| Internet-Draft | Mission Metering | August 2026 |
| McGuinness | Expires 22 February 2027 | [Page] |
Mission-Bound Authorization for OAuth 2.0 bounds an agent's authority
by resources, actions, and constraints, and its runtime enforcement
profile evaluates each consequential action at the point of use.
Neither bounds how much of an approved authority a Mission may consume.
This document defines an experimental consumption-metering extension:
four cumulative consumption bounds a Mission Intent may carry
(max_budget, max_calls, max_duration, and max_egress_volume),
an exclusivity control (exclusive, separation of duty),
the runtime metering
semantics that enforce them (atomic check-and-decrement, reserve and
commit postures, duration leases, and settlement), and the AuthZEN wire
binding for lease renewal and settlement. A consumption bound is
consented at approval and enforced only by a runtime deployment that
implements this profile; a deployment that does not meter a bound must
refuse rather than silently ignore it.¶
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-metering.html. Status information for this document may be found at https://datatracker.ietf.org/doc/draft-mcguinness-mission-metering/.¶
Source for this draft and an issue tracker can be found at https://github.com/mcguinness/mission-bound-authorization.¶
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 22 February 2027.¶
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.¶
Mission-Bound Authorization for OAuth 2.0 [I-D.draft-mcguinness-oauth-mission] (the "issuance profile") bounds what an agent may do; the runtime enforcement profile ([I-D.draft-mcguinness-mission-runtime]) enforces each consequential action at the point of use. A long-running agentic task also needs bounds on how much: cumulative spend, call volume, and wall-clock activity. Those are not per-action constraints; they are counters that deplete across the life of the Mission, and enforcing them is a metering problem with reserve, commit, retry, and distributed-consistency concerns of its own.¶
This document defines that metering layer:¶
the consumption-bounds vocabulary a Mission Intent carries,¶
the metering semantics a runtime deployment enforces, and¶
the AuthZEN wire binding ([I-D.draft-mcguinness-mission-authzen]) for settlement and duration-lease renewal.¶
This document is optional and experimental: adopt it for evaluation, not as a stable interface. Metering cumulative bounds exactly under distributed decision points is a distributed-counting problem, and the reserve, commit, lease, and settlement machinery here is expected to evolve with implementation experience.¶
A deployment that does not implement this document carries no
consumption bounds on its Missions and is fully conformant to the
issuance and runtime profiles. The issuance profile's
max_derivations is not a consumption bound: it is enforced by the
issuing Authorization Server at each derivation and needs none of this
document ([I-D.draft-mcguinness-oauth-mission]).¶
The consent-integrity rule of Section 6.1 is the boundary that makes this safe to omit: a bound is rendered to an Approver only where it is actually metered.¶
This section is informative. Promotion moves this accounting model from optional evaluation toward the architectural baseline for any Mission carrying a consumption bound, so the bar is exercised machinery, not elapsed time. Meeting these criteria supplies implementation and deployment evidence for a deliberate decision, through the applicable IETF process, about changing this document's intended status or moving selected mechanisms into a Standards Track profile; it does not itself change either status, and no criterion, met or unmet, changes anything automatically. A catalog or inventory maturity label is distinct from an IETF stream decision and follows one rather than substituting for one. Implementation listings and test reports belong in an Implementation Status section per [RFC7942], removed before RFC publication; this section states the durable criteria.¶
Promotion is gated per feature by the applicable rows:¶
| Gate | Applies to |
|---|---|
| Retry, atomicity, settlement, and crash recovery | Every consumption bound |
| Exact or qualified topology behavior | Every claimed topology profile |
| Currency and debit/refund behavior |
max_budget
|
| Call classification and counting point |
max_calls
|
| Measurement, renewal, skew, and stop behavior |
max_duration
|
| Payload dereference and message accounting |
max_egress_volume
|
| Cross-Mission counter consistency | Aggregate bounds |
| Atomic latch and release |
exclusive
|
| Independent interoperation | Each promoted wire profile |
Each gate passes on named evidence:¶
Retry, atomicity, settlement, and crash recovery. For each promoted bound class, a reproducible test suite with no skipped cases demonstrates exactly-once accounting under Section 7.2 and Section 11.1: a retry under the same idempotency key or decision identifier consumes once; a redelivered settlement neither double-commits nor double-releases; a conflicting redelivery fails closed; a crash between reserve and commit resumes to a consistent balance; and every evidence state of Section 11.1.2 is exercised. This is the decrement-side counterpart of the exactly-once reservation machinery the family's Mission-creating surfaces carry for creation.¶
Exact or qualified topology behavior. Each claimed enforcement profile of Section 7.1 is demonstrated against its own guarantee: the Exact profile never over-consumes under concurrent near-exhaustion attempts; the Bounded-consistency profile stays within its published overshoot and staleness bound and reconciles orphaned reservations on the published cadence, including a non-idempotent reservation held until affirmative evidence of non-execution. Only Exact-profile evidence qualifies a bound for promotion as a baseline hard cap; Bounded-consistency evidence qualifies only the qualified form, the cap together with its published bound, rendered per Section 6.1. Artifacts: the Enforcement Scope Statement naming the profile per counter, and the reconciliation record.¶
Currency and debit/refund behavior (max_budget). Exact decimal
arithmetic with no floating-point accumulation; refusal on a
currency the counter does not carry, or the deployment's documented
conversion rule; debit, refund, and release paths returning exact
amounts.¶
Call classification and counting point (max_calls). The
call_class mapping of Section 6 exercised across implementations,
the same consequential action counted against the same class, and
the counting point proven: the count consumed at decision, the
disposition confirmed at settlement (Section 11.1.1).¶
Measurement, renewal, skew, and stop behavior (max_duration).
The duration rules of Section 7 exercised: bounded reservation or
lease for an unknown duration, renewal through
context.prior_evaluation_id ahead of expiry by the skew margin,
stop-or-new-permit before exhaustion, the in-flight handling of an
action that cannot be safely stopped, and measured_duration commit
with release of the unused reservation.¶
Payload dereference and message accounting
(max_egress_volume). The Operation Profile measurement exercised
for both bytes and messages, including a reference-typed
parameter measured by its dereferenced payload or its action class
excluded from a claimed bytes bound (Section 7).¶
Cross-Mission counter consistency (aggregate bounds,
Section 9). A lineage-keyed budget identifier and its
authoritative shared counter metering a root Mission and its Child
Missions in the counter's own consistency domain: a lineage-wide
quota_exceeded refusal from Child Mission consumption, a Child
Mission's released reservation returned to the shared counter, and
no consent surface or Enforcement Scope Statement rendering an
aggregate bound the deployed counter does not back. Artifacts: the
refusal's Decision Evidence carrying a metering entry
(Section 10) whose requested, remaining,
counter_scope, and counter_id prove the accounting invariant
rather than only the asserted refusal.¶
Atomic latch and release (exclusive). Concurrent consequential
actions matching different selectors of one group latch exactly one
selector, atomically with the permit; the affirmative-non-execution
release restores the group; refusals carry exclusivity_latched in
Decision Evidence; the latch domain is named in the Enforcement
Scope Statement (Section 8). This gate also holds the family's
adoption of the exclusive control into the issuance profile's core
controls vocabulary: that adoption follows the applicable gates of
this section being met, and until then the control is homed here.¶
Independent interoperation. Two distinct results, both required for each promoted wire profile, aligning with the implementation and interoperation reporting convention of [RFC7942]:¶
Operational experience: at least one deployment not operated by the implementer.¶
Interoperability: an independently implemented PEP and PDP exchange the profiled messages, including duplicate delivery and failure cases, under the settlement submission contract (Section 11.1.1).¶
Across every gate, promotion additionally requires the promoted wire
surface unchanged in name, shape, and semantics across two
consecutive published revisions of this document while the evidence
accumulated: the controls members of Section 6, quota_exceeded,
exclusivity_latched, context.prior_evaluation_id,
measured_duration, the metering evidence member
(Section 10), and the settlement submission contract of
Section 11.1.1.¶
Selective promotion means the applicable rows: promoting a
per-Mission bound class does not require the aggregate or
exclusive gates, and promoting the exclusive control does not
require the bound-class gates; every promotion includes the rows
that apply to each consumption bound it covers, its claimed topology
profiles, and its wire profile.¶
The criteria are sized to be satisfiable by a reference implementation, one deployment not operated by its implementer, and one independently implemented counterpart component; they demand exercised machinery and stable surfaces, not adoption counts.¶
This document depends normatively on the issuance profile and the
runtime profile and is not implementable alone. It defines its
consumption bounds as members of the Mission Intent controls object,
using the extension seam the issuance profile provides; they are
carried on the Mission and committed by intent_hash exactly as any
other Intent member. Metering is performed by the runtime profile's
PDP within a documented enforcement scope; this document adds metering
semantics to the runtime profile's decision contract and changes no
issuance protocol, though it does add consent-surface obligations on
the Mission Issuer (Section 6.1).¶
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.¶
A cumulative bound on Mission activity that depletes as the Mission is used, as distinct from a per-action constraint that is evaluated independently on each action.¶
This profile is defined against the Mission model rather than against OAuth 2.0 mechanics: it is a substrate-neutral consumer, and this section is its consumption declaration under the rule of Mission Substrate Requirements ([I-D.draft-mcguinness-mission-substrate]) that a substrate-neutral profile declare the kernel functions and optional capabilities it consumes.¶
From the contextual-governance kernel it consumes the Mission
Identifier and issuer, the kernel's Mission Reference and Controller,
which key every consumption counter; and the immutable Approved
Context, of which the bounds are part as Mission Intent controls
members. The kernel requires no particular integrity anchors:
intent_hash is the OAuth binding's commitment to that approved
value, and it commits the bounds like any other Intent member
(Section 3).¶
It consumes these optional capabilities:¶
| Capability | Consumption | Scope of consumption |
|---|---|---|
| Lifecycle-Gated Authorization | required | Inherited scope: metering is performed by the runtime profile's PDP within a documented enforcement scope (Section 3), so every metered decision is already gated on the only-active-permits rule; this document adds counters to that gate and defines no second one |
| Structured Authority | conditional | Two consumers. A call_class value SHOULD be drawn from the actions identifiers of the entry's mission_resource_access, so the metered class maps to evaluated actions; a deployment that meters a coarser or cross-entry class defines that class's membership, and such a class is not interoperable (Section 6). The exclusive control consumes it even then: its selectors are interpreted in the identifier space of the approved Authority Set entries and compared with each consequential action, resource by equality and the invoked action by membership in actions, per group and per Mission; selector semantics are owned by this document (Section 6, Section 8) |
| State-Observable | not consumed | Mission state is established by the runtime decision this document adds counters to, under that profile's freshness rules, not by this document ([I-D.draft-mcguinness-mission-runtime]) |
| Monotonic Derivation | not consumed | A lineage-keyed budget identifier correlates a root Mission and its Child Missions to one shared counter (Section 9); lineage counters are correlation, not narrowing, and this document defines no no-broader-than comparison |
| Credential-Bound | not consumed | This document defines no binding of its own: enforcement composes through the runtime profile's Mission binding establishment step ([I-D.draft-mcguinness-mission-runtime]) |
| Independently Verifiable | not consumed | This document defines no verification artifact of its own; metered outcomes enter the runtime evidence records and inherit their verification ([I-D.draft-mcguinness-mission-runtime-evidence]) |
| Portable Evidence | not consumed | This document defines no evidence artifact of its own; metered refusals and settlement are carried in the runtime evidence records through the coordinated metering member (Section 10, [I-D.draft-mcguinness-mission-runtime-evidence]) |
The portability claim is capability-scoped rather than substrate-wide
for the reason the substrate's Capability Confusion consideration
states: every property this profile requires matches an explicit
capability claim and its scope, never the generic statement that a
binding supports Missions. Hosting under another binding also
requires a representation mapping: the bounds carried inside the
binding's own Approved Context and covered by its own commitment
mechanism, as the OAuth binding carries them in the Mission Intent
and commits them through intent_hash.¶
A Mission Intent controls object
([I-D.draft-mcguinness-oauth-mission]) MAY carry these members:¶
max_budget:OPTIONAL. An object. A hard cap on cumulative monetary
spend under the Mission. It carries the same {amount, currency}
shape as the max_amount Common Constraint the issuance profile
seeds, so the per-action cap and this cumulative cap read alike.
Has the members:¶
max_calls:OPTIONAL. An array of objects. Hard caps on the count of consequential call events. Each object has the members:¶
call_class:REQUIRED. A string. The named call class to meter. A call_class
value SHOULD be drawn from the actions identifiers of the
entry's mission_resource_access
([I-D.draft-mcguinness-oauth-mission]),
so the metered class maps to evaluated actions; a deployment that
meters a coarser or cross-entry class defines that class's
membership, and such a class is deployment-defined and not
interoperable, like a deployment-defined constraint. (Named
call_class rather than scope to avoid collision with the OAuth
scope parameter and claim.)¶
count:REQUIRED. An integer. 1 or greater.¶
max_duration:OPTIONAL. A string. An ISO 8601 duration (for
example, PT8H), matching the duration rule in Appendix A of
[RFC3339], bounding cumulative wall-clock consequential activity
under the Mission. It is distinct from the Mission's expires_at,
which bounds issuance rather than activity.¶
max_egress_volume:OPTIONAL. An object with one or both members bytes and
messages, each an integer, 1 or greater: hard caps on cumulative
egress under the Mission across consequential
external-communication and external-commitment actions
([I-D.draft-mcguinness-mission-runtime]), as the total size in
bytes of those actions' bound payload parameters and the count of
such actions. It bounds the volume of within-scope laundering; it
does not detect it.¶
exclusive:OPTIONAL. An array of exclusivity groups, each an array of two or
more selectors. A selector is an object with resource (REQUIRED,
a string) and actions (OPTIONAL, an array of strings); it matches
a consequential action whose resource equals resource and, when
actions is present, whose invoked action is within it. Within a
group, the selectors name authority the Approver consents MUST NOT
be combined under this Mission (Section 8).¶
The bounds are carried on the Mission and committed by intent_hash.
They are not enforced by the Authorization Server at issuance; they are
enforced by the runtime layer at the point of use (Section 7).¶
Example Mission Intent controls carrying three of the four bounds
alongside
the issuance profile's members:¶
{
"controls": {
"acr": "urn:example:acr:mfa",
"max_derivations": 200,
"max_budget": { "amount": "5000.00", "currency": "USD" },
"max_calls": [
{ "call_class": "journal-entries.write", "count": 50 }
],
"max_duration": "PT8H"
}
}
¶
A consumption bound is part of what the Approver consents to. Where a deployment records Consent Evidence ([I-D.draft-mcguinness-oauth-mission-consent-evidence]), the rendered authority summary MUST include the consumption bounds the Mission carries.¶
A deployment MUST NOT render a consumption bound to the Approver as enforced unless the bound is within a runtime enforcement scope that meters it under this document. A Mission Issuer whose deployment does not meter a bound MUST refuse an Intent that carries it at intake. Such an Issuer MUST NOT render the bound as an enforced limit. A deployment that accepts the bounds into the Intent but does not meter them presents an unenforced promise at the consent surface.¶
Under a multi-PDP topology (Section 7.1), a consented hard cap the deployment renders as enforced MUST either:¶
be enforced under the Exact enforcement profile, for example as per-PDP sub-budgets that sum to the cap; or¶
be rendered with the named qualifier of the Bounded-consistency enforcement profile the counter operates under (Section 7.1).¶
Either way the Approver consents to the guarantee the deployment can meet rather than to a hard number it cannot.¶
Consumption bounds are enforced by the runtime profile's PDP ([I-D.draft-mcguinness-mission-runtime]), not at issuance:¶
max_budget: the PDP performs an atomic
reserve-or-charge against the remaining balance for each
consequential action and MUST refuse when the remaining balance is
insufficient.¶
max_calls: the PDP increments an atomic
counter for the named call_class and MUST refuse a call past count.¶
max_egress_volume: the PDP adds the action's bound payload size
and increments the action count atomically with the permit and MUST
refuse an action that would exceed either cap. Payload size is
measured over the parameter bytes committed by parameter_digest
([I-D.draft-mcguinness-mission-runtime]); the Operation Profile
defines the measurement so PDPs accumulate consistently. For a
reference-typed parameter, one that carries an attachment identifier
or URL whose referent egresses while the pointer is what
parameter_digest commits, the Operation Profile MUST measure the
dereferenced payload size, or the deployment MUST exclude such
actions from a claimed bytes bound. Counting the pointer bytes
alone understates the egress.¶
max_duration: the PDP accumulates the duration of consequential
activity it reserves, commits, or permits, under these rules:¶
The PDP MUST refuse once the accumulated total would exceed the bound.¶
The PDP MUST accumulate max_duration as the sum of per-action
measured durations, not the union of activity intervals, so
concurrent actions each count against the bound.¶
For an action whose duration is not known before execution, the PDP MUST either reserve a bounded maximum duration or issue a duration lease that expires unless renewed.¶
The PEP MUST stop the action or obtain a new permit before the reservation or lease is exhausted.¶
Lease boundaries MUST carry a clock-skew margin: the PEP renews or stops ahead of expiry by at least the deployment's skew bound, so a renewal in flight does not race the expiry it extends.¶
For an action that cannot be safely stopped mid-execution, lease
exhaustion is handled as an in-flight outcome under the
orchestration profile's dispatched_not_committed or unknown
classes ([I-D.draft-mcguinness-mission-orchestration]), not by
severing the action.¶
After execution, the PEP MUST report the measured duration so the PDP can commit actual use and release any unused reservation.¶
The Operation Profile defines how a single action's duration is measured so that PDPs accumulate consistently.¶
A per-entry constraints value that expresses a cumulative consumption
bound is metered the same way. When an applicable entry or the
Mission's controls carries such a bound, the PDP MUST meter use
against it. The PDP MUST refuse a consequential action that would
exceed it. The runtime profile's fail-closed rule stands beneath all
of this: an unmetered or unrecognized consumption bound MUST cause
refusal rather than silent pass-through
([I-D.draft-mcguinness-mission-runtime]).¶
The exactness of a consumption bound depends on the decision topology, and this profile does not overpromise. A deployment enforces each counter under one of two named enforcement profiles:¶
The check and decrement are atomic against the authoritative balance: a single serializing PDP for the counter, a shared linearizable counter, or per-PDP sub-budgets that are structurally exact because they sum to the cap. The bound never over-consumes; it is a hard cap.¶
Multiple or distributed PDPs (for example, Resource Server-hosted PDPs) share the counter without linearizable coordination, a distributed-counting problem. The deployment MUST publish, per bound class, the maximum overshoot and staleness it operates under (for example, a bounded reconciliation window), and the effective guarantee is the cap plus that published bound, not exact-to-the-call enforcement. The published qualifier is part of the bound's enforced semantic and is rendered per Section 6.1.¶
A deployment MUST name the enforcement profile per counter in its Enforcement Scope Statement, and MUST NOT advertise exact consumption enforcement it cannot meet under its chosen topology. The consistency bound is part of the runtime enforcement scope the runtime profile requires a deployment to document ([I-D.draft-mcguinness-mission-runtime]).¶
For a metered permit, the PDP and PEP MUST define retry and idempotency behavior. A retry of the same normalized action under the same idempotency key or single-use decision identifier MUST NOT consume the bound twice. Reuse of an idempotency key or decision identifier for a different normalized action MUST cause refusal. For irreversible actions and external commitments, a deployment MUST define whether metering is reserved before execution and committed after success, or committed before execution. It MUST NOT leave the decrement ambiguous: a reservation settles on the evidence state (Section 11.1.2), never on a generic failure signal. The reserve/commit posture fixes when consumption is charged; the evidence state fixes how the charge settles.¶
The exclusive control (Section 6) is not a consumption bound: it
is a stateful separation-of-duty rule enforced with the same
machinery. Within an exclusivity group, the first permitted
consequential action matching a selector latches the group to that
selector, atomically with the permit; for the Mission's remaining
lifetime the PDP MUST refuse a consequential action matching any
other selector of the same group. The latch is per group and per
Mission and is PDP-side operational state like a consumption counter
(Section 7).¶
The latch tracks execution, not permit issuance: a permit whose action is affirmatively not executed never combined the group's authority, so when Execution Evidence reports that outcome within the strongly consistent latch domain the PDP releases the latch, restoring monotonic narrowing rather than breaking it. Absent that affirmative non-execution the latch does not unlatch: narrowing by exercise is monotonic, like every other narrowing in the family.¶
The latch is exempt from the Bounded-consistency enforcement profile of Section 7.1. A counter degrades gracefully under a per-PDP sub-budget or a reconciliation window; a separation-of-duty rule violated once is violated permanently, and two PDPs can latch the same group to opposite selectors within the window. Therefore:¶
An exclusivity group MUST be enforced under the Exact enforcement profile of Section 7.1, in a single strongly consistent per-Mission latch domain (the runtime profile's Mission-sharding guidance makes the Mission the consistency unit, [I-D.draft-mcguinness-mission-runtime]).¶
The Bounded-consistency enforcement profile MUST NOT be applied to
exclusive.¶
The deployment MUST name the latch domain in its Enforcement Scope Statement.¶
Exclusivity turns the quarantine deployment pattern
([I-D.draft-mcguinness-mission-architecture]) into consented,
enforceable structure: an Approver can approve a Mission that may
read a sensitive store or communicate externally, but never both.
The groups are consented at the approval event, committed by
intent_hash with the other controls members, and rendered in the
consent disclosure (Section 6.1 applies unchanged).¶
In the AuthZEN binding, a refusal under a latched group is denied
with exclusivity_latched, an extension of the runtime denial set
under the AuthZEN profile's coordinated-extension conventions
([I-D.draft-mcguinness-mission-authzen]), and recorded as the
denial_reason in Decision Evidence. A PDP that cannot establish a
group's latch state fails closed for the actions the group covers,
per the runtime profile's availability posture.¶
The bounds of this document are Mission-keyed. A deployment MAY
additionally meter the same bound classes across Missions, keyed by
the Mission's subject, by the approved client_id, or by a
lineage-keyed budget identifier shared by a root Mission and every
Child Mission derived from it, so a fleet operator can cap what an
agent identity, a Subject, or a Mission's whole derivation lineage
consumes in total rather than per task. The counter semantics,
reserve/commit postures, and refusal behavior are unchanged; only the
key differs.¶
A lineage-keyed budget identifier and its authoritative shared counter are the only mechanism this document defines for a lineage-wide aggregate consumption bound: they meter across every Mission a derivation lineage contains, not within one Mission alone. A Child Mission's own derivation counter is independent of its parent's and bounds nothing beyond that Child Mission itself ([I-D.draft-mcguinness-oauth-mission-child-delegation]); absent a deployed lineage-keyed counter, no per-Mission counter, however many Missions in a lineage carry one, adds up to an aggregate bound on the lineage. This document is experimental (Section 2), so a deployment running only the stable issuance and runtime profiles has no lineage-wide aggregate bound in force at all. A deployment MUST NOT render, in an Enforcement Scope Statement or at any consent surface, a lineage-wide or subtree aggregate bound as in force unless a lineage-keyed budget identifier and shared counter meeting this section's requirements are actually deployed and metered.¶
An aggregate bound is deployment policy: it is carried on no single
Mission Intent, is committed by no intent_hash, and is disclosed
through the deployment's Enforcement Scope Statement rather than the
approval event. A refusal under an aggregate bound is carried as
quota_exceeded, and its Decision Evidence metering entry records
the aggregate key class in counter_scope and the counter in
counter_id (Section 10).¶
Aggregate keying crosses the family's per-Mission consistency domains: a subject-keyed or lineage-keyed counter is shared by every Mission the key spans, so it cannot be sharded by Mission Identifier and is provisioned as its own consistency domain ([I-D.draft-mcguinness-mission-runtime]).¶
A metered decision is auditable only where the evidence exposes the
counter state the decision asserted. This document defines one
coordinated evidence member, metering, carried in Decision
Evidence and Execution Evidence under the runtime evidence
companion's extension conventions
([I-D.draft-mcguinness-mission-runtime-evidence]).¶
metering:An array of one or more entries, one per bound the evaluation
metered. In a deployment claiming this profile, a Decision
Evidence record emitted for a metered evaluation, permit or
refusal, MUST carry it, and an Execution Evidence record MUST
carry it where the committed quantity is conveyed by consumed
(Section 11.1.1). Each entry has the members:¶
bound:REQUIRED. A string. The metered bound: a controls member name
(Section 6) or the applicable per-entry constraint name
(Section 7).¶
counter_scope:REQUIRED. A string. The counter's key class: mission for the
Mission-keyed bounds of this document, or subject, client,
or lineage for an aggregate bound (Section 9).¶
counter_id:REQUIRED. A string. A stable identifier of the counter, the same
value on every record the counter's decisions and settlements
produce. It is committed or pseudonymous: a sensitive key (a
subject or lineage identifier the record does not otherwise
carry) appears only as a commitment, for example a sha-256:
digest of the deployment's counter key, never raw.¶
requested:REQUIRED in Decision Evidence, absent otherwise. The quantity
the evaluation sought to consume, in the bound's native unit
shape: {amount, currency} for max_budget, an integer count
for max_calls, an ISO 8601 duration for max_duration, and
an object of bytes and messages for max_egress_volume.¶
reserved:CONDITIONAL. The same shape. The quantity reserved; REQUIRED in the Decision Evidence of a permitted action under a reserving posture (Section 7.2).¶
remaining:REQUIRED in Decision Evidence, absent otherwise. The same shape. The counter's remaining quantity after the decision; after a refusal, the unchanged remaining quantity the request exceeded.¶
consumed:CONDITIONAL. The same shape. Execution Evidence only: the actual quantity for the PDP to commit, per the bound-class conveyance rules of Section 11.1.1.¶
settlement_state:CONDITIONAL. A string. Decision Evidence of a permitted metered
action only: the decision-time posture, reserved or
committed (Section 7.2). The terminal disposition, commit,
release, hold, or conflict, is applied under
Section 11.1.2 and joined through evaluation_id; it is
not restated in the immutable record.¶
The member turns a metered refusal into a checkable claim rather
than an assertion: the refusal's entry shows requested exceeding
remaining on a named counter, and the counter's accounting history
is the join of the entries sharing its counter_id. A lineage-wide
refusal under an aggregate bound reads:¶
{
"metering": [
{
"bound": "max_budget",
"counter_scope": "lineage",
"counter_id":
"sha-256:t7RnQ2xV9kM4wB1sJ6eL3yP8cA5fH0dZu2gN7bXq4Ss",
"requested": { "amount": "25.00", "currency": "USD" },
"remaining": { "amount": "10.00", "currency": "USD" }
}
]
}
¶
Where the runtime deployment uses the AuthZEN binding ([I-D.draft-mcguinness-mission-authzen]), this section defines the wire representation of metering. It defines no new metering semantics and no new constraint. The requirements of this section and Section 11.1 apply only to a deployment that adopts the AuthZEN binding under that profile's conformance; for every other deployment the AuthZEN profile remains an informative reference.¶
When metering a bound would exceed it, the PDP MUST deny with
quota_exceeded ([I-D.draft-mcguinness-mission-authzen]) instead of
returning a permit. The PDP MUST record quota_exceeded as the
denial_reason in Decision Evidence.¶
Whether a metered permit is reserved at decision time and committed on settlement, or committed at decision time, follows the deployment's documented reserve/commit posture (Section 7.2); this binding fixes neither. In a batch (boxcar) evaluation, consumption metering applies per item in request order. The exactness of the bound is the consistency bound of Section 7.1, not a property of this wire binding.¶
The metering rules require the PEP to signal actual use so the PDP
commits consumption and releases any reservation. In the AuthZEN
binding, delivery of the Execution Evidence Object
([I-D.draft-mcguinness-mission-runtime-evidence]) to the PDP is
that commit-or-release signal: on receipt the PDP settles the linked
action's consumption per Section 11.1.2, keyed to the
Execution Evidence's evaluation_id.¶
The settlement transport is the runtime evidence companion's
Execution Evidence delivery contract, profiled rather than a second
channel: exactly one Execution Evidence Object exists per final
disposition of a permit, delivery is at-least-once, and the receiver
deduplicates on execution_id
([I-D.draft-mcguinness-mission-runtime-evidence]). This profile
adds:¶
Intake and audience. The deployment MUST name the settlement intake, the PDP evidence-submission path or the event or evidence API it profiles, in its Enforcement Scope Statement. The intake MUST feed the consistency domain that holds the counter (Section 7.1, Section 9).¶
Authentication. The intake MUST accept a settlement only as an
Execution Evidence Object in the evidence companion's integrity
envelope, verified against the emitter key, scope, and audience
binding claimed in the Enforcement Scope Statement
([I-D.draft-mcguinness-mission-runtime-evidence]).¶
Acknowledgement. The intake MUST acknowledge a settlement only after the commit or release is durably applied; an unacknowledged delivery is retried under the at-least-once contract.¶
Duplicate delivery. A redelivered record, the same
execution_id with the same settlement-relevant content, MUST be
acknowledged without applying the settlement again: the commit is
idempotent on evaluation_id, so a redelivered settlement neither
double-commits the consumption nor double-releases the
reservation.¶
Conflicting redelivery. A record naming an evaluation_id
whose settlement is applied but differing in settlement-relevant
content (outcome, measured_duration, or a metering entry),
or a second execution_id for the same evaluation_id, MUST fail
closed: the applied settlement is unchanged, the conflicting
record is retained as evidence, and the conflict is surfaced on
the deployment's audit and alarm path.¶
The committed quantity is conveyed per bound class:¶
max_calls: intrinsic. The class counter is consumed at decision
time, one consequential call per evaluation_id; settlement
confirms the disposition only.¶
max_duration: the Execution Evidence measured_duration member.¶
max_budget: the Execution Evidence metering entry's consumed
member, as {amount, currency} (Section 10).¶
max_egress_volume: the Execution Evidence metering entry's
consumed member, as bytes and messages, measured per the
Operation Profile rules of Section 7, including the
dereferenced-payload rule.¶
Where an action class defines no actual measure for a bound, the Operation Profile MUST state that the reserved quantity commits.¶
A reservation settles on the evidence state, never on a generic failure signal:¶
Affirmative non-execution (outcome suppressed): the action
was permitted but affirmatively not attempted. The PDP releases
the reservation and returns the quantity to the counter. Within
the strongly consistent latch domain this same state releases an
exclusivity latch (Section 8).¶
Completed execution (outcome completed): the PDP commits
the conveyed actual quantity, or the reserved quantity where the
class defines no actual measure, and releases any reserved excess.¶
Attempted but failed (outcome failed): not affirmative
non-execution, because the attempt may have consumed real
resources. The Operation Profile MUST define the commit-or-refund
rule per metered action class; absent a defined rule the
reservation remains charged and reconciles as an unknown outcome.¶
Unknown outcome (no evidence within the reservation lease): a reservation MUST carry a bounded lease so a crashed or abandoned reservation does not consume the budget permanently. On lease expiry without settlement the PDP reconciles the reservation through the runtime profile's orphaned-evidence process ([I-D.draft-mcguinness-mission-runtime]). For an idempotent or reversible action class, expiry releases the reservation and returns the budget; for a non-idempotent action class, expiry forces reconciliation or human review rather than release. An unsettled reservation remains charged against the bound until it is reconciled, and is released only on affirmative evidence of non-execution; timeout alone never releases a reservation for a non-idempotent action class.¶
Conflicting settlement: fail closed per Section 11.1.1. The applied settlement is unchanged; the conflict is audited, never adjudicated by the intake.¶
The operational consequence: a lossy evidence channel accumulates
reservations against max_budget and max_calls until the Mission
starves on quota_exceeded, a self-inflicted denial of service. A
deployment SHOULD run orphaned-evidence reconciliation on a published
cadence sized to its evidence-channel loss rate; the reconciliation
window it publishes is how long leaked budget stays leaked.¶
For a duration-metered action the PEP reports the measured duration in
the Execution Evidence measured_duration member, and the PDP commits
that duration against max_duration. A duration-lease renewal is a new
re-evaluation request that carries the prior permit's evaluation_id
in context.prior_evaluation_id, so the PDP continues the same metered
activity rather than opening a new reservation.¶
The PDP MUST verify that the renewal's Mission, subject, action, and
audience match the evaluation named by prior_evaluation_id. A
deployment sizes lease
intervals to amortize renewals: an interval materially shorter than
the action class's staleness bound adds decision load without
tightening the revocation cutoff. This exchange requires
one request member and one evidence member:¶
context.prior_evaluation_id:OPTIONAL. A string. Present on a duration-lease renewal request,
carrying the evaluation_id of the permit being renewed. Absent on
an initial request.¶
measured_duration (Execution Evidence):REQUIRED for a duration-metered action, otherwise absent. A string
containing an ISO 8601 duration (the duration rule in Appendix A of
[RFC3339]): the PEP's measured duration for the executed action.¶
A renewal repeats the evaluation-request envelope for the same
activity and adds context.prior_evaluation_id. Here a long-running,
duration-metered ledger reconciliation renews its lease before the
prior permit expires; the action is not parameter-bound, so no
parameter_digest is carried:¶
{
"subject": {
"type": "user",
"id": "user_3p2q8mN1a0kV7tR",
"properties": {
"iss": "https://idp.example.com"
}
},
"resource": {
"type": "ledger",
"id": "ledger_main"
},
"action": { "name": "reconciliation.run" },
"context": {
"mission": {
"id": "msn_8RfX2Lqv9TqMv4z7sA2bN1k0YpEdHc9-",
"issuer": "https://as.example.com",
"authority_hash":
"sha-256:l3KvZ4mP5x0wQrR6tY2nD9bM7sX1cF8gH2vJ4kE5pNQ",
"state": "active"
},
"actor": { "client_id": "s6BhdRkqt3" },
"credential": {
"issuer": "https://as.example.com",
"expires_at": "2026-11-02T09:14:00Z"
},
"audience": "https://erp.example.com",
"freshness": {
"mode": "fresh",
"freshness_at": "2026-11-02T08:44:00Z"
},
"prior_evaluation_id": "dec_0Rt5nB8xW2qK7mJ4vS1pL9eYc"
}
}
¶
A runtime deployment that claims this profile MUST:¶
meter every consumption bound a governed Mission carries per Section 7, within its documented runtime enforcement scope ([I-D.draft-mcguinness-mission-runtime]);¶
refuse a consequential action that would exceed a bound, and refuse on any bound it cannot meter;¶
enforce every consented exclusivity group with a latch atomic with the permit (Section 8);¶
where aggregate bounds are configured, meter and disclose them per Section 9;¶
emit the metering evidence member on the records
Section 10 requires;¶
publish its consistency bound under a multi-PDP topology (Section 7.1);¶
define and document its retry, idempotency, and reserve/commit posture (Section 7.2); and¶
where the AuthZEN binding is in use, implement the settlement exchange of Section 11.1.¶
A Mission Issuer in a deployment claiming this profile MUST carry the
consented bounds on the Mission record committed by intent_hash. It
MUST render them at the approval event per Section 6.1.¶
Consumption bounds limit the blast radius of a compromised or runaway agent in a dimension authority narrowing cannot: a Mission whose every action is individually authorized can still be drained by volume. Their enforcement, however, is only as good as the metering:¶
Unenforced bounds are consent theater. A bound rendered at approval but not metered anywhere misleads the Approver about the Mission's exposure. The consent-integrity rule (Section 6.1) exists for this: it forbids a deployment that cannot meter a bound from rendering it as enforced.¶
Distributed undercounting. Under a multi-PDP topology, an attacker who can spread actions across decision points exploits the consistency gap. The published consistency bound (Section 7.1) is the honest statement of that exposure; per-PDP sub-budgets bound it structurally.¶
Settlement honesty. The PDP commits what the PEP reports. Execution Evidence is integrity-protected and signed by the PEP ([I-D.draft-mcguinness-mission-runtime-evidence]); a compromised PEP can under-report duration or spend, which is within the runtime profile's trusted-base assumptions for PEPs.¶
Lease abandonment. An agent that stops renewing a duration lease and keeps acting is stopped by the PEP, which Section 7 requires to stop the action or obtain a new permit before the lease is exhausted.¶
Reservation starvation. An attacker who opens reservations and never settles them can consume a budget with no executed action, denying the Mission its remaining authority. The bounded reservation lease (Section 11.1) caps this: an unsettled reservation is reconciled on lease expiry rather than held indefinitely, and the idempotent-release path returns the budget.¶
Latch burning. Because the first matching action latches an exclusivity group, an injected agent can try to burn a group by driving the side it wants foreclosed, denying the Mission the other side. Releasing the latch on affirmative non-execution (Section 8) keeps an unexecuted attempt from foreclosing the group permanently.¶
Metering state (spend, call counts, activity durations) is a
fine-grained record of Mission activity over time. It SHOULD be
retained under the same access controls and retention windows as
runtime enforcement evidence
([I-D.draft-mcguinness-mission-runtime]), and disclosed in decision
responses only as refusals, not as remaining-balance oracles. The
metering evidence member records remaining inside those
access-controlled records (Section 10), never in a
decision response.¶
The refusal boundary is itself a coarse balance oracle: the point at which a bound flips from permit to refusal reveals the remaining margin. A deployment SHOULD bound this probing with the AuthZEN profile's denial-oracle controls ([I-D.draft-mcguinness-mission-authzen]), per-Mission rate-limiting of access requests and evidence-logging of request provenance, so a compromised agent mapping a balance by repeated probes is visible to the humans adjudicating it.¶
This document has no IANA actions. max_budget, max_calls,
max_duration, max_egress_volume, and exclusive are Mission
Intent controls
members defined by this
profile under the issuance profile's controls extension seam;
context.prior_evaluation_id is AuthZEN extension data carried per
the AuthZEN profile's conventions
([I-D.draft-mcguinness-mission-authzen]); measured_duration and
metering are coordinated evidence members under the runtime
evidence companion's extension conventions
([I-D.draft-mcguinness-mission-runtime-evidence]).¶
This document is part of the Mission-Bound Authorization work and defines its experimental consumption-metering layer.¶