| Internet-Draft | Command Authority Across Long Delay | November 2026 |
| Morrison | Expires 22 May 2027 | [Page] |
The operational-technology command authority profile defines a Command Authority Envelope with which an enforcement point decides whether a software agent may perform a control action, on whose authority, and with which human approval. This document profiles that envelope for networks where the authority and the enforcement point are separated by round-trip times from about 2.6 seconds to about two weeks, as in the connectivity scenarios of the IP in deep space use-case document. It says how each element of the envelope is evaluated offline, how expiry and maximum age are checked against a declared clock uncertainty, how the revocation latency budget and trust-anchor staleness are declared for each scenario, how a human approval signed before a command departs stays bound to one action, how an outcome that cannot be observed across a contact gap or a reboot is recorded, what an agent may do once the origin has been silent past the revocation latency budget, and how the envelope is carried over the Space Packet Protocol on UDP, CoAP with OSCORE, and HTTP/3. It adds one grant field, with which a principal writes in advance the actions an agent may take during that silence and the interval in which it may take them. It states which way an unknown revocation check fails for each action class, and it adds one declared bound, a second revocation latency budget for the grants that field marks, beside the one budget the profile it builds on has a deployment declare.¶
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 May 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.¶
Section 2.1 of [TIPTOP-UC] says that spacecraft are controlled from Earth and "should always be manageable from Earth". Section 3 of the same document says that commands "may be time-sensitive, or have an 'expiration date'". Neither section says who may command an asset, or how the asset checks.¶
[OT-COMMAND-AUTHORITY] answers that question for operational technology. It defines a Command Authority Envelope (CAE) carrying five bindings (agent identity, principal reference, authorisation grant, binding moment and audit record) and an enforcement point that refuses any agent-originated control action whose required bindings it cannot evaluate. Section 7 of that document requires the envelope to be verifiable offline against cached trust anchors, because many OT environments are segmented from public networks for long, declared intervals. A spacecraft, a habitat or a relay out of contact with Earth is the same case at a longer interval.¶
This document profiles the CAE for the connectivity scenarios of [TIPTOP-UC]. It applies the rules of [OT-COMMAND-AUTHORITY] as they stand and adds its own requirements on time (Section 5), on revocation posture (Section 6) and on authority during silence (Section 9). It says what each rule requires when the round-trip time to the party that authorises a command is from about 2.6 seconds to about two weeks.¶
[DISCONNECTED-TRUST] defines the disconnected trust model, the terms used here for held data, staleness bounds and unevaluable checks, and twelve requirements (R1 to R12) that any answer to its questions must meet. This profile is one such answer for command authority. It answers the five questions of Section 5 of that document for the actions it covers, question 3 in Section 9.¶
This document is an individual submission. It profiles the scenarios of [TIPTOP-UC] and is offered to the TIPTOP working group for review.¶
[SCSP] specifies a space command and telemetry security protocol. It already provides, at the frame layer, signatures over a command and its header, a per-key authorisation record naming the spacecraft, virtual channel, application process and command class a ground key may address, with a validity interval (Section 4.5 of [SCSP]), a Key Revocation List signed by a master key with a monotonic sequence number (Section 3.2), replay state kept per key in non-volatile memory (Section 4.4), signed telemetry for success and error (Section 4.7), and an acceptance window for deep-space frames that keeps propagation time apart from clock, ephemeris and margin terms (Section 6.1). This profile cites those mechanisms where it meets them and redefines none of them. [SCSP] authorises a ground controller key. It does not name the person on whose authority a command is sent, it carries no per-action human approval, and it has no state for an outcome that cannot be determined. A CAE can travel inside an SCSP frame (Section 10).¶
This document covers the five CAE bindings, time, revocation, binding moments, unresolved outcomes, authority during silence and carriage. It makes no change to IP, transport, routing or key management, and like Section 7 of [OT-COMMAND-AUTHORITY] it does not define an outer encoding of the CAE. Everything this profile requires lies on the authority path. The safety carve-out of Section 6 of [OT-COMMAND-AUTHORITY] applies unchanged.¶
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.¶
The terms CAE, enforcement point, principal, authorisation grant, binding moment, audit record, and the risk classes Observe, Adjust and State-change are those of [OT-COMMAND-AUTHORITY]. Held data, carried evidence, staleness bound, out of contact and unevaluable are those of [DISCONNECTED-TRUST]. The scenario labels CS, LF, LI, MF, MI and SC, and the terms minRTT, maxRTT and holdover, are those of [TIPTOP-UC].¶
The place where a command's CAE is assembled and its binding moment is signed, typically a mission operations centre.¶
The time between a CAE leaving the origin and its evaluation at the enforcement point, including any time it waits in a buffer for a contact.¶
A condition an enforcement point evaluates for a grant. It holds when the enforcement point holds a revocation status for the grant that states a time, and that status has aged past the revocation latency budget both on the enforcement point's clock and on a monotonic counter, determined as Section 9.1 requires.¶
A grant whose silence field (Section 9.2) makes it dormant until silence holds for the grant it names, within an interval it states.¶
The grant a contingency grant names. Its coverage and expiry cap the contingency grant, and its revocation ends the contingency grant once the enforcement point receives it.¶
A second revocation latency budget, declared by a deployment as the staleness bound for revocation status relied on by an admission under a contingency grant (Section 9.2).¶
Table 1 of [TIPTOP-UC] gives minRTT and maxRTT for each scenario. Section 5.1 of that document says that both components of the round-trip time are known in advance, from orbital dynamics and from the contact schedule, and that maxRTT "should be calculated using the actual contact schedule of the mission". The maxRTT column gives the values of Table 1 of [TIPTOP-UC].¶
| Scenario | maxRTT | What changes for this profile |
|---|---|---|
| LF | around 2.6 seconds | An approver can be asked while the command waits. Echo freshness (Section 10.2) is practical. |
| LI | around 2.6 seconds plus the maximum contact gap | A CAE can wait in a buffer for the gap. Expiries are sized to it. |
| MF | 48 minutes | Every check is made without the origin. |
| MI | 48 minutes plus the maximum contact gap, 6 hours 48 minutes in the use-case example | As MF, with the gap added to transit time. |
| CS | the two-way delay at the farthest planned distance, growing slowly | Declared values are revisited as the distance grows. |
| SC | the conjunction duration, about two weeks for Mars, plus the maximum two-way propagation delay, where communications are kept active across the conjunction | Section 4.6 of [TIPTOP-UC] records a moratorium during which no commands are sent. Revocation statements issued at the origin do not arrive until it ends. |
A deployment that claims this profile for a scenario MUST declare, for that scenario:¶
its maxRTT, computed from its own contact schedule as Section 5.1 of [TIPTOP-UC] describes, and the greatest transit time it expects;¶
its time source and that source's uncertainty (Section 5);¶
the maximum age of the freshness element for each risk class (Section 5);¶
its revocation latency budget and its maximum trust-anchor staleness (Section 6);¶
a staleness bound for each other kind of held data its checks rely on, as R2 and R12 of [DISCONNECTED-TRUST] require; and¶
where it admits contingency grants, its contingency budget (Section 9.2), the monotonic counter it uses to determine silence and whether that counter survives a restart (Section 9.1), and whether the path by which revocation status reaches the enforcement point passes through an agent or through anything an agent controls (Section 9.1).¶
This document sets none of these values.¶
Each binding is evaluated by the enforcement point from held data and carried evidence (R1 of [DISCONNECTED-TRUST]). Section 4 of [OT-COMMAND-AUTHORITY] applies as written. A required binding that is absent, malformed, expired, revoked or unverifiable causes refusal, and so does an unevaluable binding.¶
The agent's key material is held at the enforcement point with a declared staleness bound. Section 8 of [OT-COMMAND-AUTHORITY] observes that discovery through a reachable resolver is not available to a segmented plant and requires offline verification against cached trust anchors. A deep-space enforcement point is in the same position for at least the scenario's transit time.¶
The principal reference is evaluated against held data under its staleness bound. This profile adds no resolution step at the time of decision.¶
The grant's expiry (G3) is checked as Section 5 requires, and its revocation state (G5) as Section 6 requires. Coverage is evaluated against the specific action. A grant may stand until its expiry and cover more than one action, and each action is admitted only on its own coverage check and, where its class requires one, its own binding moment (Section 3.3 of [OT-COMMAND-AUTHORITY]). A grant may carry a silence field (Section 9.2).¶
Section 7.¶
The enforcement point records its decision and, before forwarding an accepted action, durably records the admission (Section 3.5 of [OT-COMMAND-AUTHORITY]). The admission record also names the held data the decision used, their ages, the staleness bounds in force (R11 of [DISCONNECTED-TRUST]) and the declared clock uncertainty. Offline verification of a record needs an inclusion proof and a signed checkpoint covering it, and a verifier unable to obtain a required covering checkpoint or consistency proof cannot declare those checks satisfied (Section 3.5 of [OT-COMMAND-AUTHORITY]). Out of contact, those checks stay unsatisfied until the verifier obtains what it needs.¶
A deployment MUST declare its time source and that source's uncertainty. An enforcement point MUST evaluate every maximum age and every expiry against that uncertainty, and MUST fail closed on authority when the uncertainty exceeds the window being checked. A timestamp that is absent or zero cannot be evaluated, and the action is then refused on authority.¶
The uncertainty is that of the enforcement point's own clock, as its deployment declares it. A value carried with a request does not reduce it. Let t be the enforcement point's clock reading and u its declared uncertainty. An expiry E is satisfied only if t + u is earlier than E. A maximum age M on a timestamp T is satisfied only if (t + u) - T is no greater than M and T is no later than t + u. A check that would pass at one instant within t - u to t + u and fail at another fails.¶
Section 6.2 of [TIPTOP-UC] says that holdover should be expressed as the time until a node's clock uncertainty exceeds what its applications and protocols can tolerate, and that "lifetimes should take this clock uncertainty into account". Under this profile that tolerance is the window being checked. Once u exceeds a window, every check against that window fails closed on authority, and safety functions are unaffected (Section 6 of [OT-COMMAND-AUTHORITY]).¶
An issuer sizes each expiry it sets, on a grant or a binding moment, to exceed the greatest transit time declared for the scenario plus the uncertainty of the enforcement point's clock. The maximum age of the freshness element is declared by the deployment for each risk class (Section 3) and is sized the same way. Where a CAE carries a smaller maximum age, the enforcement point evaluates against the smaller, and it never evaluates against a larger one than the deployment declared. A window smaller than the transit time arrives expired and is refused. Section 7 of [TIPTOP-UC] asks the same of certificate renewal, whose lead time should exceed the worst-case one-way delay plus the growth of clock uncertainty. Section 7.1 of [TIPTOP-ARCH] asks that HTTP time headers cover the longest delay. A longer window widens the time in which a captured request can be replayed (Section 11).¶
Section 6.1 of [SCSP] computes a deep-space acceptance window for frames as the ground transmit epoch plus the one-way light time, plus or minus separate terms for clock error, ephemeris uncertainty and margin, and requires that these "MUST NOT be combined into a single static tolerance". That window governs the acceptance of an SCSP frame. The checks of this section govern the CAE the frame carries and are made separately. Where both apply, both pass before dispatch. Section 5.1 of [TIPTOP-UC] adds the waiting time before a contact to the propagation delay, and the transit time a CAE's windows cover includes it.¶
Section 8 of [OT-COMMAND-AUTHORITY] requires a deployment to declare its revocation latency budget and its maximum trust-anchor staleness, and forbids letting either gate a safety function.¶
Under this profile's revocation posture, when the revocation status of a grant cannot be established within the bound that applies to it, an Observe action MAY proceed and an Adjust or State-change action MUST be refused. It attaches nothing else to the action classes. An Observe action proceeds because a grant is not a binding its class requires (Section 5 of [OT-COMMAND-AUTHORITY]). Its required check does not take the grant's revocation status as an input, and no unevaluable check is accepted.¶
The bound that applies is the revocation latency budget, with one exception. A deployment that admits contingency grants declares a second budget, the contingency budget, as the bound for revocation status relied on by an admission under a contingency grant (Section 9.2). Under every other grant, and under a contingency grant past its own budget, an Adjust or State-change action whose grant's revocation status cannot be established within the bound is refused. Section 8 of [OT-COMMAND-AUTHORITY] has a deployment declare one budget, and the contingency budget is declared beside it.¶
A deployment declares both values for each scenario it claims (Section 3), as durations. Where the links are scheduled contacts, a duration can be derived from the contact plan, for example from a maximum round-trip time. A count of contacts is not a duration and does not meet the requirement. Held data older than its staleness bound is treated as unavailable, and a check that needs it is unevaluable (R3 of [DISCONNECTED-TRUST]). A trust anchor past its bound therefore causes refusal of every action whose bindings depend on it, and the approver directory is such an anchor for the binding moment (Section 8 of [OT-COMMAND-AUTHORITY]).¶
In SC, a revocation issued at the origin during the moratorium does not reach the enforcement point until the moratorium ends. Where the declared budget is shorter than the conjunction, the revocation status of an Earth-issued grant cannot be established within the bound once the bound has passed, and Adjust and State-change actions under that grant are refused. A deployment that lengthens its budget to span a conjunction keeps honouring a revoked grant for that long (Section 11). Section 9 provides for actions a principal writes in advance for such an interval.¶
The means by which held data are replaced or withdrawn is the deployment's. It works over arbitrary delay and intermittence (R9 of [DISCONNECTED-TRUST]), and this profile defines none. The Key Revocation List of [SCSP] revokes ground controller keys and accepts a list only if its sequence number exceeds the stored one (Section 3.2 of [SCSP]). It does not express the revocation of a principal's grant, and this profile does not extend it to do so.¶
A revocation cannot recall an action already released to the process (Section 8 of [OT-COMMAND-AUTHORITY]). Long delay lengthens the interval between the release of an action and the arrival of a revocation, and leaves the rule unchanged.¶
Under long delay the approving human is at the origin, and the binding moment is signed before the command departs and evaluated when it arrives. The requirements of Section 3.4 of [OT-COMMAND-AUTHORITY] apply without change. The artefact carries the approver's own signature over a digest of the action. The enforcement point verifies it offline against cached key material and the approver directory, under B1 to B5. It carries an expiry, and it is not usable more than once. The artefact also names, by content address, the grant under which the approver acted. That requirement is this profile's addition, and Section 3.4 does not make it. Section 3.5 of [OT-COMMAND-AUTHORITY] has the admission record name the grant.¶
The action is fixed at the origin. The approver signs its asset, verb and parameters, and the enforcement point evaluates and forwards that same fixed action. Section 3.4 of [OT-COMMAND-AUTHORITY] forbids re-reading an action from a mutable structure after a wait, and time in a buffer is a wait of that kind.¶
The expiry is sized as Section 5 requires. An enforcement point refuses an artefact presented after its expiry and does not treat a captured approval as valid for any window the artefact does not itself state (Section 3.4 of [OT-COMMAND-AUTHORITY]).¶
The enforcement point records each artefact it accepts as spent, and the spend record survives a restart (Section 4 of [OT-COMMAND-AUTHORITY]). Spend state held only in volatile memory weakens single use to once per uptime. Section 6.8 of [TIPTOP-UC] notes that recovery after a reboot is slow in deep space and that endpoints may keep a state context in permanent memory.¶
Sections 3.3 and 3.4 of [OT-COMMAND-AUTHORITY] keep a per-action artefact at the binding moment from becoming a standing authorisation. Commands sent together, as Section 3 of [TIPTOP-UC] says related commands may be, each carry their own binding moment where their class requires one.¶
Where a deployment requires more than one approver and declares an approval deadline, the deadline is a time check like any other and is evaluated as Section 5 requires. A change to the approver directory, or to the number of approvers required, needs the same approvals as the actions it protects.¶
[SCSP] binds a hardware attestation nonce to a frame's header and payload (Section 5.2 of [SCSP]) and authorises ground keys by record. It has no per-action human approval. A CAE carried in an SCSP frame keeps its own binding moment.¶
Section 4 of [OT-COMMAND-AUTHORITY] applies as written. An action that was accepted and dispatched and whose effect the enforcement point cannot determine is unresolved, and is reported neither as performed nor as refused. An unresolved outcome carries no authority forward. A retry is a new control action with its own CAE and, where its class requires one, its own binding moment. Fresh authorisation alone does not permit another attempt at the same unresolved operation. The deployment keeps a stable operation identifier and a durable unresolved state across restart and re-authorisation, and a party holding the identifier can later ask for the recorded outcome.¶
The admission is durably recorded before the action is forwarded (Section 3.5 of [OT-COMMAND-AUTHORITY]). After a restart, an admission with no outcome record is unresolved, and the enforcement point appends an outcome record saying so. [SCSP] writes its replay watermark to non-volatile memory at execution, after dispatch, and holds the reservation before that in volatile memory (Section 4.4 of [SCSP]). A deployment that carries a CAE in an SCSP frame still writes the admission record before dispatch.¶
An outcome report sent from the enforcement point waits for the next contact. Until it arrives, the origin knows that the command was sent and nothing more. When contact resumes, the origin can ask for the recorded outcome by operation identifier. That query dispatches nothing and spends no artefact.¶
The same holds for the length of the moratorium. A binding moment dispatched against an outcome that turns out to be unresolved is spent, and an action sent after the conjunction to achieve the same effect needs a new one.¶
Signed telemetry from [SCSP] (Section 4.7) is one form an outcome report can take, and an SCSP error for an authorisation failure is a refusal. [SCSP] drops unauthenticated frames silently, and the absence of any report is therefore never evidence of refusal. Signed per-hop custody receipts, where they exist, report custody and do not establish what the process did.¶
Section 7 of [OT-COMMAND-AUTHORITY] requires an encoding of the CAE to be verifiable offline, to carry a freshness element (a nonce and a timestamp with a declared maximum age), and to ride above the transport security of the underlying session without weakening it. Every carriage below meets those requirements in the same way. The CAE travels with the command as application data, and the enforcement point evaluates it before the command reaches the process.¶
Section 7 of [OT-COMMAND-AUTHORITY] does not specify the concrete outer encoding of the CAE, that is, how the five bindings are serialised together into one structure, and names it as a decision its current revision leaves open. That decision belongs to that document. This profile takes whatever encoding it defines and adds no requirement to it. Every check in this profile is made on the bindings and the silence field after decoding, and none depends on how they are serialised.¶
Section 7.2 of [TIPTOP-ARCH] says that a mission operations centre often uses the CCSDS Space Packet Protocol to send commands, that much common software already supports it over UDP/IP, and that command applications are "already designed for space communications use cases". Neither protocol says anything about authority. Link security below them protects frames, and the CAE carries the authority. Where a deployment uses [SCSP], a CAE MAY travel as the command payload of an SCSP frame. SCSP's signature, authorisation record and replay window then apply to the frame, the checks of this profile apply to the CAE, and both pass before dispatch.¶
An SCSP command class is a mission-profile defined string in a key's Authorization Policy Record, naming what that key may address (Section 4.5 of [SCSP]). It states nothing about physical consequence. The risk class of each action carried in an SCSP frame is assigned by the deployment from its process hazard analysis, as Section 5 of [OT-COMMAND-AUTHORITY] requires of every action. An authorisation record that admits a command class at the frame layer neither sets nor lowers that risk class, and an enforcement point does not downgrade a class to avoid a binding requirement (Section 5 of [OT-COMMAND-AUTHORITY]).¶
Section 7.1.1 of [TIPTOP-ARCH] names CoAP [RFC7252] as a candidate application protocol for space. OSCORE [RFC8613] "provides end-to-end protection between endpoints communicating using CoAP or CoAP-mappable HTTP" and supports a range of proxy operations, and the protection holds across relays. Over CoAP, this profile has the enforcement point obtain the freshness of a request through the Echo option of [RFC9175], because a CoAP server "is, in general, not able to verify when the request was sent", even under a security protocol (Section 2.1 of [RFC9175]). An Echo exchange costs a round trip, at least minRTT and up to maxRTT. Where that exceeds the maximum age of the action, the CAE's own timestamp, checked as Section 5 requires, is the freshness element. The timestamp lies under the signature of the party that assembled the CAE at the origin. Because a party that delays the request cannot sign it again, the timestamp bounds that delay. Where the enforcement point can establish freshness by neither, it acts on no authority and the action is refused (Section 5).¶
HTTP/3 [RFC9114] maps HTTP semantics over QUIC. Section 6.3 of [TIPTOP-ARCH] says that QUIC stacks in their default configurations will not work in deep space, and Section 7.1 says that HTTP time headers such as Expires must be large enough to cover the longest delay. Those headers govern caching. The CAE's expiries are its own and are evaluated as Section 5 requires. QUIC authenticates the connection, and the proof of authority is the CAE in the request.¶
The considerations of Section 8 of [OT-COMMAND-AUTHORITY] and Section 9 of [DISCONNECTED-TRUST] apply. Long delay adds the considerations below.¶
Section 2.1 of [RFC9175] notes that delayed requests are "particularly perilous in the case of actuators". An attacker who can hold a CAE and release it later can use the whole window the issuer allowed for transit. Section 10 of [TIPTOP-ARCH] notes the opposite case, that an attacker on a path with long delays and several routes might inject packets that arrive sooner than legitimate ones. An older revocation status replayed ahead of a newer one is such a packet, and Section 9.1 never lets it lower T. The window is bounded by the CAE's expiry and maximum age, evaluated against the enforcement point's declared uncertainty (Section 5), and a binding moment is spent on first use (Section 7).¶
Intermediate nodes buffer data for a long time ([TIPTOP-ARCH] Section 10). A copy of a CAE held in a buffer can be presented again. A binding moment presented twice is refused on its spend record. An Observe or Adjust action without a binding moment is bounded by its nonce and maximum age, and nonce protection lasts only as long as the enforcement point remembers the nonces it has accepted. State lost at a restart reopens the window, as Section 4 of [OT-COMMAND-AUTHORITY] observes of spend state. An enforcement point MUST keep its record of accepted nonces across a restart for at least the declared maximum age of the freshness element, or MUST refuse every CAE whose timestamp is earlier than its last restart. Section 5 of [RFC9175] requires the same of Echo values created before a timer reset. The per-key replay window of [SCSP] (Section 4.4) is a separate bound at the frame layer.¶
Section 10 of [TIPTOP-ARCH] says that data at rest in a buffer becomes available to an attacker who compromises the node. A buffered CAE reveals the agent, the principal reference, the grant reference, the asset and the verb. Its signatures stop the attacker forging a new one, and a captured copy can still be replayed within its windows. Confidentiality in transit and at rest comes from the carriage, for example OSCORE, BPSec, or SCSP's AEAD payload encryption.¶
A declared uncertainty bounds only the error the deployment knows about. A clock displaced without the deployment's knowledge lies outside it. An on-path delay in a time exchange can shift a clock by half the delay with no sign in the protocol, and Section 6.2 of [TIPTOP-UC] notes that two-way offset estimation is biased by asymmetry between the forward and return paths. An enforcement point whose clock is set back accepts expired artefacts (Section 13). One whose clock is set forward fails its expiries and maximum ages sooner. It cannot make silence hold, because Section 9.1 also requires the time elapsed on a monotonic counter to pass the budget, and a clock set back after the enforcement point has recorded a t', with no restart of its counter since, cannot stretch the contingency budget (Section 9.3).¶
An attacker who keeps an enforcement point out of contact past its revocation budget causes Adjust and State-change actions under its grants to be refused. Section 8 of [OT-COMMAND-AUTHORITY] records this cost of refusing where authority cannot be verified. The same attacker wakes every contingency grant whose capping grant falls silent within the interval the contingency grant states, and gains nothing outside it (Section 9.2). Where the agent a contingency grant names sits on the path by which revocation status arrives, it could produce that silence itself and hold a revocation off. Section 9.1 refuses every admission under such a grant. A principal therefore writes into a contingency grant only actions it accepts being taken whenever its capping grant falls silent within the interval, whatever the cause. Safety functions are unaffected.¶
A budget or staleness bound lengthened to span a conjunction widens the time in which a revoked grant or a replaced anchor is still honoured. A deployment that does so states it in its declaration. A contingency budget is such a bound for revocation status relied on by an admission under a contingency grant. A revocation of either grant that the enforcement point has not received is honoured until that budget passes or the contingency grant's interval ends.¶
A safety function MUST NOT be gated on any binding in this profile, and a safety action that a system would take autonomously MUST remain takeable when every authority is unreachable (Section 6 of [OT-COMMAND-AUTHORITY]).¶
A buffered CAE names an agent and the principal on whose authority it acts, and anyone with access to the buffer can read an unencrypted one. The audit record holds who commanded what, and it is kept for review after contact resumes. The privacy considerations of [DISCONNECTED-TRUST] apply. A deployment carries the least identifying evidence the checks need and bounds the retention and the readers of its records.¶
Every expiry and maximum age of Sections 5 and 7 rests on the enforcement point's own clock. A clock set back without the deployment's knowledge lies outside the declared uncertainty and accepts expired artefacts (Section 11). Whether a profile can require more against that displacement without new mechanism is open. Silence and admission under a contingency grant also rest on a monotonic counter (Section 9), and this issue reaches them only through a displacement before any t' is recorded, or one that stays above the latest recorded t' after the counter has restarted (Section 9.1).¶
This document has no IANA actions.¶
draft-morrison-command-authority-long-delay-00 (2026-11-18): first version. Profiles the Command Authority Envelope for the deep-space connectivity scenarios. Covers applicability by scenario, the five bindings, time checks under declared clock uncertainty, revocation and anchor staleness by scenario, binding moments signed before departure, unresolved outcomes across gaps and reboot, contingency grants dormant until silence and what an enforcement point may do once every bound has passed (answering question 3 of the disconnected trust model), and carriage over SPP/UDP, CoAP with OSCORE and HTTP/3, with SCSP command classes kept apart from risk classes and the outer encoding left to the profile this one builds on.¶
None yet.¶