2026-11-02T08:14:00Z" }, "actor": { "client_id": "s6BhdRkqt3" }, "credential": { "issuer": "https://as.example.com", "expires_at": "2026-11-02T08:19:00Z", "authority": [ { "type": "mission_resource_access", "resource": "https://erp.example.com", "actions": ["journal-entries.write"], "constraints": { "max_amount": { "amount": "500.00", "currency": "USD" } } } ] } } }, { "resource": { "type": "journal-entry", "id": "je_2026Q3_inv_9310", "properties": { "audience": "https://erp.example.com", "operation_target": "https://erp.example.com" } }, "action": { "name": "journal-entries.write", "properties": { "parameters": { "amount_usd": "780.00", "source_invoice_id": "inv_2026Q3_931" }, "parameter_digest": "sha-256:mzFwtXAT6_hY0v8_NFHMDJG39HFuWY2fRcOCSFGDyyE" } }, "context": { "mission": { "id": "msn_8RfX2Lqv9TqMv4z7sA2bN1k0YpEdHc9-", "issuer": "https://as.example.com" }, "mission_state_observation": { "state": "active", "mode": "fresh", "freshness_at": "2026-11-02T08:14:00Z" }, "actor": { "client_id": "s6BhdRkqt3" }, "credential": { "issuer": "https://as.example.com", "expires_at": "2026-11-02T08:19:00Z", "authority": [ { "type": "mission_resource_access", "resource": "https://erp.example.com", "actions": ["journal-entries.write"], "constraints": { "max_amount": { "amount": "500.00", "currency": "USD" } } } ] } } } ] } The response returns one decision per item, in request order; the first is a permit and the second a parameter_violation deny, whose failing max_amount key is listed in that item's Decision Evidence contributing_constraints: { "evaluations": [ { "decision": true, "context": { "evaluation_id": "dec_2FpQ8kV5nR1tX7mB4sJ9eL6wYc", "conditions": { "parameter_digest": "sha-256:WPVi6EnQ7H9Fh-qk9ADxmTg8zruOdVUX1esl-v3TfCI", "valid_until": "2026-11-02T08:15:00Z", "use_limit": 1 } } }, { "decision": false, "context": { "evaluation_id": "dec_6JwN3xT9rQ4mV8kP1sB5eZ2yLd", "reason": "parameter_violation" } } ] } 5.23. Timeouts, Retries, and Overload An evaluation exchange carries metered and single-use consequences. Its transport behavior is as follows: * A PEP MUST bound each evaluation call with a timeout inside the action class's staleness budget and treat expiry as pdp_unreachable ([I-D.draft-mcguinness-mission-runtime-evidence]): fail closed, never a permit. * A PEP MAY retry an evaluation whose response was lost, carrying the same normalized parameters and idempotency key. The runtime profile's retransmission rule then returns the prior decision where its permit is unexpired and unconsumed ([I-D.draft-mcguinness-mission-runtime]). Whether the retried request consumes a metered bound is the metering companion's rule ([I-D.draft-mcguinness-mission-metering]). * A PDP under overload sheds load explicitly with HTTP 429 and Retry-After. This is transport backpressure, not an authorization result: the PEP MUST NOT retry before Retry-After elapses, and when no decision is obtainable within its policy window it fails closed as pdp_unreachable under the runtime profile. A successful evaluation that denies transiently is expressed with context.next_action: retry (Section 6.1), never as an HTTP error. * A PDP MAY publish a maximum batch size and refuse an oversized boxcar with HTTP 413; the PEP splits the batch rather than dropping items. 6. PDP Response The PDP returns its permit or denial in the AuthZEN response [AUTHZEN]: a boolean decision and an optional context object. Runtime denials are successful evaluations, represented as decision: false with the context members below, not as transport errors. 6.1. Response Decision Context This profile defines the following AuthZEN response context members: evaluation_id: REQUIRED. A string. The correlation identifier for this evaluation, ARAP's evaluation_id [ARAP]. ABNF: 1*64( ALPHA / DIGIT / "-" / "_" ). At least 128 bits of entropy. Decision Evidence records this same value as its own evaluation_id member, alongside the evidence record's distinct evidence_id ([I-D.draft-mcguinness-mission-runtime-evidence]). evaluated_at: OPTIONAL. An RFC 3339 [RFC3339] timestamp of the evaluation, ARAP's evaluated_at member [ARAP]. RECOMMENDED on a denial that carries access_request or next_action. reason: REQUIRED when decision is false. A string: ARAP's reason member [ARAP], with a value from Section 6.5, including approval_required, which this profile defines, and any specification-defined extension under that section's extensibility rule, which also governs an unrecognized value. A constraint violation uses parameter_violation; the specific failing constraints keys are carried in the Decision Evidence contributing_constraints, not here. obligations: OPTIONAL. An array of obligation objects ([AUTHZEN-OBL]), present when the decision carries a genuine attached obligation, on either a permit or a denial. The PEP processes them as Section 6.2 specifies; an obligation on a denial confers nothing and does not make the denial a permit. conditions: REQUIRED when decision is true for a consequential action. An object carrying the permit's decision conditions (Section 6.1.1). access_request: OPTIONAL. An object: ARAP's requestable-denial context object, imported as ARAP defines it; this profile adds no members and no additional required members. Present on an out_of_authority or approval_required denial when the deployment exposes it as requestable (Section 6.6). The access request the PEP submits references the denied evaluation by the top-level context.evaluation_id, as [ARAP] specifies. Its presence does not change the decision: false result and does not grant access. next_action: OPTIONAL on a denial. A string: request, retry, or none ([ARAP]). request MUST be accompanied by a fresh context.access_request. retry marks the denial transient: the same request is expected to succeed on re-evaluation after a delay. none marks it terminal for policy and local action alike. ARAP defines next_action for the denial of a re-evaluation that presented an approval; this profile extends it, with identical semantics, to any denial. retry_after: OPTIONAL, only with next_action: retry. An integer: seconds the PEP waits before re-evaluating ([ARAP]). decision_evidence: OPTIONAL. An object: the complete signed Decision Evidence Object the PDP emitted for this evaluation, envelope included ([I-D.draft-mcguinness-mission-runtime-evidence]). The PDP constructs and signs it on its own emission path; a requesting PEP verifies it and retains it as received, byte for byte. Carrying it on the response does not change its treatment: it is retrospective evidence of what the PDP decided, never authorization to act (Section 6.9). The permit is the decision itself. 6.1.1. Decision Conditions The conditions object carries the permit's decision conditions ([I-D.draft-mcguinness-mission-runtime]): declarative constraints on relying on the permit, each evaluated at every use. Members: parameter_digest: CONDITIONAL. A string. The parameter binding the permit is bound to; REQUIRED, and equal to the request's action.properties.parameter_digest, when the action was parameter- bound. The PEP MUST enforce the exact request binding. This is the response's only carriage of the digest. valid_until: REQUIRED. An RFC 3339 [RFC3339] timestamp. The permit's validity bound, past which the permit MUST NOT be used. valid_until MUST NOT exceed the earliest of: * the presented credential's expiry, context.credential.expires_at (Section 5.6), when present; * the approval's approved_until, where the permit satisfies approval_required (Section 5.16); * the state valid-through; and * any policy-defined maximum permit lifetime for the action's class. The state valid-through is defined by source: a signed Mission Status Response's mission.fresh_until (not its exp), or a lease's end, both carried in a supplied context.mission_state_observation.mission_status_expires_at; or, for a polled observation reporting neither, freshness_at plus the deployment's published maximum staleness bound for the action's class. In fresh mode with no supplied mission_status_expires_at, the PDP derives the state valid-through from its own state view under the same rule. The observation timestamp (freshness_at) alone is never the state valid-through, so no mode leaves this window unbounded. use_limit: OPTIONAL. An integer. A consumption bound on evaluation_id. The PDP MUST set use_limit: 1 for a permit in the high-consequence classes ([I-D.draft-mcguinness-mission-runtime]), and the PEP MUST treat a high-consequence permit lacking it as invalid. Absent otherwise, the permit carries no PDP-enforced use limit. The PEP applies its own deployment classification floor to this duty, independent of any class the PDP applied internally: for an action the PEP's own deployment classification or a resource floor places in the high-consequence classes, the PEP MUST refuse a permit lacking use_limit: 1, whatever class the PDP recorded for itself in Decision Evidence ([I-D.draft-mcguinness-mission-runtime-evidence]). No class label is echoed on this response for the PEP to compare. The PDP is stateless with respect to consumption: it declares use_limit on the decision, and the PEP owns the consumed- identifier store and refuses a re-presented consumed evaluation_id. evaluation_context_digest: CONDITIONAL. A string. REQUIRED on a permit for an operation covered by Section 5.10; MUST equal the validated request's context digest. The executing PEP MUST recompute it under the descriptor the binding committed, from authoritative facts, and MUST refuse before an effect on a missing or unequal binding, a descriptor the operation's declaration no longer names, an unrecoverable descriptor, or an unavailable fact. A mismatch or inability to re-resolve after a permit is Execution Evidence suppressed, not a PDP denial reason, with the error the runtime evidence profile's classification names ([I-D.draft-mcguinness-mission-runtime-evidence]). action_phase: CONDITIONAL. A string from preflight, prepare, commit, or compensate; REQUIRED on a permit for an operation that is a phase of a compound action. It MUST equal the validated request phase (Section 5.8). At every use, the PEP MUST compare this condition with the phase from its own trusted Operation Profile and MUST NOT release an effect on inequality, a missing or malformed condition where a phase is required, or a phase it cannot establish. A phase condition on a crossing whose profile declares no phase is invalid there. The request context and the retrospective evidence phase never substitute for this condition. resource_contract: CONDITIONAL. An object; REQUIRED on a permit for an action to which a resource contract applies, and equal to the validated request's resource.properties.resource_contract (Section 5.3). At every use, the executing PEP MUST resolve this reference against its own current contract declarations and MUST NOT release an effect unless it resolves and equals the reference of the contract the PEP is about to perform the operation under. A missing condition where a contract applies, an unresolvable reference, or an unequal one is Execution Evidence suppressed / resource_contract_unresolvable, not a PDP denial. work_item: CONDITIONAL. An object; REQUIRED on a permit for a decision whose request carried context.work_item (Section 5.15), and equal to it. At every use, the executing PEP MUST compare it with the work item it attributes to the request it is executing, and MUST NOT release an effect when they differ or when either is present without the other. Such a refusal is Execution Evidence suppressed / work_item_mismatch, not a PDP denial. The PEP MUST honor every condition present, at every use of the permit. A permit carrying a condition member the PEP does not recognize is invalid: the PEP MUST NOT release the action's effect under it. New conditions are coordinated extensions under this document's extensibility rule (Section 6.5). If AuthZEN standardizes decision conditions or must-understand response semantics, that mechanism governs and this profile aligns with it. This profile defines no must-understand marker, because a marker inside the response cannot bootstrap itself: a component that predates the marker ignores it too. The guarantee is deployment- established instead: a PDP MUST NOT return a permit carrying conditions to a PEP it has not established, through deployment configuration and the mutually authenticated channel, as conforming to this profile. 6.2. Obligations An obligation is mandatory PEP work under the existing decision, whichever it is. It confers no authority, has no approver, and creates no governance state: unlike the governance lane (Section 6.7), it involves no adjudication, no new authority, no approval record, and no task handle. An obligation MAY involve subject interaction or an external side effect (a step-up challenge to the subject, a notification to another system); it never acquires authority from an adjudicator. 6.2.1. Obligation Processing An obligation MAY accompany a permit or a denial (Section 6.1): on a permit the PEP MUST fulfill each obligation before releasing the action's effect; on a denial the PEP MUST still deny the action and MUST additionally execute every returned obligation. The PEP MUST treat an obligation it does not recognize, or cannot fulfill, as an effective deny and MUST release no protected effect; it still executes every obligation it understands and can perform, as on any denial. Fail-closed processing does not depend on prior negotiation: a PDP MAY attach an obligation the PEP has not declared support for, and a PEP that cannot perform an attached obligation treats it as unfulfillable. This profile mandates correct obligation processing, not any particular obligation type. On a permit, obligation processing is recorded in Execution Evidence's obligation_outcomes ([I-D.draft-mcguinness-mission-runtime-evidence]), one entry per attached obligation, including a failure that suppresses an already- permitted action before release. A denial-side obligation is executed but not separately evidenced: it inherits the AuthZEN Obligations Profile's no-proof-of-execution posture [AUTHZEN-OBL], and a deployment that needs proof of its execution composes its own audit mechanism. A PDP claiming the Obligations feature profile (Section 2) MUST publish in its metadata supported_obligations array each obligation type it supports, the step-up obligation included, as the AuthZEN Obligations Profile requires [AUTHZEN-OBL]. A PEP MAY declare supported_obligations in the request context, as that profile defines. The declaration is advisory, a usability optimization on the fail-closed baseline: a PDP MAY use it to avoid attaching an obligation that is guaranteed to fail, sparing a round trip. A PEP that omits the declaration has an unknown capability, not support for no obligation type, and the omission never relaxes obligation processing. 6.2.2. Step-Up Obligation The PDP MAY attach the [AUTHZEN-OBL] step-up obligation (authentication-context properties such as acr_value and amr_values, per that profile) when Resource policy requires a stronger authentication context than the actor presents. It has two fulfillment shapes, both in the obligations lane: In-process step-up: The obligation is carried on a permit only when the PEP can establish the stronger authentication context without changing any credential-bound authorization input, for example a session re-verification or a local challenge. The PEP MUST achieve the required context before releasing the action's effect, and the original permit continues. OAuth step-up (RFC 9470): Obtaining a new access token changes credential-derived inputs, so this shape cannot fulfill an obligation attached to a permit. The PDP denies the original request with reason resource_policy and the step-up obligation carrying the challenge parameters, under the deny-side processing of Section 6.2.1. The PEP challenges the actor, obtains a new access token through RFC 9470 step-up authentication [RFC9470] at the protected resource, and a fresh evaluation follows under the new credential. The original permit is never continued across an RFC 9470 step-up. Denial does not turn step-up into governance: no access request is submitted and no approval record is created (Section 6.7). The requirement is Resource policy, never a Mission constraint. The OAuth binding's approval-time acr_values/max_age carriage is a requirement on the Approver at the approval event, a distinct fact from this per-action, token-associated requirement: it is neither carried on derived tokens nor evaluated per action, and the OAuth binding defines no per-action amr constraint ([I-D.draft-mcguinness-oauth-mission]). The resource server's own challenge for a weak or stale token-associated authentication is the RFC 9470 insufficient_user_authentication challenge ([I-D.draft-mcguinness-oauth-mission]); it composes with either shape but is not a Mission-defined denial reason. 6.2.3. Further Obligation Types and Advice A deployment MAY define further obligation types for genuine attached PEP actions, for example watermark (apply a provenance watermark to output before release), under the AuthZEN Obligations Profile's registration and extensibility rule [AUTHZEN-OBL]. An obligation MUST NOT carry human approval, role or relationship creation, Mission expansion, a task handle, residual policy, or retry instructions; evaluation_id and evaluation timestamps are decision facts outside it. Obligation-array ordering is not normative. This profile defines no advice collection. A future context.advice carrying safely-ignorable instructions is reserved; an obligation MUST NOT be marked optional or downgraded to advice, and permit request binding, permit expiry, use limits, evidence emission, step- up, and mandatory notification are never advice. 6.3. Permit Response When the PDP permits an action, it returns AuthZEN decision: true and the context the PEP needs to enforce the permit. The evaluation_id binds the response to the Decision Evidence. conditions carries the permit's decision conditions, including the parameter_digest that binds the permit to the request inputs the PDP evaluated; obligations is present only when a genuine obligation applies. The following is an example of a permit: { "decision": true, "context": { "evaluation_id": "dec_8K2nP4qV9rL3tY6sB1zN0eF7jB", "conditions": { "parameter_digest": "sha-256:WPVi6EnQ7H9Fh-qk9ADxmTg8zruOdVUX1esl-v3TfCI", "valid_until": "2026-11-02T08:15:00Z", "use_limit": 1 } } } 6.4. Denial Response A runtime denial is returned as decision: false with the context members of Section 6.1. The following is an example of a denial: { "decision": false, "context": { "evaluation_id": "dec_5WcR2mT8xN4qV7kB1sJ6eL9yPd", "reason": "stale_state" } } Alternatively, the PDP marks the same denial transient: { "decision": false, "context": { "evaluation_id": "dec_5WcR2mT8xN4qV7kB1sJ6eL9yPd", "reason": "stale_state", "next_action": "retry", "retry_after": 10 } } Malformed requests, authentication failures, or PDP processing errors that prevent evaluation MAY be returned as AuthZEN or transport-level errors; so is an in-scope request refused for missing Mission context (Section 5.20). A deployment MAY additionally carry [RFC9457] problem details for structured error information when the PDP is consumed over HTTP outside the AuthZEN envelope. 6.5. Runtime Denial Classification When the PDP denies a consequential action, the failure condition is one the runtime profile defines. This section gives the denial- reason identifier for each condition, carried in context.reason and in Decision Evidence's denial_reason: * out_of_authority: the action is not within the Authority Set, or not within the presented credential's authority (Section 5.6); Decision Evidence's authority_bound records which (Section 7). * approval_required: deployment or Resource policy requires an action-bound approval for this action ([I-D.draft-mcguinness-mission-runtime]) and no context.approval is presented. This is the first-evaluation reason. The runtime profile owns the approval artifact; the PEP carries a presented approval in context.approval (Section 5.16). This profile defines the value; [ARAP] uses the same spelling only in non-normative examples. The value alone does not make a denial requestable (Section 6.6). * approval_expired, out_of_scope, grant_pending, policy_denied, approval_unverifiable: ARAP's five registered re-evaluation reasons [ARAP], returned when a re-evaluation that presented context.approval fails a Mission or ARAP check, with the default next_action values Section 5.16 lists. * taint_context_missing: the deployment declares PDP-enforced taint for a class that applies to the action, as the harness profile defines ([I-D.draft-mcguinness-mission-harness]), and the decision request carries no context.taint (Section 5.13). * work_item_context_missing: the deployment claims work-item attribution for a class that applies to the action and the decision request carries no context.work_item (Section 5.15). A malformed one is the pre-evaluation refusal work_item_invalid instead. * stale_state: the Mission state the decision relies on is not current: the PEP-supplied freshness is outside the deployment's staleness bound (a freshness-bound violation), the PDP cannot establish the state it is required to establish, or the request's state version differs from the PDP's (Section 5.20). A Mission the PDP knows is not active is mission_inactive instead. Where a fresh read or reconciliation is expected to resolve the condition, the PDP SHOULD mark the denial transient with next_action: retry (Section 6.1); otherwise it does not recommend a retry. * view_inconsistent: the request's Mission id or issuer, or (when carried) authority_hash or policy_view_id, does not equal the committed values in the materialized policy view the PDP loaded, or in the Mission record it evaluates directly, so the request and the PDP disagree on which Mission or view is in force. This is a view inconsistency, not staleness. * mission_inactive: the Mission state is not active. * mission_binding_failed: in externally-established Mission binding mode ([I-D.draft-mcguinness-mission-runtime]), the PDP could not verify the supplied Mission reference against the acting credential under the binding's join, so no Mission is established. * actor_invalid: the required act chain is missing or malformed, so the PDP cannot establish the runtime actor context ([I-D.draft-mcguinness-mission-runtime]). * credential_invalid: token-derived credential facts supplied by the PEP are expired, inconsistent, or otherwise not usable for a runtime decision. * parameter_violation: parameters violate a constraint the PDP evaluated, the recomputed digest does not match, or a required parameter_digest is absent for a parameter-bound action. * duplicate_suppressed: the request's idempotency_key and operation identity (Section 5.18) match a prior claim whose outcome is unresolved (transient) or completed (terminal), so the PDP suppresses a duplicate execution of the same normalized action rather than issue a second permit (Section 5.7.1). * idempotency_conflict: the request reused an idempotency_key with a different operation identity (Section 5.18) than the prior claim it was made under (Section 5.7.1). This condition is terminal in every claim state, carries next_action: none (Section 6.1), and is never marked requestable (Section 6.6). * resource_policy: Resource policy refuses the action independently of Mission authority. * quota_exceeded: a metered runtime bound is exhausted. The runtime profile fixes the fail-closed posture for consumption bounds ([I-D.draft-mcguinness-mission-runtime]); the metering semantics and settlement exchange are defined by the experimental metering companion ([I-D.draft-mcguinness-mission-metering]). * unsupported_authorization_type: the action targets an authorization_details type the PDP does not understand or cannot enforce, so it refuses rather than guess the type's semantics ([I-D.draft-mcguinness-mission-runtime]). * constraint_unsupported: an applicable constraint or consumption bound on the entry is unrecognized or unmetered, so the PDP cannot enforce it and refuses ([I-D.draft-mcguinness-mission-runtime]). This is distinct from parameter_violation, which is a constraint the PDP evaluated and found violated. * resource_contract_drift: a resource contract applies to the action and the applicable entry's reference is missing or does not resolve under the resource's declared contract states ([I-D.draft-mcguinness-mission-runtime]), or the request's resource.properties.resource_contract is absent or differs from it (Section 5.3). The entry names the action, but the contract it was approved under no longer holds, so this is distinct from out_of_authority, and from capability_drift, which compares one capability's extracted definition. * history_not_satisfied: a deployment- or Resource-policy-required history predicate (Section 5.17) is not satisfied: the PDP established it false, could not establish it, could not consult its evidence store within the declared staleness bound, or does not recognize the presented predicate value. The per-predicate outcomes are in Decision Evidence ([I-D.draft-mcguinness-mission-runtime-evidence]), so the unsatisfied and unavailable cases stay distinguishable after the fact. Authentication step-up has no dedicated denial-reason value (Section 6.2.2): an in-process step-up carries the obligation on a permit, and an RFC 9470 step-up carries it on a resource_policy denial. That denial carries no next_action value: the step-up obligation is the actionable instruction, and the fresh evaluation that follows depends on the PEP obtaining a new token, not on a timed retry or a requestable workflow. A Resource-policy refusal with no step-up path is also resource_policy. A terminal policy refusal SHOULD carry next_action: none, except a denial carrying the step-up obligation, which carries no next_action; a denial the deployment exposes as requestable carries next_action: request with context.access_request (Section 6.1). This document defines no other denial-reason values. ARAP's five re- evaluation reasons are values ARAP owns and registers under its own extensibility rule [ARAP], carried here rather than redefined, and this profile does not require them to satisfy the naming rule below. A companion profile MAY extend the set by specification; an extension value MUST be either a collision-resistant name (following the Collision-Resistant Name guidance of Section 4.2 of [RFC7519]) or a name coordinated within this document family, so values cannot collide. A consumer of a denial reason, wherever it is carried (the Decision Evidence denial_reason, [I-D.draft-mcguinness-mission-runtime-evidence], or the response context.reason, Section 6.1), MUST treat an unrecognized value as a deny and MUST NOT attach any other semantics to it, mirroring the OAuth binding's open lifecycle state space ([I-D.draft-mcguinness-oauth-mission]). 6.6. Requestable Denials A deny is terminal for the attempted action: the agent does not proceed on a denial. A deny need not end the task, however. For an out_of_authority or approval_required denial, the PDP MAY mark the denial *requestable* by including a context.access_request object, composing this profile with ARAP [ARAP]. The PEP then submits an ARAP access request bound to the denied evaluation; an independent approver or policy adjudicates it, synchronously when policy auto- approves and otherwise asynchronously through ARAP's task handle; and on approval the PEP re-evaluates. This is the demand-driven counterpart to the pre-consented drawdown of the experimental progressive authorization companion ([I-D.draft-mcguinness-oauth-mission-progressive]): the agent starts narrow and requests the authority it discovers it needs, instead of holding it up front. Auto-approval is bounded the same way in-ceiling drawdown is ([I-D.draft-mcguinness-oauth-mission-progressive]): a deployment SHOULD rate-limit and anomaly-check synchronous auto-approval, and MUST NOT auto-approve a request for an approval_required denial in the irreversible, external-commitment, or privileged-administration classes without an independent approver, so a compromised agent cannot drive the request loop to escalate itself unattended. An ARAP approval is input context, not a bearer grant: the PDP remains authoritative at enforcement, so the PEP MUST obtain a fresh decision, and any resulting permit and evidence remain subject to this profile. The action-bound approval an approval_required denial calls for ([I-D.draft-mcguinness-mission-runtime]) is exactly such an approval, and ARAP's approval.id or signed approval.state is its carrier. An ARAP completion realizes one of two outcomes: * In-authority approval gate: the action is within the Authority Set, but policy requires a fresh approval. The approval attribute alone satisfies the re-evaluation; no Mission change occurs. * Missing authority (out_of_authority, or a subject lacking a role or relationship): the governance realization MAY be a Mission expansion ([I-D.draft-mcguinness-oauth-mission-expansion]), a policy-adjudicated in-ceiling expansion where the progressive companion is deployed ([I-D.draft-mcguinness-oauth-mission-progressive]), a role or relationship grant, or another authority-state change. The fresh evaluation observes that new state. Expansion is one realization, not ARAP's completion model. The following example shows a requestable denial: deployment policy requires an action-bound approval for the journal-entry write, no valid fresh approval is present, and the PDP marks the denial requestable: { "decision": false, "context": { "evaluation_id": "dec_7YbK4nQ9tR2xV6mL1sP8eJ3wZc", "evaluated_at": "2026-11-02T08:14:00Z", "reason": "approval_required", "next_action": "request", "access_request": { "endpoint": "https://requests.example.com/access-requests", "expires_at": "2026-11-02T09:14:00Z", "binding_token": "eyJhbGciOiJFUzI1NiIsInR5cCI6InBkcC1k..." } } } The example uses ARAP's PDP-signed binding_token form. The PEP refuses the action, submits the access request referencing the top- level context.evaluation_id, and re-evaluates only after approval. 6.7. Choosing the Lane Three lanes and three non-lanes classify a response by one governing test: what changes before access can proceed. Lane 1, obligations and advice: PEP work under the existing decision, whichever it is (Section 6.2). An obligation is mandatory and fail-closed (Section 6.2.1). Advice (reserved, Section 6.2.3) is safely ignorable and fail-open: a PEP that does not act on it proceeds unaffected. Neither confers authority, has an approver, or creates governance state. Lane 2, governance (ARAP): A governance process creates approval, authority, role, or relationship state: context.access_request outbound, context.approval on return, and a portable task handle tracking the workflow in between (Section 6.6). Lane 3, partial evaluation: Residual policy, with no workflow and nothing conferred. Partial evaluation is a non-goal: this profile returns decisions and instructions, never executable residual policy. parameter_attributes (Section 5.7), context.mission_history (Section 5.17), materialized policy views (Section 4), and embedded PDPs are inputs or deployment shapes, not residual evaluation. Composition with a future AuthZEN partial-evaluation profile is reserved, and a PDP MUST NOT return residual policy under this profile. Not a lane, the RAR remediation grain: insufficient_authorization and authorization_remediation of [I-D.draft-ietf-oauth-rar-metadata-remediation], as the OAuth binding adopts them into the family's remediation ([I-D.draft-mcguinness-oauth-mission]). The grain is payload: it describes authority that could be requested, MAY travel as payload within an ARAP request, and confers nothing by itself. A deployment MAY expose both grains on the same denial. The RAR- details grain names what authority to request, as actionable authorization_details routed per the OAuth binding's remediation routing ([I-D.draft-mcguinness-oauth-mission]); the AuthZEN grain names how the denial is escalated, into a governed access request that an independent approver or policy adjudicates before the PEP re-evaluates. Not a lane, transient denial: An outcome carrying retry guidance (next_action: retry, retry_after), not a remediation path: nothing is remediated, and the condition is expected to clear with time. Not a lane, decision conditions: Constraints of the decision itself (conditions, Section 6.1.1): declarative bounds on relying on it, evaluated at every use. They are not work the PEP performs; a condition the PEP does not recognize makes the permit invalid, where an unrecognized obligation makes the result an effective deny. Lanes coexist on one response: a requestable denial can carry a notification obligation under Lane 1's deny-obligation rule (Section 6.2.1), and an approval-satisfied permit can carry a watermarking obligation, without either changing the other's semantics. 6.8. Failure-Condition Coverage Every runtime failure condition, whether named in the runtime profile's failure-mode table or in its other normative requirements ([I-D.draft-mcguinness-mission-runtime]), surfaces through exactly one of four carriers in this profile: a Refusal Record for a PEP or PDP refusal before any PDP decision ([I-D.draft-mcguinness-mission-runtime-evidence]), a PDP denial (reason in the decision context, denial_reason in Decision Evidence), a permit obligation (Section 6.2), or an Execution Evidence error for a failure after a permit ([I-D.draft-mcguinness-mission-runtime-evidence]). A PDP denial MAY additionally carry an obligation; the OAuth step-up composition of Section 6.2.2 is the one case this document defines, and its row still names a single condition surfaced by one denial. The following table is the normative mapping for the conditions it names; extension identifiers remain governed by each carrier's extensibility rule. +=========================+===========+=================================+ |Runtime failure condition|Carrier |Identifier | +=========================+===========+=================================+ |Token validation fails |Refusal |token_invalid | | |Record | | +-------------------------+-----------+---------------------------------+ |Required mission claim |Refusal |mission_claim_missing | |absent |Record | | +-------------------------+-----------+---------------------------------+ |PEP-PDP channel |Refusal |channel_failure | |authentication or |Record | | |integrity fails | | | +-------------------------+-----------+---------------------------------+ |PDP unreachable |Refusal |pdp_unreachable | | |Record | | +-------------------------+-----------+---------------------------------+ |Mission state not |Refusal |state_unavailable | |establishable at the PEP |Record | | +-------------------------+-----------+---------------------------------+ |Required phase not |Refusal |phase_unavailable | |establishable at the PEP |Record | | |before the decision | | | |request | | | +-------------------------+-----------+---------------------------------+ |Committed effect a |Refusal |compensation_link_unavailable | |compensation offsets, or |Record | | |the Decision it executed | | | |under, not establishable | | | |at the PEP from retained,| | | |verified evidence before | | | |the decision request | | | +-------------------------+-----------+---------------------------------+ |Authoritative evaluation |Refusal |evaluation_context_unresolvable | |context of a covered |Record | | |operation not capturable | | | |at the PEP before the | | | |decision request | | | +-------------------------+-----------+---------------------------------+ |Presented credential's |Refusal |credential_authority_insufficient| |own authority does not |Record | | |cover the request, | | | |established before any | | | |decision request | | | +-------------------------+-----------+---------------------------------+ |Enforcement surface |Refusal |request_unsupported | |implements no such action|Record | | +-------------------------+-----------+---------------------------------+ |Supplied arguments fail |Refusal |request_invalid | |the PEP-established |Record | | |action's required input | | | |schema | | | +-------------------------+-----------+---------------------------------+ |Named target object not |Refusal |target_unresolvable | |resolvable at the |Record | | |enforcement surface | | | +-------------------------+-----------+---------------------------------+ |Operation target not |Refusal |target_unresolvable | |establishable as one |Record | | |unambiguous target at the| | | |PEP | | | +-------------------------+-----------+---------------------------------+ |Capability definition the|Refusal |capability_source_unresolvable | |PEP must present not |Record | | |resolvable before the | | | |decision request | | | +-------------------------+-----------+---------------------------------+ |Resource contract the |Refusal |resource_contract_unresolvable | |operation requires not |Record | | |resolvable before the | | | |decision request | | | +-------------------------+-----------+---------------------------------+ |Decision Evidence for a |Refusal |decision_evidence_unverifiable | |permit absent or not |Record | | |verifiable, so no relied-| | | |upon decision was | | | |obtained | | | +-------------------------+-----------+---------------------------------+ |In-scope request reaches |Refusal |mission_context_missing | |the PDP without the |Record | | |Mission decision context | | | +-------------------------+-----------+---------------------------------+ |Operation target absent, |Refusal |operation_target_invalid | |malformed, or outside the|Record | | |requesting PEP's | | | |enforcement scope at the | | | |PDP | | | +-------------------------+-----------+---------------------------------+ |No action class |Refusal |action_class_missing | |established for a |Record | | |mediated request where | | | |the Enforcement Scope | | | |Statement declares action| | | |classes, or a carried | | | |class other than the one | | | |the PDP's declaration | | | |assigns, at the PDP | | | +-------------------------+-----------+---------------------------------+ |action_phase outside the |Refusal |phase_invalid | |closed set, or absent |Record | | |where the trusted | | | |operation declaration | | | |requires a phase, at the | | | |PDP | | | +-------------------------+-----------+---------------------------------+ |compensates_evaluation_id|Refusal |compensation_link_invalid | |absent where action_phase|Record | | |is compensate, or | | | |malformed, at the PDP | | | +-------------------------+-----------+---------------------------------+ |Evaluation-context |Refusal |evaluation_context_invalid | |binding inputs missing, |Record | | |malformed, or | | | |incompatible with the | | | |declared operation | | | |binding, for a covered | | | |operation, at the PDP | | | +-------------------------+-----------+---------------------------------+ |context.work_item present|Refusal |work_item_invalid | |but not an object with |Record | | |exactly the string | | | |members issuer and id, at| | | |the PDP | | | +-------------------------+-----------+---------------------------------+ |Action outside the |PDP denial |out_of_authority | |Authority Set (including | | | |an invoked identity | | | |outside the approved set | | | |with no recorded source | | | |binding), or the request | | | |would broaden it | | | +-------------------------+-----------+---------------------------------+ |Duration-lease renewal |PDP denial |out_of_authority | |names no lease the PDP | | | |issued or an ended | | | |activity, or fails | | | |continuity with the | | | |evaluation it renews | | | +-------------------------+-----------+---------------------------------+ |Resource policy requires |Obligation |step-up obligation | |a stronger authentication|on permit |([AUTHZEN-OBL]) | |context, satisfiable by | | | |in-process step-up with | | | |no change to credential- | | | |bound inputs | | | +-------------------------+-----------+---------------------------------+ |Resource policy requires |PDP denial |resource_policy with the step-up | |a stronger authentication|(with |obligation ([AUTHZEN-OBL]) | |context satisfiable only |obligation)| | |by a new access token | | | |(RFC 9470) | | | +-------------------------+-----------+---------------------------------+ |Resource policy requires |PDP denial |resource_policy | |a stronger authentication| | | |context and refuses | | | |outright, no step-up path| | | |available | | | +-------------------------+-----------+---------------------------------+ |Required action-bound |PDP denial |approval_required | |approval absent (first | | | |evaluation) | | | +-------------------------+-----------+---------------------------------+ |Re-evaluation's presented|PDP denial |ARAP's approval_expired, | |approval fails a Mission | |out_of_scope, grant_pending, | |or ARAP check | |policy_denied, or | | | |approval_unverifiable ([ARAP]) | +-------------------------+-----------+---------------------------------+ |Mission state not |PDP denial |stale_state | |current: a freshness- | | | |bound violation, required| | | |state not establishable | | | |at the PDP, or a state- | | | |version mismatch | | | +-------------------------+-----------+---------------------------------+ |Request Mission id or |PDP denial |view_inconsistent | |issuer, or (when carried)| | | |authority_hash or | | | |policy_view_id, | | | |inconsistent with the | | | |loaded view or the | | | |directly evaluated | | | |Mission record | | | +-------------------------+-----------+---------------------------------+ |Mission not active, |PDP denial |mission_inactive | |including a passed | | | |expires_at | | | +-------------------------+-----------+---------------------------------+ |External Mission-binding |PDP denial |mission_binding_failed | |join verification fails | | | +-------------------------+-----------+---------------------------------+ |Work-item context absent |PDP denial |work_item_context_missing | |where work-item | | | |attribution is claimed | | | |for a class that applies | | | |to the action | | | +-------------------------+-----------+---------------------------------+ |Required act chain |PDP denial |actor_invalid | |missing or malformed | | | +-------------------------+-----------+---------------------------------+ |Credential facts expired |PDP denial |credential_invalid | |or inconsistent | | | +-------------------------+-----------+---------------------------------+ |Credential authority |PDP denial |credential_invalid | |absent, or carrying an | | | |entry the PDP cannot | | | |evaluate in full | | | +-------------------------+-----------+---------------------------------+ |No entry of the presented|PDP denial |out_of_authority | |credential's own | | | |authority names the | | | |action's resource and | | | |action, established by | | | |the PDP | | | +-------------------------+-----------+---------------------------------+ |Entries of the presented |PDP denial |parameter_violation | |credential's own | | | |authority name the | | | |action, but its | | | |parameters violate a | | | |constraint on each | | | +-------------------------+-----------+---------------------------------+ |Parameter constraint |PDP denial |parameter_violation | |violated, PDP digest | | | |mismatch, or required | | | |digest absent | | | +-------------------------+-----------+---------------------------------+ |Idempotency key and |PDP denial |duplicate_suppressed | |operation identity match | | | |a prior unresolved or | | | |completed claim, except a| | | |retransmitted settled | | | |duration-lease renewal, | | | |which receives its permit| | | |while valid | | | +-------------------------+-----------+---------------------------------+ |Idempotency key reused |PDP denial |idempotency_conflict | |with a different | | | |operation identity | | | +-------------------------+-----------+---------------------------------+ |Resource policy refuses |PDP denial |resource_policy | |the action | | | +-------------------------+-----------+---------------------------------+ |Consumption bound |PDP denial |quota_exceeded | |exhausted | | | +-------------------------+-----------+---------------------------------+ |Unsupported |PDP denial |unsupported_authorization_type | |authorization_details | | | |type | | | +-------------------------+-----------+---------------------------------+ |Unrecognized or unmetered|PDP denial |constraint_unsupported | |constraint | | | +-------------------------+-----------+---------------------------------+ |Resource contract |PDP denial |resource_contract_drift | |reference missing where | | | |required, or not | | | |resolvable, established | | | |by the PDP | | | +-------------------------+-----------+---------------------------------+ |Presented approval bound |PDP denial |ARAP's out_of_scope ([ARAP]) | |to a resource contract | | | |reference other than the | | | |request's | | | +-------------------------+-----------+---------------------------------+ |Policy-required history |PDP denial |history_not_satisfied | |predicate unsatisfied or | | | |not establishable | | | +-------------------------+-----------+---------------------------------+ |Effective parameters |Execution |parameter_mismatch | |differ at the executing |Evidence | | |PEP in a parameter no | | | |claimed evaluation- | | | |context binding declares | | | |as a fact | | | +-------------------------+-----------+---------------------------------+ |Permit's bound phase |Execution |phase_mismatch | |differs from the |Evidence | | |crossing, is missing or | | | |malformed where a phase | | | |is required, or cannot be| | | |established at use | | | +-------------------------+-----------+---------------------------------+ |A claimed evaluation- |Execution |target_drift | |context binding differs |Evidence | | |or cannot be established | | | |at use | | | +-------------------------+-----------+---------------------------------+ |Operation target at use |Execution |target_drift | |differs from the |Evidence | | |request's | | | |operation_target, or | | | |cannot be established at | | | |use | | | +-------------------------+-----------+---------------------------------+ |Named target object the |Execution |target_drift | |permit was evaluated for,|Evidence | | |or the effective | | | |parameters it determines,| | | |not establishable at use | | | +-------------------------+-----------+---------------------------------+ |Permit validity window |Execution |permit_expired | |passed at execution |Evidence | | +-------------------------+-----------+---------------------------------+ |Consumed single-use |Execution |permit_consumed | |identifier presented |Evidence | | |again | | | +-------------------------+-----------+---------------------------------+ |Different evaluation |Execution |operation_already_claimed | |identifier presented for |Evidence | | |an operation identity | | | |already claimed | | | +-------------------------+-----------+---------------------------------+ |Single-use authorization |Execution |operation_identity_conflict | |identifier turned to a |Evidence | | |different operation | | | |identity | | | +-------------------------+-----------+---------------------------------+ |Permit carries a decision|Execution |condition_unrecognized | |condition the executing |Evidence | | |PEP does not recognize | | | +-------------------------+-----------+---------------------------------+ |Store the executing PEP |Execution |consumption_unavailable | |takes single use in is |Evidence | | |unreachable | | | +-------------------------+-----------+---------------------------------+ |Capability definition the|Execution |capability_source_unresolvable | |permit was decided |Evidence | | |against not re-resolvable| | | |at use | | | +-------------------------+-----------+---------------------------------+ |Resource contract the |Execution |resource_contract_unresolvable | |permit was decided under |Evidence | | |not re-resolvable at use,| | | |or unequal to the | | | |contract the operation | | | |would be performed under | | | +-------------------------+-----------+---------------------------------+ |Obligation attached to a |Execution |obligation_unfulfilled | |permit could not be |Evidence | | |fulfilled | | | +-------------------------+-----------+---------------------------------+ |Operator or safety |Execution |kill_switch | |control suppressed |Evidence | | |execution | | | +-------------------------+-----------+---------------------------------+ |Permit's work-item |Execution |work_item_mismatch | |condition differs from |Evidence | | |the work item attributed | | | |to the executing request,| | | |or one is present without| | | |the other | | | +-------------------------+-----------+---------------------------------+ Table 2 The following rules separate conditions that share a carrier or an identifier: * Phase: An at-use phase failure is recorded as Execution Evidence with outcome suppressed and error phase_mismatch, with no effect released. A phase unestablishable before the decision request is instead a local configuration failure at the PEP, refused before any decision request and recorded as a Refusal Record with phase_unavailable ([I-D.draft-mcguinness-mission-runtime-evidence]); it does not invent a PDP denial or an Execution Evidence record for a permit never obtained. * Compensation: A PEP that cannot establish the committed effect a compensation offsets refuses before any decision request, with compensation_link_unavailable. The compensated action's permit presented at a compensating crossing is a post-permit failure of its phase binding, phase_mismatch as Execution Evidence, never a Refusal Record. * Unusable inputs: phase_invalid, compensation_link_invalid, evaluation_context_invalid, work_item_invalid, and action_class_missing are PDP refusals before evaluation. They describe decision inputs the PDP cannot use, not an evaluated policy denial, and a PDP claiming only the Decision Base records them in its own internal form (Section 2). * Target: A target-determining argument that changed under a parameter-bound permit is a parameter deviation, parameter_mismatch; target_drift records an operation target that differs while the bound parameters do not. * Context: under a claimed Evaluation-Context Binding a moved declared fact can change the parameter digest too. The recorded error is target_drift unless the context is established and an agent-supplied parameter changed, and every observable digest pair is kept ([I-D.draft-mcguinness-mission-runtime-evidence]). evaluation_context_unresolvable is the PEP's capture failure before the decision request, and evaluation_context_invalid the PDP's refusal of the inputs it received. * Duplicates: The duplicate_suppressed and idempotency_conflict rows key on the same (idempotency scope, idempotency_key) claim (Section 5.7.1), and whether the new request's operation identity (Section 5.18) equals the claim's decides between them. Both are PDP denials, refused before a permit exists. Their post-permit counterparts, reached at use under a permit the PEP already holds, are operation_already_claimed and operation_identity_conflict, decided by the same equality. A duplicate detected at use is never carried as a PDP denial reason, and a duplicate refused before a decision is never carried as an Execution Evidence error. Both post-permit errors remain distinct from permit_consumed, the same evaluation identifier presented twice. * Capability source: capability_source_unresolvable appears on two carriers, and the carrier records whether the resolution failure happened before the decision request or at use; a resolution failure is never reported as parameter_violation or parameter_mismatch. * Resource contract: resource_contract_unresolvable likewise appears on two carriers, the carrier recording whether the PEP found the contract unresolvable before the decision request or at use; resource_contract_drift is the PDP's denial between them. A contract failure is never reported as a parameter deviation or as target_drift, even when the parameters and the target are unchanged. 6.9. Permit Binding in Split Topologies The requesting component and the executing PEP MUST be the same enforcement identity, on the same mutually authenticated channel that requested the permit. A permit is valid only on that channel and identity, and MUST NOT be relayed to another component as a bearer grant. Signed Decision Evidence ([I-D.draft-mcguinness-mission-runtime-evidence]) MUST NOT be accepted as authorization to act: it is retrospective evidence of what the PDP decided, never a live credential a distinct component presents to execute. A deployment whose requesting component and executing component differ is out of scope for this profile. That split is reserved for a future split-execution profile, which would define an execution authorization artifact distinct from evidence: audience-bound to the executor, short-lived, and one-use. The single enforcement identity owns the consumed-identifier store and honors the permit's valid_until and use_limit (Section 6.1.1) under the runtime profile's consumed-identifier rules ([I-D.draft-mcguinness-mission-runtime]), so a permit cannot be executed twice or after its lease. A PEP permit cache MUST key on the request's cache key (Section 5.18), which already excludes observation telemetry by definition, so a per-request timestamp cannot prevent a cache hit. The materialized view identifier is not echoed on the response (Section 4), so a cache cannot key on it; the permit's own valid_until and use_limit bound reuse instead, so a permit issued against a view the PDP has since retired cannot outlive its lease. A cached permit cannot be reused for a request whose cache key differs. At each use of a permit, cached or not, the executing PEP establishes the operation target again from its operation contract and the arguments it executes with. It MUST NOT release an effect on a target other than the request's operation_target, and requests a new decision for the changed operation instead (Section 6.8). For a parameter-bound action, the arguments that determine the target are effectful parameters, so the digest reverification also covers them ([I-D.draft-mcguinness-mission-runtime]). A parameter-bound class keys on parameter_digest, and a high- consequence permit is single-use, so a permit cache amortizes only repeat-identical, non-single-use actions: reads and idempotent re- checks. Consequential writes take one evaluation per action, and a deployment sizes PDP capacity and placement to that rate, not to a cache hit ratio ([I-D.draft-mcguinness-mission-runtime]). 6.10. Evaluation Identifier Propagation The resource request a permit authorizes is commonly served by a resource server that did not see the PDP exchange. The PEP SHOULD propagate the permit's evaluation_id to the resource request in the Mission-Decision request header field (Section 11); the field value is the evaluation_id, whose ABNF (Section 6.1) is field-value-safe. A PEP that propagates the field MUST send the resource request carrying it over TLS 1.2 or later; the TLS this profile requires of the PDP endpoint (Section 9) does not cover that hop. Where the deployment signs resource requests, the signature MUST cover the field. A resource server that logs the received evaluation_id with the access it serves joins that access to the decision that permitted it: Decision Evidence, Execution Evidence, and the access log share one identifier, with no timestamp correlation. This adds the evaluation identifier to the OAuth binding's recommendation that a resource server log the mission claim's id and the token jti with each decision ([I-D.draft-mcguinness-oauth-mission]). The field is a correlation aid, not an authorization. Its presence or value grants nothing, the resource server's token validation and PEP obligations are unchanged, and a resource server MUST NOT treat it as a permit; the permit-binding rules (Section 6.9) govern. 7. Runtime Evidence This section defines the Runtime Evidence feature profile (Section 2). A PDP and PEP that claim it emit the Decision Evidence, Execution Evidence, and Refusal Records of the runtime evidence companion ([I-D.draft-mcguinness-mission-runtime-evidence]) for every decision, execution outcome, and pre-decision refusal this profile produces. A PEP/PDP pair claiming only the Decision Base satisfies the runtime profile's baseline evidence duty in its own internal form instead ([I-D.draft-mcguinness-mission-runtime]). The response's evaluation_id (Section 6.1) is the correlation key; each record carries its own record identifier (evidence_id, execution_id, or refusal_id). The runtime evidence companion defines every core Decision Evidence and Execution Evidence member. Under that companion's coordinated-extension rule, this profile defines two Decision Evidence extension members: mission_history, the policy- selected history predicates and their outcomes (Section 5.17), and authority_bound (below). It also carries coordinated extension members owned elsewhere, for example taint, protected_data and work_item (owned by the harness profile, [I-D.draft-mcguinness-mission-harness]) and capability_source (owned by Mission Capability Binding, [I-D.draft-mcguinness-mission-capability-binding]). authority_bound: A string, recorded at the top level of Decision Evidence: credential when the presented credential's own authority (Section 5.6) decided the deny, mission when the Mission's own authority did. It is REQUIRED on an out_of_authority or parameter_violation deny that one of those two bounds decided, and absent on every other decision, including a permit and a deny by Resource policy. When both bounds exclude the action, the PDP records the one it evaluated first. This profile maps its wire members into Decision Evidence as follows, in addition to the members other sections require the PDP to record: * mission, subject, resource, action, audience, credential, parameter_digest, obligations, taint, protected_data, and work_item are populated from the correspondingly named members of the PDP request and response (Section 5, Section 6). * mission_history records the predicates the PDP evaluated and their outcomes (Section 5.17); the PDP computes it, never copying it from context.mission_history. * conditions is populated from the response context.conditions (Section 6.1.1). * denial_reason carries the value returned in context.reason (Section 6.5). * action_class, action_classes, class_source, and mission_state_version are not on this profile's wire: the PDP computes them and records them directly. The runtime evidence companion's integrity envelope is this profile's one signing convention for evidence objects: the default envelope format is jws-compact, the protected typ is the one value that envelope fixes, and the protected cty names the record's own registered media type ([I-D.draft-mcguinness-mission-runtime-evidence]). 7.1. Evaluation Request Digest Input Under this profile, the input to evaluation_request_digest ([I-D.draft-mcguinness-mission-runtime-evidence]) is the complete JSON request body the PEP submitted to the evaluation or evaluations endpoint: every member it carried, extension members included, before the PDP applies any default or enrichment. HTTP header fields and transport framing are not part of it. The digest is sha-256: followed by the base64url, unpadded, encoding of the SHA-256 of the body's JCS [RFC8785] canonical bytes, so member order and whitespace do not change it, while an omitted member and a member set to null yield different digests. For a batch (Section 5.22), the input is the batch request body as submitted, before its default members are merged into the items; each item's Decision Evidence carries that one digest. The digest commits to its input; it does not hide an input a party can guess. For example, for this abbreviated consequential-read request, submitted without a parameter_digest: { "subject": { "type": "user", "id": "user_3p2q8mN1a0kV7tR" }, "resource": { "type": "journal-entry", "id": "je_2026Q3_inv_8421", "properties": { "audience": "https://erp.example.com" } }, "action": { "name": "journal-entries.read" }, "context": { "mission": { "id": "msn_8RfX2Lqv9TqMv4z7sA2bN1k0YpEdHc9-", "issuer": "https://as.example.com" } } } the value is: evaluation_request_digest = sha-256:v_QKMU8dffVDdoAnQkBenjWgtp5U_ffH5ORwT39iaoo 7.2. Execution Evidence Requirement Independent of anything the PDP returns, a PEP that claims Runtime Evidence MUST emit one Execution Evidence Object ([I-D.draft-mcguinness-mission-runtime-evidence]) for the final disposition of every consequential permit: completed, failed, or suppressed before release. This is a profile-level PEP requirement, not an obligation. A PEP claiming only the Decision Base satisfies the runtime profile's baseline execution-evidence duty in its own internal form instead, as it does for Decision Evidence and Refusal Records (Section 7). The runtime evidence companion defines the record's delivery contract, keyed on its stable execution_id. A permit whose action is never released still ends in Execution Evidence with outcome suppressed; only a refusal before any PDP decision is a Refusal Record. If evidence emission fails after an irreversible effect, the effect stands: nothing retroactively permits or denies it. The PEP MUST record the emission failure durably and surface it on its operational alarm path, and the deployment's audit posture treats the execution as unevidenced. Where the deployment's transaction assurance depends on that later evidence ([I-D.draft-mcguinness-mission-runtime]), the PEP MUST durably reserve evidence capacity for the execution, an evidence reservation, before acting, and MUST fail closed when the reservation cannot be made. 8. Conformance This profile adds AuthZEN-specific requirements to the runtime profile's enforcement contract. An implementation conforms to this profile only for the resources, action classes, and PDPs in the runtime enforcement scope it documents ([I-D.draft-mcguinness-mission-runtime]). Each requirement is normative in its owning section; the checklists below cite them without restating their mechanics. Mission Capability Binding ([I-D.draft-mcguinness-mission-capability-binding]) and the harness profile ([I-D.draft-mcguinness-mission-harness]) are normative references. The requirements that cite them bind only a deployment that adopts Capability Binding, or the harness capability the requirement names (PDP-enforced taint, for the taint members; exposure qualification, for the protected-data member; and work-item attribution, for the work-item member). A PEP conforming to the Decision Base MUST: * carry the Mission and actor decision inputs from validated token claims only (Section 5), and carry in resource.properties.operation_target the operation target it establishes from its trusted operation contract, refusing an operation for which it cannot establish one (Section 5.2); * supply action.properties.parameter_digest where required (Section 5.7), and, where the deployment adopts Mission Capability Binding, context.capability_source as that companion requires ([I-D.draft-mcguinness-mission-capability-binding]); * process obligations as Section 6.2.1 requires; * honor every decision condition present (parameter_digest, valid_until, use_limit, action_phase, evaluation_context_digest, resource_contract, work_item) at every use of the permit, applying its own deployment classification floor to that duty independent of any class the PDP applied internally (Section 6.1.1); * be the same enforcement identity as the requesting component, on the same mutually authenticated channel, and never relay a permit or treat signed Decision Evidence as authorization to act (Section 6.9); and * key permit caches on the request's cache key, and release an effect only on the operation target a permit was evaluated for (Section 6.9). A PEP additionally claiming the Runtime Evidence profile (Section 2) MUST emit a Refusal Record ([I-D.draft-mcguinness-mission-runtime-evidence]) for a pre-decision refusal, and an Execution Evidence Object for the final disposition of every consequential permit (completed, failed, or suppressed before release, Section 7.2). A PEP claiming only the Decision Base records a refusal or an execution outcome in its own internal form, never represented as this profile's portable schema (Section 2). A PDP conforming to the Decision Base MUST: * refuse an in-scope consequential request that lacks the Mission decision context, and perform the PDP-side consistency checks (Section 5.20); * match mission_resource_access entries against resource.properties.operation_target, never against audience, type or id (Section 5.1); * classify every denial per Section 6.5; * return the decision context of Section 6.1, including a conditions object on every consequential permit and an obligations array only when a genuine obligation applies (Section 6.2); and * advertise the urn:ietf:params:authzen:mission-runtime capability URN in its metadata capabilities array (Section 8.1). A PDP additionally claiming the Runtime Evidence profile (Section 2) MUST emit the records of [I-D.draft-mcguinness-mission-runtime-evidence] for every decision, and a Refusal Record for an in-scope request it refuses before evaluation (Section 5.20). A PDP claiming only the Decision Base records a decision or refusal in its own internal form, sufficient to satisfy the runtime profile's baseline evidence duty ([I-D.draft-mcguinness-mission-runtime]) without this profile's portable schema. A PDP claiming the Obligations feature profile publishes its supported obligation types in metadata as Section 6.2.1 requires. A deployment whose requesting component and executing component differ is out of scope for conformance to this profile (Section 6.9). Record producer and consumer conformance, including evidence signing- key verification and the classification of orphaned Decision Evidence, is defined by the runtime evidence companion ([I-D.draft-mcguinness-mission-runtime-evidence]). 8.1. Capability Discovery A PDP conforming to the Decision Base MUST advertise the full URN urn:ietf:params:authzen:mission-runtime, never the bare capability name. A PDP that additionally conforms to a feature profile MUST advertise that profile's own capability URN as well; a PDP MUST NOT advertise a feature-profile URN it does not conform to. Section 11.1 registers the URNs. A PEP MUST NOT rely on a feature profile whose URN the PDP's advertised capabilities does not include: an unadvertised profile fails closed exactly as an absent one does elsewhere in this document (Section 2). This in-band discovery complements, and does not replace, the deployment-established conformance gating this profile otherwise requires (Section 6.1.1). Capability advertisement is PDP-declared, never requester-selected: nothing in a PEP's request selects, enables, or negotiates a feature profile, and a requester cannot omit a required profile to downgrade enforcement (Section 2). 9. Security Considerations The runtime profile's Security Considerations ([I-D.draft-mcguinness-mission-runtime]) apply in full: placement and bypass, classification integrity, freshness and consumption honesty, Resource policy authority, TOCTOU and replay, and the limits of a compromised PEP or PDP. Threats specific to the evidence records are covered by the runtime evidence companion ([I-D.draft-mcguinness-mission-runtime-evidence]), and threats specific to capability-source binding by Mission Capability Binding ([I-D.draft-mcguinness-mission-capability-binding]). This section covers threats specific to this profile. 9.1. Unbound-Evaluation Downgrade A PDP that serves both Mission-bound and ordinary AuthZEN traffic has two policy surfaces, and the request selects between them by carrying context.mission. If an in-scope request that lost its Mission context (a PEP defect, or a path an attacker can influence) were evaluated against the non-Mission surface, a generic allow rule would stand in for the Mission evaluation this profile requires. The consistency requirement of Section 5.20 refuses such a request outright; deployments SHOULD additionally alert on in-scope requests arriving without Mission context, since each one is a PEP defect or a probe. 9.2. Access Request Service in the Trusted Base A deployment that composes with ARAP adds the Access Request Service to its trusted base: it adjudicates requestable denials and issues the approvals the PDP consumes as input. A compromised or misconfigured Access Request Service can auto-approve escalations, so it MUST be trusted, authenticated, and access-controlled like the PDP, and its auto-approval is bounded as Section 6.6 specifies. 9.3. Denial Oracle The denial-reason identifiers and any contributing_constraints are a decision oracle: an agent can probe them to map authority it does not hold. The PEP SHOULD minimize the denial detail it relays to the agent; a generic refusal suffices to stop the action, and the full reason and contributing constraints belong in evidence, not in the agent-facing response. To bound probing through the request loop, deployments SHOULD rate-limit access requests per Mission and surface request provenance to Approvers, so a compromised agent driving repeated requestable denials is visible to the humans adjudicating them. 9.4. Materialized View Fidelity A PDP that evaluates against a materialized view enlarging the Authority Set's bounds violates the bounded property of Section 4. The Authority Set that authority_hash commits is the upper bound. The id and issuer check, and policy_view_id where the PEP supplies it, let the PDP detect that the view it loaded does not match the Mission the PEP referenced and deny with view_inconsistent (Section 5.20). 9.5. Transport The PDP endpoint MUST be served over TLS 1.2 or later (TLS 1.3 RECOMMENDED). PEP-to-PDP authentication MUST be mutual, satisfying the integrity and mutual-authentication requirement the runtime profile places on the PEP-PDP channel ([I-D.draft-mcguinness-mission-runtime]). Evidence transport and storage requirements are the runtime evidence companion's ([I-D.draft-mcguinness-mission-runtime-evidence]). 10. Privacy Considerations The runtime profile's evidence-privacy guidance ([I-D.draft-mcguinness-mission-runtime]) applies in full. The runtime evidence companion covers the privacy properties of Decision Evidence, Execution Evidence, and Refusal Records, including their status as PII sinks, parameter exposure, and actor-chain correlation ([I-D.draft-mcguinness-mission-runtime-evidence]). Of this profile's two Decision Evidence extension members, mission_history (Section 7) carries behavioral history predicates and their outcomes. It inherits that guidance in full and receives the same PII-sink, access-control, and retention treatment as the rest of the Decision Evidence record, with no exemption. authority_bound names only an authority bound. This profile defines no other record content. 11. IANA Considerations This document requests the IANA actions in the following subsections. The runtime evidence companion, not this document, registers the Decision Evidence, Execution Evidence, and Refusal Record media types ([I-D.draft-mcguinness-mission-runtime-evidence]). The request and response members this document defines (Section 5, Section 6.1) are AuthZEN extension data and are not registered in an IANA registry. Members a companion adds are that companion's: context.capability_source ([I-D.draft-mcguinness-mission-capability-binding]), and context.taint, context.protected_data, context.work_item, and action.properties.external_communication ([I-D.draft-mcguinness-mission-harness]). The Mission-bound token claims this profile consumes are registered by [I-D.draft-mcguinness-oauth-mission]. This document registers no obligation type in the AuthZEN Obligation Types registry; the one obligation type it uses, step-up (Section 6.2.2), is registered by [AUTHZEN-OBL]. 11.1. AuthZEN Policy Decision Point Capabilities Registration This document requests registration of the following seven capabilities in the AuthZEN Policy Decision Point Capabilities registry established by [AUTHZEN]: one for the Decision Base and one for each feature profile (Section 2). * Capability Name: :mission-runtime * Capability URN: urn:ietf:params:authzen:mission-runtime * Capability Description: support for the Mission-Bound Runtime Enforcement AuthZEN profile's Decision Base. * Change Controller: the author of this document * Specification Document(s): this document * Capability Name: :mission-runtime-transaction-assurance * Capability URN: urn:ietf:params:authzen:mission-runtime- transaction-assurance * Capability Description: support for the Transaction Assurance feature profile (Section 2). * Change Controller: the author of this document * Specification Document(s): this document * Capability Name: :mission-runtime-evidence * Capability URN: urn:ietf:params:authzen:mission-runtime-evidence * Capability Description: support for the Runtime Evidence feature profile (Section 2). * Change Controller: the author of this document * Specification Document(s): this document * Capability Name: :mission-runtime-obligations * Capability URN: urn:ietf:params:authzen:mission-runtime- obligations * Capability Description: support for the Obligations feature profile (Section 2). * Change Controller: the author of this document * Specification Document(s): this document * Capability Name: :mission-runtime-arap * Capability URN: urn:ietf:params:authzen:mission-runtime-arap * Capability Description: support for the ARAP feature profile (Section 2). * Change Controller: the author of this document * Specification Document(s): this document * Capability Name: :mission-runtime-history * Capability URN: urn:ietf:params:authzen:mission-runtime-history * Capability Description: support for the History feature profile (Section 2). * Change Controller: the author of this document * Specification Document(s): this document * Capability Name: :mission-runtime-batch * Capability URN: urn:ietf:params:authzen:mission-runtime-batch * Capability Description: support for the Batch feature profile (Section 2). * Change Controller: the author of this document * Specification Document(s): this document 11.2. HTTP Field Name Registration This document registers the following in the "Hypertext Transfer Protocol (HTTP) Field Name" registry ([RFC9110]): * Field Name: Mission-Decision * Status: permanent * Reference: this document, Section 6.10 * Comments: none 12. References 12.1. Normative References [ARAP] OpenID Foundation, "AuthZEN Access Request and Approval Profile - Draft 1", 2026, . [AUTHZEN] OpenID Foundation, "OpenID AuthZEN Authorization API 1.0", 2026, . [AUTHZEN-OBL] OpenID AuthZEN Working Group, "AuthZEN Obligations Profile 1.0", 2026, . [I-D.draft-mcguinness-mission-capability-binding] McGuinness, K., "Mission Capability Binding", 2026, . [I-D.draft-mcguinness-mission-harness] McGuinness, K., "Mission-Aware Agent Harnesses", 2026, . [I-D.draft-mcguinness-mission-runtime] McGuinness, K., "Mission-Bound Runtime Enforcement", 2026, . [I-D.draft-mcguinness-mission-runtime-evidence] McGuinness, K., "Mission Runtime Evidence", 2026, . [I-D.draft-mcguinness-mission-substrate] McGuinness, K., "Mission Substrate Requirements", 2026, . [I-D.draft-mcguinness-oauth-client-instance-id] McGuinness, K., "OAuth 2.0 Client Instance Identifiers", Work in Progress, Internet-Draft, draft-mcguinness-oauth- client-instance-id-01, 2 November 2026, . [I-D.draft-mcguinness-oauth-mission] McGuinness, K., "Mission-Bound Authorization for OAuth 2.0", 2026, . [I-D.draft-mcguinness-oauth-mission-cross-domain] McGuinness, K., "Mission Cross-Domain Projection for OAuth 2.0", 2026, . [I-D.draft-mcguinness-oauth-mission-resource-access] McGuinness, K., "Mission Resource Access Profile for OAuth 2.0", 2026, . [I-D.draft-mcguinness-oauth-mission-status] McGuinness, K., "Mission Status and Lifecycle for OAuth 2.0", 2026, . [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, . [RFC3339] Klyne, G. and C. Newman, "Date and Time on the Internet: Timestamps", RFC 3339, DOI 10.17487/RFC3339, July 2002, . [RFC3986] Berners-Lee, T., Fielding, R., and L. Masinter, "Uniform Resource Identifier (URI): Generic Syntax", STD 66, RFC 3986, DOI 10.17487/RFC3986, January 2005, . [RFC6234] Eastlake 3rd, D. and T. Hansen, "US Secure Hash Algorithms (SHA and SHA-based HMAC and HKDF)", RFC 6234, DOI 10.17487/RFC6234, May 2011, . [RFC7519] Jones, M., Bradley, J., and N. Sakimura, "JSON Web Token (JWT)", RFC 7519, DOI 10.17487/RFC7519, May 2015, . [RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017, . [RFC8259] Bray, T., Ed., "The JavaScript Object Notation (JSON) Data Interchange Format", STD 90, RFC 8259, DOI 10.17487/RFC8259, December 2017, . [RFC8785] Rundgren, A., Jordan, B., and S. Erdtman, "JSON Canonicalization Scheme (JCS)", RFC 8785, DOI 10.17487/RFC8785, June 2020, . [RFC9110] Fielding, R., Ed., Nottingham, M., Ed., and J. Reschke, Ed., "HTTP Semantics", STD 97, RFC 9110, DOI 10.17487/RFC9110, June 2022, . 12.2. Informative References [I-D.draft-ietf-oauth-rar-metadata-remediation] Zehavi, Y., "OAuth 2.0 RAR Metadata and Error Remediation", Work in Progress, Internet-Draft, draft- ietf-oauth-rar-metadata-remediation-00, 23 August 2026, . [I-D.draft-mcguinness-mission-architecture] McGuinness, K., "An Architecture for Mission-Bound Authorization", 2026, . [I-D.draft-mcguinness-mission-metering] McGuinness, K., "Mission Consumption Metering", 2026, . [I-D.draft-mcguinness-mission-runtime-oauth] McGuinness, K., "Mission-Bound Runtime Enforcement: OAuth 2.0 Profile", 2026, . [I-D.draft-mcguinness-oauth-mission-approved-set-verification] McGuinness, K., "Mission Approved-Set Verification for OAuth 2.0", 2026, . [I-D.draft-mcguinness-oauth-mission-containment] McGuinness, K., "Mission Containment for OAuth 2.0", 2026, . [I-D.draft-mcguinness-oauth-mission-expansion] McGuinness, K., "Mission Expansion for OAuth 2.0", 2026, . [I-D.draft-mcguinness-oauth-mission-progressive] McGuinness, K., "Mission Progressive Authorization for OAuth 2.0", 2026, . [RFC9457] Nottingham, M., Wilde, E., and S. Dalal, "Problem Details for HTTP APIs", RFC 9457, DOI 10.17487/RFC9457, July 2023, . [RFC9470] Bertocci, V. and B. Campbell, "OAuth 2.0 Step Up Authentication Challenge Protocol", RFC 9470, DOI 10.17487/RFC9470, September 2023, . Appendix A. Document History [[ To be removed from the final specification ]] * History Context: the specification defining a predicate type states whether it selects an operation and which of its members names it. Where a type selects one, its mission_history entry carries that operation's identifier as action; an action_class_completed entry, or one for a type the PDP does not recognize, carries none (#2048). * Protected-Data Context: context.protected_data is a harness-owned extension point carrying the exposure determination that the harness profile's exposure qualification relies on. It is part of the authorization binding and recorded in Decision Evidence, and taint_context_missing is unchanged (#2044). * History Context: names work_item_prerequisite_completed, the harness profile's coordinated predicate type (#815). * Changes a requirement (#1994): a mediated request carried with an action class other than the one the PDP's trusted declaration assigns its operation and phase has no class established, and the PDP refuses it before evaluation as action_class_missing. * Taint, Work-Item and History Context: PDP-enforced taint, the external_communication member and work-item attribution apply when a class that applies to the action calls for them, and action_class_completed counts a permit when the named class is in its action_classes, or is its action_class where the record carries no action_classes (#1994). Every such permit needs a completion witness, a linked completed record whose effective digests equal the authorized ones; a digest deviation is none, a failed use or suppressed replay removes none, a single-use permit's completion other than the one that took its use creates none, a missing effective digest leaves completion undetermined, and the evaluation relies on a complete history read (#2029). * Credential Decision Context: requiring a credential's whole authority description to be evaluable is this profile's credential validation policy, stricter than Resource Access's alternative- entry rule and not a consequence of the prohibition on falling back to the Mission's authority; an entry the PDP evaluates and finds unsatisfied still lets another complete entry authorize; the interoperability cost is stated (#1806). * Changes a requirement (#956, D814): where the runtime Enforcement Scope Statement declares action classes, the PDP refuses a mediated request with no action class established before evaluation, records the Refusal Record action_class_missing, and never evaluates it under a default class or records one in Decision Evidence. * Work-Item Context: context.work_item is a harness-owned extension point the PEP populates only from the harness's authenticated attachment. A malformed member is refused before evaluation as work_item_invalid, and an absent one where work-item attribution is claimed for the action's class is denied work_item_context_missing. The member is part of the authorization binding and the cache key, a permit binds it as the work_item condition, and a permit for another work item is Execution Evidence suppressed / work_item_mismatch (#815). * Failure-Condition Coverage maps a covered operation's evaluation context not capturable at the PEP to the Refusal Record evaluation_context_unresolvable, and limits the parameter_mismatch row to a parameter no claimed binding declares as a fact; a context failure at use carries the error the runtime evidence profile's classification names; the context digest condition is recomputed under the descriptor the binding committed (#773). * Compensation link (#252). context.compensates_evaluation_id is REQUIRED on a compensate-phase request, established by the PEP from retained, verified evidence of the committed effect and its Decision, never from an agent argument. PDP-side consistency check 12 refuses a missing or malformed link before evaluation as compensation_link_invalid, and Failure-Condition Coverage maps it and the PEP's compensation_link_unavailable. * Failure-Condition Coverage maps a named target object, or the effective parameters it determines, not establishable at use to Execution Evidence target_drift, the post-permit counterpart of the pre-decision target_unresolvable (#786). * Owner-ruled follow-ups to the editorial pass (#1629). A PDP claiming the Obligations feature publishes its supported obligation types; a PEP's supported_obligations declaration is optional and its omission relaxes nothing. Obligation Processing holds the effective-deny rule for an unrecognized or unfulfillable obligation. A PEP claiming Runtime Evidence emits Execution Evidence for every consequential permit. Failure-Condition Coverage maps phase_invalid, evaluation_context_invalid, and phase_unavailable, and stale_state covers freshness-bound violations, state the PDP cannot establish, and state-version mismatch. approval_required is this profile's value. Metering owns the retried-request rule. Capability Binding and the harness are normative references, scoped to deployments adopting them. The history request object is context.mission_history. Transaction Assurance is required, and advertised, for the parameter-bound and high-consequence classes. A propagated Mission-Decision field travels over TLS. * Editorial and structural pass. Mission-Bound Authorization for OAuth 2.0 is called the OAuth binding throughout. Each repeated rule keeps one home, with pointers elsewhere: obligation processing, the idempotency claim (now its own subsection), ARAP's re-evaluation reasons, unrecognized denial reasons, evidence as authorization, and the step-up shapes. Decision conditions and obligations gain subsections. Mission Substrate folds into the Introduction and Mission Status Composition into the state observation section. The capability advertisement rules move from IANA Considerations to a Capability Discovery subsection of Conformance, and IANA Considerations drop the withdrawn composite obligation's history. Mission Runtime Evidence is a normative reference and RAR metadata remediation an informative one. No requirement changed. * Failure-condition coverage: a duration-lease renewal that names no lease the PDP issued or an ended activity, or fails continuity with the evaluation it renews, is denied out_of_authority; a retransmission of a settled renewal receives that renewal's permit while it is valid, not duplicate_suppressed (#1464). * Runtime Denial Classification and Failure-condition coverage: a resource contract reference missing where required, or not resolvable, is denied resource_contract_drift; the PEP records resource_contract_unresolvable before the decision request or at use (#249). * Resource Contract: resource.properties.resource_contract carries the reference the PEP will perform under, and the PDP requires it to equal the applicable entry's. The approval binds it through the approved Resource, and an approval under another reference is denied out_of_scope. The permit returns it as conditions.resource_contract, which the executing PEP compares before release (#249). * resource.properties.audience carries the runtime profile's boundary identifier, the boundary the PEP established, byte for byte: on OAuth, the protected resource's identifier, never an aud value or alias. The authorization binding, and so the cache key, names it. Operation-target matching is unchanged (#853). * resource.properties.operation_target carries the operation target the PEP establishes from its trusted operation contract. The PDP matches mission_resource_access entries in the credential's and the current effective authority against it, never against audience, type or id, and refuses a request whose target is absent, malformed or outside the PEP's enforcement scope (operation_target_invalid). The target is part of the operation identity, the authorization binding and the cache key; a target that changes at use is target_drift (#1276). * context.mission's id and issuer are the binding's Mission Reference identifier and Controller namespace in that binding's representation, with the OAuth realization unchanged, and issuer is distinct from context.credential.issuer. Terminology states the same (#1169). * mission_state_observation gains an OPTIONAL termination, the Mission's termination when state is terminated. The PDP rule, which keys only on exact active, is unchanged (#705). * Decision Evidence gains authority_bound, registered by this profile: credential or mission, the authority bound that decided an out_of_authority or parameter_violation deny (#825). * The failure-condition table maps supplied arguments that fail the PEP-established action's required input schema to the Refusal Record's request_invalid (#1106). * context.credential carries authority, the verified credential's own authorization_details, on every decision; the PDP enforces it as the credential authority bound, independently of the matched Mission entry, and denies credential_invalid when it is absent or not evaluable (#825). An action no credential entry names is out_of_authority; one whose parameters violate a constraint on every entry that names it is parameter_violation. * The RAR remediation grain cites the working-group successor draft- ietf-oauth-rar-metadata-remediation and defers its routing to the issuance profile's remediation routing instead of restating it. No requirement changed. * Mission-to-Policy Materialization describes the materialized policy view as one evaluation path: a PDP that evaluates the Mission's recorded authority directly meets the runtime profile's authority input without a view and omits policy_view_id. The PDP- side consistency checks compare the request against the loaded view or the directly evaluated Mission record, and the policy_view_id check applies only where a view is used. * Defined the input to evaluation_request_digest: the complete request body as submitted, extension members included, before any default or enrichment, with header fields and transport framing excluded; a batch carries the one digest of its body (#971). * The state observation's mission_status_issued_at is the RFC 3339 form of a Mission Status Response's iat, which freshness_at precedes by no more than the published maximum clock skew, and mission_status_expires_at and the state valid-through are its mission.fresh_until, not its exp, or a lease's end. Acknowledgments This document is the AuthZEN profile of Mission-Bound Runtime Enforcement and builds on the OpenID AuthZEN Authorization API. The author thanks the OpenID AuthZEN community and the Mission-Bound Authorization implementer community for feedback. Author's Address Karl McGuinness Independent Email: public@karlmcguinness.com