<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE rfc [ <!ENTITY nbsp "&#160;"> ]>
<rfc xmlns:xi="http://www.w3.org/2001/XInclude" category="info" docName="draft-das-consequence-bound-execution-authority-00" ipr="trust200902" submissionType="independent" xml:lang="en" version="3" tocInclude="true" tocDepth="3" sortRefs="true" symRefs="true">
<front>
<title abbrev="Consequence-Bound Execution Authority">Consequence-Bound Execution Authority for Autonomous Systems: Operation Reconstruction, Independent Validation, and Protected Effectuation Profiles</title>
<seriesInfo name="Internet-Draft" value="draft-das-consequence-bound-execution-authority-00"/>
<author fullname="Sangam Kumar Das" initials="S." surname="Das">
<organization>Independent Inventor</organization>
<address>
<postal><city>Balasore</city><region>Odisha</region><code>756001</code><country>India</country></postal>
<phone>+91 98613 63532</phone>
<email>info@sangamdas.com</email>
<!-- TODO(author): add the LinkedIn profile URL as <uri>https://www.linkedin.com/in/...</uri>; the URL was not available to the editor -->
</address>
</author>
<date year="2026" month="November" day="16"/>
<area>Security</area>
<workgroup>Independent Submission</workgroup>
<keyword>execution finality</keyword><keyword>AI agents</keyword><keyword>authorization</keyword><keyword>effectuation boundary</keyword><keyword>operation reconstruction</keyword><keyword>Finality Sink</keyword>
<abstract>
<t>Authorization to compute, propose, prepare, simulate or request an operation is not, by itself, authorization to cause the operation's actual consequential effect. An authorized operation may afterwards be transformed, delegated, delayed, replicated, compiled, offloaded or executed under different technical conditions. This document describes an architecture, and a set of candidate interoperability requirements, for separating those authorizations: a proposed operation is held as a Candidate Act in a Non-Effective State; the Actual Pending Operation is reconstructed; current source authority, context, dependencies and policy and revocation epochs are evaluated by an independent validator; a scoped, non-bearer execution authorization is issued; and a protected Finality Sink at the actual effectuation boundary re-derives the operation, enforces the authorization, and supports reconciliation against authoritative outcome evidence.</t>
<t>The document is an individual, informational technical proposal. It credits OAuth, JWT/CWT, RBAC/ABAC, zero-trust, remote attestation, transaction logs, hardware isolation and cryptographic signatures for their established functions and does not characterize them as defective. It includes a threat model, 12 figures, 39 consequence-specific profiles, a non-normative European applicability discussion and an examiner-oriented terminology appendix. It is not an IETF standard, an EU regulatory instrument, a telecommunications or financial standard, or an endorsement by any named company.</t>
</abstract>
<note title="Source and Patent Information">
<t>The technical subject matter is attributed to the human inventor Sangam Kumar Das. Foundational reference: PCT/IB2026/055615, filed 4 June 2026, published 16 July 2026 as WO 2026/150382 A2. Later Indian provisional application 202631136555, dated 15 November 2026 (as reported by the applicant). See <xref target="ipr"/> for the full IPR and attribution statement.</t>
<t>Where to read the full PCT filing: search "Sangam Das" on WIPO PATENTSCOPE; search PCT/IB2026/055615 (pctib202655615); or read the PCT file directly in Technical Disclosure Commons at https://www.tdcommons.org/dpubs_series/12542/.</t>
<t>AI-assisted drafting was used to prepare this explanatory text, keyword index and industry mapping. Readers must verify patent numbers, dates, claim quotations, product capabilities and standards references against primary sources.</t>
</note>
</front>
<middle>
<section anchor="intro">
<name>Introduction</name>
<section anchor="problem">
<name>Problem Statement</name>
<t>Systems increasingly separate the component that decides what to do from the component that makes it happen. An AI agent proposes a tool call; a workflow engine queues it; a database trigger fans it out; a SmartNIC compiles it into a hardware rule; a retry worker repeats it after a timeout. Each hop can be individually authenticated and individually permitted, yet the operation that finally produces an effect may differ from the operation that was approved, may be executed under authority that has since been revoked, or may have an effect that is only visible after composition with other work.</t>
<t>This document uses the term *consequence* for the externally effective result of an operation at the point where it can no longer be undone by the originating software alone. The architecture's organizing question is: at the boundary where a consequence would become real, what independent evidence shows that this exact effect, now, under current authority, is permitted?</t>
</section>
<section anchor="distinction">
<name>Central Distinction</name>
<t>The central principle preserved throughout this document is:</t>
<t>Authorization to compute, propose, prepare, simulate, or request an operation is not automatically authorization to cause its actual consequential effect.</t>
<t>The following are distinct questions with distinct evidence. A deployment can answer some and not others, and conflating them is the failure mode this document addresses.</t>
<table anchor="tbl-distinct">
<name>Five questions that are not interchangeable</name>
<thead><tr><th>Question</th><th>Typical established mechanism</th><th>What it does not establish</th></tr></thead>
<tbody>
<tr><td>Who is the caller? (identity authentication)</td><td>TLS client auth, OAuth client auth, workload identity</td><td>That this operation is permitted</td></tr>
<tr><td>May the caller invoke this tool? (tool-invocation permission)</td><td>OAuth scopes, RBAC, ABAC, API gateways</td><td>That the operation actually reaching the target is the one evaluated</td></tr>
<tr><td>May the caller stage this operation? (staging permission)</td><td>Draft/queue/propose APIs, change requests</td><td>That staged work may later become effective</td></tr>
<tr><td>May this exact effect be committed now? (effect-commit permission)</td><td>Proposed here: act-bound capability checked at a Finality Sink</td><td>That the intended outcome was achieved</td></tr>
<tr><td>Did the intended effect occur? (outcome evidence)</td><td>Destination-of-record confirmation, settlement evidence</td><td>That a later continuation is authorized</td></tr>
</tbody>
</table>
</section>
<section anchor="relation">
<name>Relationship to Existing DAS Internet-Drafts and Contribution of This Document</name>
<t>The same author has published several individual Internet-Drafts on execution finality. This document was compared with the following, using the text available to the editor at the time of drafting:</t>
<ul>
<li>draft-das-receipt-gated-iev <xref target="I-D.das-receipt-gated-iev"/> (individual, Experimental, October 2026): staged effectuation, Bounded Trial Effect, receipt-gated continuation and the Interim Effectuation Validator. The editor read this draft's text. This document reuses its IEV idea only as an optional phase profile (<xref target="fig-f4"/>) and does not redefine it.</li>
<li>draft-das-execution-finality-foundations <xref target="I-D.das-execution-finality-foundations"/> and draft-das-agent-execution-finality <xref target="I-D.das-agent-execution-finality"/>: the editor could not retrieve the full text of either from the Datatracker during preparation. This document therefore makes no claim about their precise content. Overlap in the core chain (Candidate Act, Finality Sink, non-bearer authorization) is expected and should be resolved by the author before submission.</li>
<li>Other Datatracker drafts by the same author exist in the same family, including draft-das-interoperable-consequence-authority <xref target="I-D.das-interoperable-consequence-authority"/> (Part 1 of a series), draft-das-finality-sink-enforcement <xref target="I-D.das-finality-sink-enforcement"/> (Part 4), draft-das-causal-effect-receipts <xref target="I-D.das-causal-effect-receipts"/> (Part 6), draft-das-execution-finality-protocol-layer, and draft-das-ef-registries <xref target="I-D.das-ef-registries"/> (which proposes registries for the family). The editor read only search excerpts of these.</li>
</ul>
<t>The intended additional contribution of this document is limited to the following, and each should be checked for duplication:</t>
<ol>
<li>A single taxonomy that separates identity, tool-invocation, staging, effect-commit and outcome evidence (<xref target="tbl-distinct"/>).</li>
<li>A catalogue of 39 consequence-specific profiles, each stated as an enforcement predicate of the form 'X does not independently authorize Y', with its effectuation boundary and failure modes (<xref target="appendix-emb"/>). These profiles derive from the later Indian provisional filing.</li>
<li>Seventeen candidate interoperability requirements with an enforcement rationale and a testable failure condition for each (<xref target="reqs"/>).</li>
<li>An 18-item threat analysis with attacker capability, trust boundary, protected invariant, mitigation, residual risk and a concrete test (<xref target="threats"/>).</li>
<li>Explicit treatment of route coverage, provider-hosted tool execution, uncertain outcomes and latency-aware placement.</li>
</ol>
<t>If, on comparison, the core chain material of this document substantially duplicates an existing draft, the recommended course is to revise that draft or to publish only the profile catalogue and threat analysis as a narrowly scoped companion document, rather than submitting a redundant document.</t>
</section>
<section anchor="nonendorse">
<name>Status and Non-Endorsement</name>
<t>This is an individual technical proposal. It is not an endorsed IETF standard, not an EU regulatory approval, not a telecommunications standard, not a financial certification and not an implementation by any named company. References to companies and products identify potential technical integration points only. They are not evidence of adoption, sponsorship, copying, infringement or commercial affiliation. Regulatory references (<xref target="eu-applicability"/>) are engineering relevance statements and are not claims of legal compliance.</t>
</section>
</section>
<section anchor="terminology">
<name>Terminology</name>
<section anchor="reqlang">
<name>Requirements Language</name>
<t>The key words "<bcp14>MUST</bcp14>", "<bcp14>MUST NOT</bcp14>", "<bcp14>REQUIRED</bcp14>", "<bcp14>SHOULD</bcp14>", "<bcp14>SHOULD NOT</bcp14>", "<bcp14>RECOMMENDED</bcp14>", "<bcp14>NOT RECOMMENDED</bcp14>", "<bcp14>MAY</bcp14>" and "<bcp14>OPTIONAL</bcp14>" in this document are to be interpreted as described in BCP 14 <xref target="RFC2119"/> <xref target="RFC8174"/> when, and only when, they appear in all capitals, as shown here.</t>
<t>Scope of normative language: because this document is Informational, normative keywords bind only an implementation that elects to claim the property to which the statement relates. Requirements are collected in <xref target="reqs"/>; keywords used elsewhere restate those requirements, state architectural recommendations, or restate IETF process obligations. Such language does not oblige any deployment to claim a property and does not define a wire format, registry or transport. Statements marked [ARCH] are architectural recommendations.</t>
</section>
<section anchor="defs">
<name>Definitions</name>
<t>Definitions are listed with industry-standard synonyms where they exist. This document avoids new acronyms where a plain term suffices; "IEV" is retained only because it is used by the cited companion draft.</t>
<dl newline="true">
<dt>Candidate Act</dt>
<dd><t>A proposed consequential operation produced by a human, software or AI agent. Proposal confers no effectuation authority. Synonyms: proposed operation, deferred operation, staged transaction.</t></dd>
<dt>Non-Effective State</dt>
<dd><t>A protected state in which a Candidate Act exists without having produced its governed consequence. Synonym: pending non-committed state.</t></dd>
<dt>Actual Pending Operation</dt>
<dd><t>The operation that would really occur at the effectuation boundary, derived (reconstructed) from the Candidate Act, current resource state and applicable execution parameters, rather than taken from the proposer's description. Synonym: reconstructed effective operation.</t></dd>
<dt>Operation Reconstruction</dt>
<dd><t>The process of deriving the Actual Pending Operation, including the effective principal, target, parameters, material dependencies and derived child effects.</t></dd>
<dt>Source-Authority Envelope</dt>
<dd><t>The record of the originating authority for an act: principal, delegation chain, permitted task and operation class, limits, purpose and applicable epochs. Synonym: originating authorization context.</t></dd>
<dt>Protected Enforcement Domain</dt>
<dd><t>An environment whose integrity and policy state are protected against the operation source, in which validation occurs and protected state is kept. Realizations include a trusted execution environment, a hardware security module, a separate process or host, or a trusted application boundary.</t></dd>
<dt>Independent Validator</dt>
<dd><t>A component, outside the control of the operation source, that evaluates the Actual Pending Operation against current authority, policy, dependencies and evidence. Synonym: runtime policy revalidation service.</t></dd>
<dt>Scoped Execution Capability</dt>
<dd><t>A non-bearer authorization bound to a specific act, scope, destination, validity period or consumption count, designated Finality Sink and current epochs. Possession alone is not sufficient to cause the effect. Synonym: operation-bound execution authorization.</t></dd>
<dt>Finality Gate</dt>
<dd><t>The decision logic at or before the effectuation boundary that admits or rejects an operation.</t></dd>
<dt>Finality Sink</dt>
<dd><t>The component at the first boundary at which the operation can become externally effective, which verifies a Scoped Execution Capability against the Actual Pending Operation and consumes it. Synonyms: protected effectuation boundary, execution enforcement point.</t></dd>
<dt>Consequence-Bound Authorization</dt>
<dd><t>Authorization whose predicate is evaluated over the effect that would actually result, including aggregate and transitive effects, rather than over the request alone. Synonym: effect-aware authorization.</t></dd>
<dt>Authority Continuity</dt>
<dd><t>The property that originating restrictions and delegated limits remain binding through transformation, summarization, delegation, queueing, translation and retry. Synonym: delegated authority preservation.</t></dd>
<dt>Protected Policy Epoch</dt>
<dd><t>A protected, monotonically advancing version of the effect-relevant policy. Synonym: policy generation counter.</t></dd>
<dt>Revocation Epoch</dt>
<dd><t>A protected, monotonically advancing version of revocation state, used to reject authority issued before a revocation.</t></dd>
<dt>Evidence-Carrying Receipt</dt>
<dd><t>A record, bound to an act and phase, carrying authenticated evidence of what occurred. Self-reported receipts from the act source are not Authoritative Outcome Evidence.</t></dd>
<dt>Effect Confirmation Receipt</dt>
<dd><t>A receipt whose evidence originates at the destination of record or an independent observer rather than at the operation source. Synonym: authenticated destination effect evidence.</t></dd>
<dt>Interim Effectuation Validation (IEV)</dt>
<dd><t>A validation in which authenticated evidence of an earlier real (bounded) effect is evaluated before a continuation condition for a later effect is established. Synonym: receipt-gated continuation validation.</t></dd>
<dt>Ledger-Anchored Validation Receipt</dt>
<dd><t>A validation receipt whose existence and ordering are anchored in an append-only or transaction-log structure, so that omission or reordering is detectable. Anchoring is optional and does not by itself establish validity.</t></dd>
<dt>Effective Operation Descriptor</dt>
<dd><t>A structured description of the effective operation as reconstructed, including, where relevant, installed match/action behavior of a hardware path (Effective Hardware Action Descriptor).</t></dd>
<dt>Material Dependency Graph</dt>
<dd><t>The graph of resources, services, derived operations and state on which the effect depends or which it affects materially. Synonym: causal effect graph.</t></dd>
<dt>Protected State Commitment</dt>
<dd><t>A cryptographic or hardware-protected commitment to state used in validation (for example an epoch, ledger or watermark).</t></dd>
<dt>Authoritative Outcome Evidence</dt>
<dd><t>Evidence from the system of record, or from an independent observer with adequate coverage, sufficient to classify an operation as committed, conclusively not committed, or unresolved.</t></dd>
<dt>Stale Authority</dt>
<dd><t>Authority that was valid at issuance but whose epoch, revocation state, target, context or time bound no longer holds at effectuation.</t></dd>
<dt>Transitive Effect</dt>
<dd><t>An effect reachable from the authorized operation through a chain of triggers, workflows, queues, delegations or calls, not named by the entry point.</t></dd>
<dt>Aggregate Consequence</dt>
<dd><t>The combined effect of several operations or work products that are each individually acceptable.</t></dd>
<dt>Bounded Real Effect</dt>
<dd><t>A real but intentionally limited effect, not a simulation, preview or dry run, performed before broader effectuation. Synonyms: canary effect, limited trial effect.</t></dd>
<dt>Effectuation-Boundary Enforcement</dt>
<dd><t>Enforcement at the point where the operation actually becomes effective, as opposed to enforcement only at request, gateway or logging layers.</t></dd>
<dt>Control-to-Effect Route Coverage</dt>
<dd><t>The extent to which every discovered route capable of producing a governed effect is subject to the required control, demonstrated by evidence of rejection at the sink rather than by policy configuration or logs.</t></dd>
</dl>
</section>
<section anchor="established">
<name>Established Mechanisms and Their Functions</name>
<t>The following mechanisms are not criticized here. Each answers a real question, and the architecture below relies on them.</t>
<dl newline="true">
<dt>OAuth 2.0 and token exchange</dt>
<dd><t>Delegated authorization of a client to call resource servers (<xref target="RFC6749"/>, <xref target="RFC8693"/>); proof-of-possession and sender-constraining (<xref target="RFC9449"/>, <xref target="RFC8705"/>, <xref target="RFC7800"/>); rich authorization details (<xref target="RFC9396"/>). These can carry fine-grained scopes and bind tokens to keys. They do not, by themselves, make the resource server re-derive the effective operation or guarantee that every route to the effect is covered.</t></dd>
<dt>JWT, CWT and COSE</dt>
<dd><t>Compact signed or MACed claim formats (<xref target="RFC7519"/>, <xref target="RFC8392"/>, <xref target="RFC9052"/>). They are suitable carriers for a Scoped Execution Capability; this document does not select a format.</t></dd>
<dt>RBAC and ABAC</dt>
<dd><t>Role-based and attribute-based access control (<xref target="NIST-ABAC"/>). They decide whether a principal may perform an operation class. The profiles here concern whether the particular effect is within scope at effectuation.</t></dd>
<dt>Zero trust</dt>
<dd><t>Continuous verification and least privilege (<xref target="NIST-ZTA"/>). The architecture is consistent with it and adds a requirement to verify at the actual effect boundary.</t></dd>
<dt>Remote attestation</dt>
<dd><t>Evidence about an attester's state, appraised by a verifier for a relying party (<xref target="RFC9334"/>, <xref target="RFC9711"/>). The attestation profile (E9) concerns correspondence between remediation, evidence and policy per consequence class and does not replace RATS roles.</t></dd>
<dt>Transaction logs and audit</dt>
<dd><t>Durable records of what was done. Logging is detection and accountability; it is not prevention unless the log write gates the effect.</t></dd>
<dt>Hardware isolation and secure elements</dt>
<dd><t>Isolation of keys and execution. They bound what compromise can do under intact-hardware assumptions; E35 considers the case where those assumptions fail.</t></dd>
<dt>Cryptographic signatures</dt>
<dd><t>Integrity and origin of a message. A valid signature over a request does not show that the request's effect is permitted now.</t></dd>
</dl>
</section>
</section>
<section anchor="scope">
<name>Scope and Non-Goals</name>
<t>In scope: an architecture for controlling the transition from a proposed operation to its actual consequence; candidate interoperability requirements for implementations that claim specific properties; threat analysis; and a catalogue of consequence-specific profiles.</t>
<t>Non-goals:</t>
<ul>
<li>This document defines no wire format, message encoding, registry or transport. Where an encoding is needed, existing formats (for example COSE or JWT) are candidates.</li>
<li>It does not claim that the architecture prevents every attack, nor that it can control a system whose effectuation path lies outside the protected boundary and cannot be integrated.</li>
<li>It does not replace identity, authorization, attestation, safety or regulatory controls; it is intended to be layered with them.</li>
<li>It does not address real-time safety loops, PHY/MAC scheduling or other microsecond-scale control paths, except to state how they relate to the slower boundaries that this document does address (see <xref target="latency"/> and <xref target="eu-applicability"/>).</li>
<li>It makes no experimental, performance, latency or adoption claim. None is reported.</li>
</ul>
</section>
<section anchor="architecture">
<name>Architecture</name>
<t><xref target="fig-f1"/> shows the principal flow. The ten elements that the architecture relates are: (1) Originating Authority, (2) Candidate Act, (3) Non-Effective State, (4) Actual Pending Operation reconstruction, (5) current execution context and dependencies, (6) independent validation, (7) scoped execution authorization, (8) Protected Finality Sink, (9) actual resulting effect, and (10) authoritative outcome reconciliation.</t>
<figure anchor="fig-f1">
<name>Principal protected execution-finality flow (mandatory conceptual flow)</name>
<artwork type="ascii-art"><![CDATA[
+------------------------------+
| AI Agent / Software / Human  |
|       Operation Source       |
+--------------+---------------+
               |  (1) propose
               v
+------------------------------+
|        Candidate Act         |
|  Proposed consequential act  |
+--------------+---------------+
               |  (2) stage; no real effect yet
               v
+------------------------------+
|   Protected Non-Effective    |
|            State             |
+--------------+---------------+
               |  (3) derive what would really happen
               v
+------------------------------+
|   Actual Pending Operation   |
|        Reconstruction        |
+--------------+---------------+
               |  (4) bind current facts
               v
+------------------------------+
| Source Authority + Context   |
| Dependencies + Policy Epoch  |
+--------------+---------------+
               |  (5) evaluate independently
               v
+------------------------------+
|   Independent Consequence-   |
|      Bound Validation        |
+--------------+---------------+
               |
               v
        +-------------+
        | Valid now?  |
        +------+------+
          NO   |   YES
       +-------+-------+
       v               v
+--------------+  +------------------+
| HOLD / DENY  |  | Scoped Execution |
| / RECONCILE  |  |  Authorization   |
+--------------+  +--------+---------+
                           |  (6) present at the boundary
                           v
                  +-------------------+
                  | Protected Finality|
                  |       Sink        |
                  +--------+----------+
                           |  (7) sink re-derives and consumes
                           v
                  +-------------------+
                  | Actual Consequence|
                  +--------+----------+
                           |  (8) observe at the destination
                           v
                  +-------------------+
                  | Outcome Evidence  |
                  | & Reconciliation  |
                  +-------------------+]]></artwork>
</figure>
<section anchor="transitions">
<name>Transitions, Conditions, Trust Assumptions and Failures</name>
<t>Each numbered transition in <xref target="fig-f1"/> is described with its mandatory condition, the trust assumption it relies on and its principal failure. Unless stated, a failed condition results in HOLD or DENY (fail closed).</t>
<dl newline="true">
<dt>(1) Source to Candidate Act: propose</dt>
<dd><t>Condition: the proposal is recorded with a stable act identity and the source's authority envelope. Trust: none placed in the source's own description of the operation. Failure: source supplies an inaccurate description, or an unbounded number of proposals (rate and quota limits apply).</t></dd>
<dt>(2) Candidate Act to Non-Effective State: stage</dt>
<dd><t>Condition: staging produces no governed consequence; staging area is itself protected from tampering by the source after submission. Trust: staging store integrity. Failure: staged act modified after validation; staged act becomes effective through an unmodelled path (route coverage failure).</t></dd>
<dt>(3) Non-Effective State to Actual Pending Operation: reconstruct</dt>
<dd><t>Condition: the effective principal, target, parameters, dependencies and derived child effects are derived from current resource state, not copied from the proposal. Trust: the reconstruction logic and its inputs are accurate. Failure: incomplete reconstruction (an effect branch is not modelled), reconstruction on stale state, or state changing between reconstruction and effect (time-of-check to time-of-use).</t></dd>
<dt>(4) Reconstruction to context binding</dt>
<dd><t>Condition: the current Source-Authority Envelope, material dependencies, and policy and revocation epochs are bound to the operation. Trust: epoch sources are authentic and fresh. Failure: epoch source unreachable or stale (treated as not current).</t></dd>
<dt>(5) Context to Independent Validation</dt>
<dd><t>Condition: the validator is independent of the source and evaluates the reconstructed operation, not the proposal. Trust: validator integrity. Failure: validator compromise or substitution (<xref target="threats"/>, T7); mitigated by attestation of the validator and by the sink's own re-derivation.</t></dd>
<dt>Decision: valid now?</dt>
<dd><t>Condition: every required predicate for the consequence class is true. UNKNOWN is not true. Failure: a predicate cannot be evaluated (HOLD); a predicate is mis-specified (policy error; residual risk).</t></dd>
<dt>NO branch: HOLD, DENY or RECONCILE</dt>
<dd><t>Condition: no capability is issued. RECONCILE applies where an earlier unresolved outcome must be classified first. Failure: denial of service by inducing HOLD (<xref target="threats"/>, T17); mitigated by bounded holds and safe-state rules.</t></dd>
<dt>(6) YES branch: scoped authorization</dt>
<dd><t>Condition: a Scoped Execution Capability is created, bound to the act, scope, destination, epochs, validity period or consumption count, and the designated sink. Trust: capability is non-bearer, so possession alone does not authorize. Failure: capability replay or widening (T5).</t></dd>
<dt>(7) Sink: re-derive and consume</dt>
<dd><t>Condition: the sink independently reconstructs the operation it is about to make effective, compares it with the capability, checks sink-local epoch and consumption state, then consumes the capability atomically with the effect where technically required. Trust: the sink controls the first effectual boundary. Failure: alternate route around the sink (T11), crash between consumption and effect (REQ-13).</t></dd>
<dt>(8) Effect to Outcome Evidence and Reconciliation</dt>
<dd><t>Condition: the outcome is classified from Authoritative Outcome Evidence as committed, conclusively not committed, or unresolved. Trust: evidence originates at the destination or an independent observer. Failure: forged, missing, delayed or contradictory evidence (T8, T9); unresolved outcomes remain quarantined (REQ-12).</t></dd>
</dl>
</section>
<section anchor="src-vs-eff">
<name>Source Authority Versus Reconstructed Effective Operation</name>
<t><xref target="fig-f2"/> illustrates why the validator compares the authorized envelope with the reconstructed operation. The nominal target, derived children, effective principal, and epochs may differ between approval and effectuation.</t>
<figure anchor="fig-f2">
<name>Source authority versus reconstructed effective operation</name>
<artwork type="ascii-art"><![CDATA[
 WHAT WAS AUTHORIZED            WHAT WILL ACTUALLY HAPPEN
 (Source-Authority Envelope)    (Actual Pending Operation)
+---------------------------+  +---------------------------+
| principal / delegator     |  | effective principal       |
| permitted task class      |  | effective operation       |
| nominal target "T"        |  | resolved target "T'"      |
| scope, limits, purpose    |  | derived children, writes  |
| policy epoch  Pe=17       |  | current epoch  Pe=19      |
| revocation epoch Re=4     |  | current Re=5              |
+-------------+-------------+  +-------------+-------------+
              |                              |
              +--------------+---------------+
                             v
              +------------------------------+
              |  Consequence-bound compare   |
              |  T == T'?  children inside   |
              |  scope?  epochs current?     |
              |  limits composed correctly?  |
              +--------------+---------------+
                             |
          any mismatch       |        all hold
       +---------------------+---------------------+
       v                                           v
 +-------------+                         +------------------+
 | HOLD / DENY |                         | scope-limited    |
 | fail closed |                         | authorization    |
 +-------------+                         +------------------+]]></artwork>
</figure>
</section>
<section anchor="staging-vs-commit">
<name>Distinguishing Staging, Commitment and Evidence</name>
<t>Permission to stage an operation (to write it into a queue, change request or draft) <bcp14>MUST</bcp14> be treated, in any implementation claiming effect-commit control, as separate from permission to commit its effect. Permission to commit the effect is in turn separate from evidence that the intended effect was achieved: a continuation or customer-facing statement that depends on the outcome requires the outcome evidence (<xref target="reqs"/>, REQ-07 and REQ-12).</t>
</section>
</section>
<section anchor="components">
<name>Component Roles and Trust Assumptions</name>
<t><xref target="fig-f3"/> shows trust boundaries B1 (protected enforcement domain), B2 (Finality Sink) and B3 (effect domain).</t>
<figure anchor="fig-f3">
<name>Independent Validator, protected state, and Finality Sink trust boundaries</name>
<artwork type="ascii-art"><![CDATA[
+-----------------------------------------------------------------+
| UNTRUSTED / PARTIALLY TRUSTED ZONE                              |
|  model, agent runtime, orchestrator, tool adapters, retry       |
|  workers, caches, application-side gateways                     |
+--------------------------------+--------------------------------+
                                 | Candidate Act (no authority)
                                 v
+-----------------------------------------------------------------+
| PROTECTED ENFORCEMENT DOMAIN  (trust boundary B1)               |
|  +-------------------+     +--------------------------------+   |
|  | Independent       |<--->| Protected state                |   |
|  | Validator         |     |  policy epoch, revocation      |   |
|  | (re-derives facts)|     |  epoch, consumption record,    |   |
|  +---------+---------+     |  exposure ledger, commitments  |   |
|            |               +--------------------------------+   |
|            | Scoped Execution Capability (non-bearer)           |
+------------+----------------------------------------------------+
             v
+-----------------------------------------------------------------+
| FINALITY SINK  (trust boundary B2)                              |
|  verifies capability vs ACTUAL operation, checks sink-local     |
|  epoch/consumption, then admits or rejects                      |
+--------------------------------+--------------------------------+
                                 | only admitted operations
                                 v
+-----------------------------------------------------------------+
| EFFECT DOMAIN: database, network element, payment rail, device, |
| actuator, accelerator, message channel (trust boundary B3)      |
+-----------------------------------------------------------------+
 Assumption: B2 is the first boundary where the act can become
 externally effective. If another route reaches B3 around B2, the
 route-coverage requirements (REQ-09, REQ-16) are violated.]]></artwork>
</figure>
<section anchor="roles">
<name>Component Roles</name>
<dl newline="true">
<dt>Operation Source</dt>
<dd><t>AI agent, software component or human. Untrusted for the purpose of describing its own operation. May hold identity credentials and tool-invocation permissions.</t></dd>
<dt>Staging Store</dt>
<dd><t>Holds Candidate Acts in the Non-Effective State. Integrity-protected; the source can append but not mark an act validated.</t></dd>
<dt>Reconstruction Function</dt>
<dd><t>Derives the Actual Pending Operation and Material Dependency Graph from current state. May be a part of the validator or of the sink, or both; the sink <bcp14>MUST</bcp14> be able to re-derive for the operation it admits where the sink claims to enforce REQ-01 to REQ-04.</t></dd>
<dt>Independent Validator</dt>
<dd><t>Evaluates predicates and issues the Scoped Execution Capability. Runs in a Protected Enforcement Domain. May be co-located with the sink for latency (see <xref target="latency"/>).</t></dd>
<dt>Protected State</dt>
<dd><t>Epochs, consumption records, exposure ledgers, watermarks, commitments. Crash-consistent where replay protection is claimed.</t></dd>
<dt>Finality Sink</dt>
<dd><t>The enforcement point at the first effectual boundary. Holds the credential or capability to cause the effect; the operation source does not.</t></dd>
<dt>Effect Observer / Destination of Record</dt>
<dd><t>Produces Authoritative Outcome Evidence. Independent of the operation source.</t></dd>
<dt>Reconciler</dt>
<dd><t>Classifies unresolved outcomes, authorizes retry, compensation or recovery under current authority.</t></dd>
<dt>Interim Effectuation Validator (optional)</dt>
<dd><t>Evaluates evidence of a bounded real effect before continuation authority is created (<xref target="fig-f4"/>).</t></dd>
</dl>
</section>
<section anchor="trust">
<name>Trust Assumptions</name>
<ul>
<li>The operation source is untrusted for its own descriptions and may be compromised, prompt-injected or malicious.</li>
<li>The Protected Enforcement Domain and the Finality Sink are trusted to the extent of their stated integrity guarantees (for example hardware isolation or a separate administrative domain). Their compromise is within the threat model and is addressed by redundancy, attestation and bounded blast radius, not assumed away.</li>
<li>Effect-relevant epochs are authentic and monotonic; where an epoch source is unreachable the epoch is treated as not current.</li>
<li>The sink is the first boundary at which the act becomes externally effective, or the deployment has explicitly declared and tested the uncovered routes (REQ-09, REQ-16).</li>
<li>Destination evidence is authentic. Where a destination cannot produce evidence, outcomes remain unresolved and the architecture offers containment (quarantine), not certainty.</li>
<li>Time sources used for validity periods are accurate within a stated tolerance.</li>
</ul>
</section>
</section>
<section anchor="reqs">
<name>Candidate Interoperability Requirements</name>
<t>This section lists candidate interoperability requirements. Each applies only to an implementation that elects to claim the named property. Each has an enforcement rationale and a testable failure condition. The tag PROFILE denotes a candidate profile requirement; ARCH denotes an architectural recommendation. No mandatory wire format, registry or transport is imposed.</t>
<section anchor="req-01">
<name>REQ-01: Stable act identity and canonical operation representation</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>An implementation claiming this property <bcp14>MUST</bcp14> derive a stable act identity from a canonical representation of the operation, such that two representations that cause the same effect map to the same identity and a materially different effect maps to a different one. The canonicalization <bcp14>MUST</bcp14> be deterministic and <bcp14>MUST</bcp14> be applied at validation and again at the sink.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Without a canonical form, an attacker or a translation layer can present a different serialization to the validator and to the sink (semantic or encoding substitution).</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Present the sink with an operation that differs only in a field the validator ignored; the sink <bcp14>MUST</bcp14> reject. Present two encodings of one effect; both <bcp14>MUST</bcp14> map to one identity.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E1-E38; T1, T14</t></dd>
</dl>
</section>
<section anchor="req-02">
<name>REQ-02: Current source authority and delegation restrictions</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>The validator <bcp14>MUST</bcp14> evaluate the current Source-Authority Envelope including the delegation chain, and <bcp14>MUST</bcp14> ensure that each delegation narrows or preserves, and never widens, the permitted scope. Restrictions present at origin <bcp14>MUST</bcp14> remain binding after summarization, handoff, forwarding or translation.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Authority that is reconstructed from model-visible or forwarded content can be lost or invented in transit; authority must come from a record the transforming component cannot rewrite.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Delegate A to B to C and have C request a wider scope; the request <bcp14>MUST</bcp14> be denied. Summarize context and drop a prohibition; the later effect <bcp14>MUST</bcp14> still be denied.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E4, E15, E17, E20, E24, E36; T3, T1</t></dd>
</dl>
</section>
<section anchor="req-03">
<name>REQ-03: Actual destination and effective target identity</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>The sink <bcp14>MUST</bcp14> resolve and compare the actual destination and effective target (including effective principal) with those bound in the capability. A nominal target name <bcp14>MUST NOT</bcp14> be accepted where resolution can change between approval and effect.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Aliases, redirects, re-resolution, account reassignment and target drift are the commonest TOCTOU form.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Change DNS, alias or account mapping after validation; the sink <bcp14>MUST</bcp14> reject the changed destination.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E22, E23, E24, E28, E34; T6</t></dd>
</dl>
</section>
<section anchor="req-04">
<name>REQ-04: Effect-relevant policy, configuration and authority epochs</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>The capability <bcp14>MUST</bcp14> bind the policy, revocation and (where relevant) configuration epochs under which it was issued, and the sink <bcp14>MUST</bcp14> reject a capability issued under an epoch older than the sink's current epoch for that consequence class.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Epoch binding makes revocation and policy change effective for already issued but unconsumed authority.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Advance the policy epoch after issuance; presenting the capability <bcp14>MUST</bcp14> fail.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E3, E9, E10, E13, E25; T4, T5, T13</t></dd>
</dl>
</section>
<section anchor="req-05">
<name>REQ-05: Protection against replay, duplication and stale authorization</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>A capability <bcp14>MUST</bcp14> be single-use or bounded-use as designated, and the sink <bcp14>MUST</bcp14> record consumption so that a second presentation, including from a restored replica or a retry worker, is rejected.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Possession of a valid capability must not allow repeated effect.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Present the same capability twice, including after a failover; the second <bcp14>MUST</bcp14> be rejected.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E5, E14, E32, E33; T5, T13, T15</t></dd>
</dl>
</section>
<section anchor="req-06">
<name>REQ-06: Current revocation and consumed-capability checks</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>Where revocation is claimed to be effective, the sink <bcp14>MUST</bcp14> check the current Revocation Epoch (or equivalent) at effectuation, and <bcp14>MUST</bcp14> define the maximum staleness it tolerates for revocation data.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Revocation that is only evaluated at issuance does not bind authority already in flight.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Revoke after issuance and before presentation; presentation <bcp14>MUST</bcp14> fail within the stated staleness bound.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E1, E3, E9; T4, T5</t></dd>
</dl>
</section>
<section anchor="req-07">
<name>REQ-07: Independent evidence qualification</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>Evidence relied upon for a validation or continuation decision <bcp14>MUST</bcp14> originate outside the operation source, <bcp14>MUST</bcp14> be bound to the act and phase, and <bcp14>MUST</bcp14> be checked for coverage and freshness (including, for absence-of-event evidence, sensing opportunity). Self-reported completion by the source <bcp14>MUST NOT</bcp14> be accepted as Authoritative Outcome Evidence.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Evidence the source can write or withhold gives the source control of the decision that is supposed to constrain it.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Have the source write a PASS record; it <bcp14>MUST NOT</bcp14> advance the act. Remove a sensor from coverage; a 'no event' inference <bcp14>MUST NOT</bcp14> be admitted.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E12A, E14, E16, E19, E27; T8, T9</t></dd>
</dl>
</section>
<section anchor="req-08">
<name>REQ-08: Effect-bound authorization</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>The authorization predicate <bcp14>MUST</bcp14> be evaluated over the reconstructed effect, including aggregate and transitive effects relevant to the consequence class, and the capability's scope <bcp14>MUST</bcp14> be no wider than the validated effect.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>A request-level check can pass while the combined or derived effect is outside scope.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Submit two individually acceptable work products whose combination violates a limit; the aggregate <bcp14>MUST</bcp14> be withheld.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E2, E4, E7, E21, E26, E31, E33; T3</t></dd>
</dl>
</section>
<section anchor="req-09">
<name>REQ-09: Actual effectuation-route enforcement</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>Enforcement <bcp14>MUST</bcp14> be located at the actual effectuation boundary for each covered route. A deployment claiming coverage <bcp14>MUST</bcp14> identify the routes it covers and <bcp14>MUST</bcp14> demonstrate, for each, that an operation lacking valid authority is rejected at the sink.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Controls located upstream of an alternate route do not constrain the alternate route.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Invoke the effect through an alternate API, credential or hardware path; if the route is declared covered, the sink <bcp14>MUST</bcp14> reject.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E25, E10, E36; T11, T12</t></dd>
</dl>
</section>
<section anchor="req-10">
<name>REQ-10: Compatibility with local protected state and cryptographic mechanisms</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>ARCH</t></dd>
<dt>Statement</dt>
<dd><t>Implementations <bcp14>SHOULD</bcp14> allow the capability to be realized by cryptographic means (for example signed or MACed tokens), by protected local state, by hardware interlocks, or by a hybrid, and <bcp14>SHOULD</bcp14> state which is used for which consequence class. Fast paths <bcp14>SHOULD</bcp14> use pre-validated local state with cryptographic validation at lease creation, renewal, change and revocation rather than per event.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>A single mechanism cannot meet all latency and trust constraints; the property to preserve is the binding, not a specific cipher.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Run the same scenario under cryptographic and local-state realizations; observable accept/reject outcomes <bcp14>MUST</bcp14> be equivalent.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E1, E5, E10, E13, E38; T7</t></dd>
</dl>
</section>
<section anchor="req-11">
<name>REQ-11: Multi-phase progression where appropriate</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>ARCH</t></dd>
<dt>Statement</dt>
<dd><t>Where a consequence is progressive, implementations <bcp14>SHOULD</bcp14> permit a bounded real effect followed by authenticated effect evidence and a continuation condition (<xref target="fig-f4"/>). A later phase <bcp14>MUST NOT</bcp14> exceed the maximum authorized cumulative envelope merely because earlier phases succeeded.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Evidence of a small real effect can justify a larger one without granting authority for the whole session.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Request a next phase exceeding the remaining envelope; it <bcp14>MUST</bcp14> be denied even after a successful first phase.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E37, E38; T8</t></dd>
</dl>
</section>
<section anchor="req-12">
<name>REQ-12: Handling of uncertain or partially completed effects</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>A timeout, lost acknowledgment, worker failure or restart <bcp14>MUST NOT</bcp14> be treated as evidence that an effect did not occur. The implementation <bcp14>MUST</bcp14> keep a logical effect identity distinct from attempt identity, <bcp14>MUST</bcp14> hold conflicting continuations while the outcome is unresolved, and <bcp14>MUST</bcp14> classify the outcome from Authoritative Outcome Evidence before authorizing retry, compensation or recovery.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Blind retry after an unknown outcome can duplicate an irreversible effect.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Drop the acknowledgment of a committed payment; a retry <bcp14>MUST NOT</bcp14> be admitted until the outcome is reconciled.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E14, E32, E33, E34; T9, T15, T18</t></dd>
</dl>
</section>
<section anchor="req-13">
<name>REQ-13: Crash-consistent consumption and replay protection where technically required</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>Where single-use is claimed, consumption of the capability and the governed effect <bcp14>MUST</bcp14> be atomic or recoverable such that a crash cannot yield both a retained valid capability and a completed effect, nor a consumed capability without a recorded effect state.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Non-atomic consume-then-act or act-then-consume orders create duplicate or lost effects across crashes.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Kill the sink between consume and effect and between effect and record; on recovery the state <bcp14>MUST</bcp14> be classifiable and no duplicate <bcp14>MUST</bcp14> occur.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E5, E14, E32; T5, T13</t></dd>
</dl>
</section>
<section anchor="req-14">
<name>REQ-14: Cross-platform semantic equivalence</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>Where an operation, policy or authority contract is translated between protocols, engines or platforms, the implementation <bcp14>MUST</bcp14> verify, per consequence class, that the destination preserves the semantics the contract requires, and <bcp14>MUST</bcp14> fail closed (deny or declare uncovered) for features the destination does not preserve.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Syntactic acceptance of a translated request does not show that restrictions survived.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Translate a request with a prohibition into a protocol lacking that feature; the result <bcp14>MUST</bcp14> be denial or an explicit uncovered declaration.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E20, E30; T14</t></dd>
</dl>
</section>
<section anchor="req-15">
<name>REQ-15: Provider-hosted tool execution that cannot be intercepted by an application-side gateway</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>Where a tool runs in a provider-hosted environment that an application-side gateway cannot intercept (<xref target="fig-f6"/>), authority <bcp14>MUST</bcp14> be enforced at the end-target sink, or the tool <bcp14>MUST</bcp14> be disabled for covered consequence classes, or the path <bcp14>MUST</bcp14> be declared uncovered.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>An application-side control does not see a call that never passes through the application.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Enable a hosted tool that calls a covered target; the end target <bcp14>MUST</bcp14> reject without valid authority, or the tool <bcp14>MUST</bcp14> be unavailable.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E20, E25; T10, T11</t></dd>
</dl>
</section>
<section anchor="req-16">
<name>REQ-16: Alternate-path and bypass analysis</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>PROFILE</t></dd>
<dt>Statement</dt>
<dd><t>A deployment claiming non-bypassability <bcp14>MUST</bcp14> perform and document an alternate-path analysis covering credentials, APIs, hardware paths, management planes, retry workers and administrative routes, and <bcp14>MUST</bcp14> treat any known act-equivalent path as disabled, mediated, restricted or declared uncovered.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Claims of non-bypassability are only as strong as the route inventory.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Add a new credential with direct access to the target; a route-coverage test <bcp14>MUST</bcp14> fail or the credential <bcp14>MUST</bcp14> be rejected at the sink.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E25, E10, E6, E36; T11, T12</t></dd>
</dl>
</section>
<section anchor="req-17">
<name>REQ-17: Latency-aware validation at actual consequential boundaries</name>
<dl newline="true">
<dt>Tag</dt>
<dd><t>ARCH</t></dd>
<dt>Statement</dt>
<dd><t>Implementations <bcp14>SHOULD</bcp14> place full validation at slower consequential boundaries (management, configuration, authorization, settlement, deployment) and <bcp14>SHOULD</bcp14> use pre-validated bounded local state for low-latency loops, validating at lease creation, renewal, change and revocation. No latency figure is specified or claimed here.</t></dd>
<dt>Enforcement rationale</dt>
<dd><t>Per-event cryptographic validation can be infeasible in microsecond loops; the control point must match the time scale of the consequence.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Measure that a revocation takes effect within the declared bound for a fast path without per-event signature verification.</t></dd>
<dt>Related embodiments and threats</dt>
<dd><t>E2, E10, E13, E38; T17</t></dd>
</dl>
</section>
</section>
<section anchor="latency">
<name>Latency-Aware Placement</name>
<t>The architecture distinguishes two time scales. First, consequential management, configuration, authorization, settlement and deployment boundaries, where a validation step can be placed before the effect without making the operation infeasible. Second, low-latency radio, scheduling and safety-control loops, where a per-event cryptographic check may be infeasible. For the second class this document proposes only that validation occur when protected local state is created, renewed, changed or revoked, with the fast path enforcing the resulting bounded state using local protected checks (<xref target="reqs"/>, REQ-10 and REQ-17). No latency figure is stated or implied. Whether a given loop can be covered at all, and with what safety consequences, is a matter for measurement in the deployment and for the safety case of that system.</t>
<t>Safety consequence: for cyber-physical effects, validator unavailability <bcp14>MUST NOT</bcp14> translate into an unsafe default. The safe state (hold, continue last safe configuration, or stop) is system specific and is chosen by the safety case, not by this document.</t>
</section>
<section anchor="profiles">
<name>Application Profiles</name>
<t>The following profiles apply the architecture to specific boundaries. They are illustrative of disclosed mechanisms and are not wire protocols.</t>
<section anchor="p-iev">
<name>Bounded Real Effect and Continuation</name>
<t>Where a consequence is progressive, a first phase produces a bounded real effect. Authenticated evidence of that effect is evaluated, and only then is a continuation condition created for the next phase (<xref target="fig-f4"/>). Evidence is a structural dependency, not a log: the next sink verifies or consumes the continuation condition. This is the idea developed in the cited companion draft on receipt-gated IEV; it is included here as an optional profile and not redefined. Unknown or indeterminate evidence holds the act; it never defaults to pass.</t>
<figure anchor="fig-f4">
<name>Bounded real effect, authenticated effect evidence, and optional continuation authorization</name>
<artwork type="ascii-art"><![CDATA[
 Candidate Act A, Envelope_MAX
        |
        v
+--------------------+    scope1 <= Envelope_MAX
| Validation Phase 1 |--------------------------+
+---------+----------+                          |
          v                                     |
+--------------------+                          |
| Finality Sink FS1  |                          |
+---------+----------+                          |
          v                                     |
 [ BOUNDED REAL EFFECT E1 ]  (not a simulation) |
          |                                     |
          v                                     |
+--------------------------+                    |
| Effect Evidence R1       |  destination-      |
| (authenticated, bound to |  origin, not       |
|  act, phase, nonce)      |  self-reported     |
+------------+-------------+                    |
             v                                  |
+--------------------------+                    |
| Interim Effectuation     |  expected X1 vs    |
| Validation (IEV)         |  observed O1       |
+------+-----------+-------+                    |
 PASS  |           | FAIL / UNKNOWN             |
       v           v                            |
+-------------+  +-----------------------+      |
| Continuation|  | HOLD / SAFE STATE /   |      |
| condition   |  | RECONCILE (no blind   |      |
| G2 created  |  | retry)                |      |
+------+------+  +-----------------------+      |
       v                                        |
+--------------------+  verifies/consumes G2    |
| Finality Sink FS2  |<-------------------------+
+--------------------+  scope2 <= Envelope_MAX - spent]]></artwork>
</figure>
</section>
<section anchor="p-agents">
<name>Cross-Agent Composition (E21)</name>
<t>When several agents produce work products, each may pass local validation while their composition does not. The profile materializes the composition at a shared commit boundary, evaluates cross-branch reads and writes, negative constraints and cumulative limits, and authorizes the accepted aggregate or a validated subset (<xref target="fig-f5"/>).</t>
<figure anchor="fig-f5">
<name>Cross-agent work-product composition and aggregate-effect validation</name>
<artwork type="ascii-art"><![CDATA[
 +---------+   +---------+   +---------+
 | Agent A |   | Agent B |   | Agent C |
 | WP-A ok |   | WP-B ok |   | WP-C ok |
 +----+----+   +----+----+   +----+----+
      |  locally valid, each one      |
      +------------+-----------------+
                   v
       +-------------------------+
       | Composition snapshot    |  materialized merge result,
       | (shared commit boundary)|  not three separate diffs
       +------------+------------+
                    v
       +-------------------------+
       | Aggregate validator     |  cross-branch reads/writes,
       |                         |  negative constraints,
       +------+-------------+----+  cumulative limits
        reject|             |accept (whole or valid subset)
              v             v
        +-----------+  +-------------------+
        | withhold  |  | scoped capability |
        | aggregate |  | for accepted state|
        +-----------+  +---------+---------+
                                 v
                        +-------------------+
                        | Merge/commit sink |
                        +-------------------+]]></artwork>
</figure>
</section>
<section anchor="p-mcp">
<name>Cross-Protocol Tool Dispatch (E20)</name>
<t>Conversion between model APIs, MCP adapters and provider-hosted or client-managed tool execution can drop restrictions or change effect semantics. The profile keeps a canonical authority contract outside the model context, a feature and exception matrix, and an effective tool-location map. For provider-hosted tools that an application-side gateway cannot intercept, enforcement moves to the end-target sink (<xref target="fig-f6"/>). This document does not assert which specific providers expose which hosted-tool modes; that must be verified per provider and version.</t>
<figure anchor="fig-f6">
<name>Cross-protocol MCP/tool-dispatch authority preservation; provider-hosted versus client-managed execution</name>
<artwork type="ascii-art"><![CDATA[
 +----------------------------------------------+
 | Canonical authority contract (outside model  |
 | context): prohibitions, scope, tool-location |
 | map, feature/exception matrix                |
 +--------------------+-------------------------+
                      | consulted at every hop
 Agent runtime        v
 +----------+   +-------------+   +-----------------------+
 | model    |-->| protocol    |-->| tool-call dispatch    |
 | output   |   | bridge /    |   +-----+-----------+-----+
 +----------+   | MCP adapter |         |           |
                +-------------+  CLIENT-MANAGED   PROVIDER-HOSTED
                                        |           |
                                        v           v
              +-------------------+   +-------------------------+
              | app-side gateway  |   | provider executes tool; |
              | can intercept     |   | app gateway cannot      |
              +---------+---------+   | intercept the call      |
                        |             +------------+------------+
                        v                          |
              +-------------------+                |
              | tool server / MCP |                |
              +---------+---------+                |
                        +-------------+------------+
                                      v
                       +-------------------------------+
                       | END-TARGET FINALITY SINK      |
                       | (independent of both paths)   |
                       +-------------------------------+
 Rule: where a path cannot be intercepted, authority must be
 enforced at the end target, or the path must be disabled or
 declared uncovered.]]></artwork>
</figure>
</section>
<section anchor="p-offload">
<name>Hardware Offload Equivalence (E10)</name>
<t>A software decision that is compiled into wildcarded, prioritized, cached or stateful hardware rules may have a different real effect from the decision. The profile compares the effective installed behavior (read back from the device where supported) with the permitted effect and fences stale fast-path authority on policy change (<xref target="fig-f7"/>). Compared with existing practice (P4 runtime verification, match-action verification, consistent updates), the distinguishing element to examine is the binding of installed-effect equivalence to the software authorization and to an authority epoch (<xref target="appendix-exam"/>).</t>
<figure anchor="fig-f7">
<name>Hardware-offload equivalence between authorized software action and installed fast-path behavior</name>
<artwork type="ascii-art"><![CDATA[
 Slow path (software)               Fast path (hardware)
+---------------------------+      +---------------------------+
| Slow-Path Authorization   |      | Effective Hardware Action |
| Witness  W                |      | Descriptor  H             |
| + Offload Effect Contract |      | (wildcards, priority,     |
|   permitted behavior      |      |  caches, state, actions)  |
+-------------+-------------+      +-------------+-------------+
              |        compile / install         |
              +--------------+-------------------+
                             v
               +-----------------------------+
               | Offload-equivalence check:  |
               | effects(H) subset-of        |
               |   permitted(W, Contract)?   |
               | + Fast-Path Authority Epoch |
               +--------+-----------+--------+
                  FAIL  |           | PASS
                        v           v
              +-------------+  +----------------------+
              | do not      |  | enable offload; read |
              | enable;     |  | back installed rules |
              | remove rule |  | and compare          |
              +-------------+  +-----------+----------+
                                           v
                              +--------------------------+
                              | epoch change / revocation|
                              | => stale fast-path rules |
                              | fenced (barrier)         |
                              +--------------------------+]]></artwork>
</figure>
</section>
<section anchor="p-coverage">
<name>Execution-Route Discovery and Coverage (E25)</name>
<t>A dashboard, policy document or compliant sample trace does not show that every effect-producing route is controlled. The profile builds a route inventory, relates required controls to actual enforcement points, and obtains evidence of rejection at the sink for each route, using non-consequential probes where safe (<xref target="fig-f8"/>). Routes that cannot be probed safely or covered are declared uncovered, and authority for the associated consequence classes is lowered accordingly.</t>
<figure anchor="fig-f8">
<name>Execution-route discovery and governance enforcement coverage</name>
<artwork type="ascii-art"><![CDATA[
+-------------------+      +----------------------------------+
| Route inventory   |      | Governance coverage contract     |
| APIs, credentials,|      | required control per effect class|
| hardware, network |      +----------------+-----------------+
| paths, tools      |                       |
+---------+---------+                       |
          +-------------------+-------------+
                              v
                +-------------------------------+
                | Governance-to-effect graph    |
                +---------------+---------------+
                                v
                +-------------------------------+
                | Coverage probe: attempt a     |
                | non-consequential test effect |
                | on each route (where safe)    |
                +--------+---------------+------+
                 sink    |               | effect got through
                 rejects |               | without control
                         v               v
              +--------------+   +--------------------+
              | route COVERED|   | route UNCOVERED:   |
              | (evidence)   |   | disable, restrict, |
              +--------------+   | or lower authority |
                                 +--------------------+
 Logs and dashboards are not coverage evidence; sink rejection is.]]></artwork>
</figure>
</section>
<section anchor="p-db">
<name>Database Commit and Downstream Effects (E31, E32)</name>
<t>A permitted stored procedure can reach triggers, definer-rights code, autonomous transactions and external functions beyond its entry point (E31). A downstream consumer of an outbox or change-data-capture event should not treat an uncommitted intent, a stale outbox row or an unrelated transaction's event as authority; release is bound to a source-commit certificate and current effect state, with reconciliation after uncertain acknowledgments (E32). <xref target="fig-f9"/> shows the flow. The transactional outbox and CDC are established patterns (<xref target="OUTBOX"/>); the profile concerns binding consumer effect authority to the source commit.</t>
<figure anchor="fig-f9">
<name>Database source commit, downstream event provenance, and consequential effectuation</name>
<artwork type="ascii-art"><![CDATA[
+--------------------+  txn  +-----------------------------+
| Agent / app        |------>| Source database             |
| proposes write     |       | (stored programs, triggers) |
+--------------------+       +--------------+--------------+
                                            |
          transitive effect closure of the  | COMMIT
          entry point is reconstructed      v
                                +-----------------------------+
                                | Commit log / outbox / CDC   |
                                | source-commit certificate   |
                                +--------------+--------------+
 uncommitted intent  X  no authority            |
 stale outbox row    X  no authority            v
 unrelated txn event X  no authority +----------------------------+
                                     | Downstream consumer        |
                                     | derives child operation    |
                                     +-------------+--------------+
                                                   v
                                     +----------------------------+
                                     | Effect Finality Sink       |
                                     | checks: commit certificate,|
                                     | policy epoch, child scope  |
                                     +-------------+--------------+
                                                   v
                                     +----------------------------+
                                     | External effect + ack      |
                                     | uncertain ack => reconcile |
                                     +----------------------------+]]></artwork>
</figure>
</section>
<section anchor="p-ran">
<name>6G and Open RAN Management-Plane Boundaries (E2, E37, E38)</name>
<t>At the non-real-time management plane, an AI-generated configuration or slice change is reconstructed into its effective multi-cell and slice state, validated, and applied as a limited real change; attributable impact evidence bounds expansion (<xref target="fig-f10"/>). Slice capacity lending (E38) additionally requires conservation (lending creates no capacity), preservation of the lender's entitlement floor and an enforceable reclaim bound. Real-time loops are out of scope and use pre-validated local state (<xref target="latency"/>). Relevant established mechanisms include 3GPP network slicing and the O-RAN architecture (<xref target="TS23501"/>, <xref target="ORAN-ARCH"/>).</t>
<figure anchor="fig-f10">
<name>6G/Open RAN authorization at network-management and configuration-effect boundaries</name>
<artwork type="ascii-art"><![CDATA[
 NON-REAL-TIME / MANAGEMENT PLANE (in scope)
+--------------+   +----------------+   +---------------------+
| rApp / AI    |-->| SMO / Non-RT   |-->| Candidate config    |
| optimizer    |   | RIC policy     |   | or slice change     |
+--------------+   +----------------+   +----------+----------+
                                                   v
                                   +-------------------------------+
                                   | Reconstruct effective state:  |
                                   | multi-cell, neighbors, slices,|
                                   | affected-service footprint    |
                                   +---------------+---------------+
                                                   v
                                   +-------------------------------+
                                   | Validator + Finality Sink at  |
                                   | the configuration-effect point|
                                   +---------------+---------------+
                                                   v
                                   +-------------------------------+
                                   | Limited real change (canary)  |
                                   +---------------+---------------+
                       attributable impact evidence|
                                                   v
                                   +-------------------------------+
                                   | Causal impairment evaluator   |
                                   | expand only if bounded impact |
                                   +-------------------------------+
 OUT OF SCOPE: PHY / MAC / HARQ and other real-time loops. Those use
 prevalidated local state; validation occurs at lease or policy
 creation, renewal, change and revocation, not per transport block.]]></artwork>
</figure>
</section>
<section anchor="p-pay">
<name>Agentic Payment Exposure (E33, E34, E35)</name>
<t>Multiple pending authorizations, captures and agent-issued instructions can collectively exceed a mandate. The profile maintains a conservative exposure ledger, reserves capacity before admitting an operation, and keeps unresolved exposure held until authoritative evidence resolves it (<xref target="fig-f11"/>). Existing network and issuer controls remain authoritative; the profile is complementary. For cross-border payments the nominal instruction, settlement stages and beneficiary-net outcome are distinguished (E34, with ISO 20022 tracking as established context <xref target="ISO20022"/>). For offline digital cash, E35 addresses delegated spending authority and post-compromise fork reconciliation and does not claim that secure-element protection is ineffective under intact-hardware assumptions.</t>
<figure anchor="fig-f11">
<name>Agentic payment exposure reservation, actual settlement or beneficiary evidence, and uncertainty reconciliation</name>
<artwork type="ascii-art"><![CDATA[
+---------------------------+
| Protected Exposure Mandate|  principal, agents, aggregate limit,
+-------------+-------------+  beneficiary/merchant restrictions
              v
+---------------------------+      +--------------------------+
| Exposure ledger           |<---->| reserved | authorized |  |
| conservative accounting   |      | captured | released   |  |
+-------------+-------------+      | UNRESOLVED (kept held)   |
              v                    +--------------------------+
+---------------------------+
| Reserve capacity, then    |  capacity must exist BEFORE the
| issue transaction-bound   |  operation is admitted
| capability                |
+-------------+-------------+
              v
+---------------------------+   authorization --> capture -->
| Payment finality sink     |   clearing --> settlement
+-------------+-------------+
              v
+---------------------------+  timeout or lost ack is NOT failure
| Outcome evidence:         |------------+
| settlement / beneficiary  |            v
| net amount, FX, fees      |   +--------------------------+
+-------------+-------------+   | quarantine; query        |
              v                 | authoritative status;    |
+---------------------------+   | retry/compensate only    |
| Reconcile ledger          |<--| under current authority  |
+---------------------------+   +--------------------------+]]></artwork>
</figure>
</section>
</section>
<section anchor="states">
<name>Operational States and Transitions</name>
<t><xref target="fig-f12"/> shows the state machine with the attack and recovery paths. Every transition not drawn defaults to DENIED or HELD.</t>
<figure anchor="fig-f12">
<name>Attack and recovery state machine: replay, stale epochs, substitution, revocation, timeout, fail-closed</name>
<artwork type="ascii-art"><![CDATA[
                 stage
 [PROPOSED] ------------> [NON_EFFECTIVE]
                               | reconstruct
                               v
                        [RECONSTRUCTED]--substitution/mismatch--+
                               | validate                       |
                               v                                v
                        [VALIDATING]--stale epoch/revoked-->[DENIED]
                               | all predicates true             ^
                               v                                 |
                        [AUTHORIZED]--expiry-->[EXPIRED]         |
                               | present at sink                 |
                               v                                 |
      replay / consumed -->[SINK_CHECK]---mismatch/no cap--------+
      capability = reject      | admit + consume (atomic)
                               v
                        [EFFECTUATING]
                          |        |
       effect evidence    |        | validator or evidence
                          v        | unavailable
                    [CONFIRMED]    v
                          [HELD] <---- [INDETERMINATE] (timeout or
                             |              ^           lost ack)
                             +-- reconcile--+
                                   | authoritative evidence
                                +--> [CONFIRMED] or [NOT_EFFECTED]
 Every arrow not drawn defaults to DENIED or HELD (fail closed).
 Retry from INDETERMINATE requires fresh authority and
 reconciliation.]]></artwork>
</figure>
<table anchor="tbl-states">
<name>Operational states</name>
<thead><tr><th>State</th><th>Entered when</th><th>Leaves to</th></tr></thead>
<tbody>
<tr><td>PROPOSED</td><td>Source submits act</td><td>NON_EFFECTIVE</td></tr>
<tr><td>NON_EFFECTIVE</td><td>Act staged; no consequence</td><td>RECONSTRUCTED</td></tr>
<tr><td>RECONSTRUCTED</td><td>Actual Pending Operation derived</td><td>VALIDATING; DENIED (mismatch)</td></tr>
<tr><td>VALIDATING</td><td>Predicates being evaluated</td><td>AUTHORIZED; DENIED; HELD</td></tr>
<tr><td>AUTHORIZED</td><td>Capability issued</td><td>SINK_CHECK; EXPIRED</td></tr>
<tr><td>SINK_CHECK</td><td>Capability presented at sink</td><td>EFFECTUATING; DENIED</td></tr>
<tr><td>EFFECTUATING</td><td>Capability consumed with effect</td><td>CONFIRMED; INDETERMINATE; HELD</td></tr>
<tr><td>INDETERMINATE</td><td>Timeout, lost ack, partial effect</td><td>HELD; CONFIRMED; NOT_EFFECTED</td></tr>
<tr><td>HELD</td><td>Validator or evidence unavailable, or quarantine</td><td>VALIDATING (fresh); DENIED</td></tr>
<tr><td>CONFIRMED / NOT_EFFECTED / DENIED / EXPIRED</td><td>Terminal for this act identity</td><td>(new act required)</td></tr>
</tbody>
</table>
<t>Retry from INDETERMINATE is not a transition back to AUTHORIZED. It requires a fresh validation under current authority after reconciliation, and it carries the same logical effect identity so that a duplicate is detectable (<xref target="reqs"/>, REQ-12).</t>
</section>
<section anchor="interop">
<name>Interoperability Considerations</name>
<t>Because no wire format is defined, interoperability here means preserving the binding between act, effect, authority and epoch across components from different vendors. Candidate carriers for a Scoped Execution Capability include CWT/COSE and JWT (<xref target="RFC8392"/>, <xref target="RFC9052"/>, <xref target="RFC7519"/>); sender-constraint mechanisms exist (<xref target="RFC9449"/>, <xref target="RFC8705"/>); attestation evidence may use RATS roles and EAT (<xref target="RFC9334"/>, <xref target="RFC9711"/>). Existing deployments can often remain unchanged above and below the sink, provided the effect-holding credential is held by the sink rather than by the source. Where a component cannot be integrated, the route <bcp14>MUST</bcp14> be declared uncovered rather than assumed protected (REQ-16).</t>
<t>Legacy components that may remain unchanged include identity providers, existing RBAC/ABAC policy engines, transaction logs and message brokers; the architecture adds a binding and a check at the effect point. Whether any particular legacy component can host a sink is a deployment question this document does not answer.</t>
</section>
<section anchor="failure">
<name>Failure Handling and Recovery</name>
<dl newline="true">
<dt>Validator unavailable</dt>
<dd><t>Operations HOLD; no default allow. Fast paths continue under pre-validated bounded state until its expiry, then fail to the declared safe state.</t></dd>
<dt>Epoch source unreachable or stale</dt>
<dd><t>The epoch is treated as not current for consequence classes that require it.</t></dd>
<dt>Sink crash between consume and effect</dt>
<dd><t>Recovery classifies the act from sink and destination state before any retry (REQ-13).</t></dd>
<dt>Timeout or lost acknowledgment</dt>
<dd><t>Outcome is INDETERMINATE; competing operations are held; reconciliation against the authoritative outcome criterion precedes retry or compensation (REQ-12).</t></dd>
<dt>Contradictory evidence</dt>
<dd><t>Treated as unresolved; escalated to human or multi-party resolution as the profile specifies; never defaulted to success.</t></dd>
<dt>Revocation during flight</dt>
<dd><t>Unconsumed capability is rejected at the sink; in-flight effects follow the profile's fencing rule (for example E12 for queued segments, E3 for desired-state controllers).</t></dd>
<dt>Partial or irreversible physical effect</dt>
<dd><t>Contain, hold further phases, move to safe state, then reconcile; compensation requires fresh authority.</t></dd>
<dt>Reclaim and unwind</dt>
<dd><t>Where lending or delegation was made (E38, E35), reclamation or fork reconciliation proceeds under the profile's deadline and evidence rules.</t></dd>
</dl>
</section>
<section anchor="threats">
<name>Threat Model</name>
<section anchor="adv">
<name>Assumptions and Adversary Capabilities</name>
<t>The adversary may control or influence the operation source and its inputs, may observe and replay messages, may race operations, may obtain credentials for routes that bypass the intended path, and may attempt to disable the validator. The adversary is not assumed to control both a genuine validator and a genuine sink simultaneously; if it does, the architecture offers no protection. The architecture does not promise that it prevents every attack; each threat below lists residual risk.</t>
</section>
<section anchor="t1">
<name>T1: Compromised AI agents and untrusted model-generated operations</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Controls or influences the model output; can propose arbitrary Candidate Acts and describe them misleadingly.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Operation source to protected enforcement domain (B1).</t></dd>
<dt>Protected invariant</dt>
<dd><t>No act becomes effective unless the reconstructed operation is validated against current authority.</t></dd>
<dt>Mitigation</dt>
<dd><t>Validation over the reconstructed operation (REQ-01, REQ-08); sink re-derivation (REQ-03); quotas on proposals.</t></dd>
<dt>Residual risk</dt>
<dd><t>an operation within valid scope but harmful in intent; mandate limits bound the harm but do not remove it.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>A compromised agent submits an act whose description says 'read' but whose effect is 'write'; the sink <bcp14>MUST</bcp14> reject.</t></dd>
</dl>
</section>
<section anchor="t2">
<name>T2: Prompt injection and tool-output manipulation</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Places instructions in retrieved content, tool results or messages that the model treats as authority.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Model context versus authority record.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Instructions found in data never create or widen authority.</t></dd>
<dt>Mitigation</dt>
<dd><t>Authority taken from a record outside the model context (REQ-02); quoted or forwarded content not operative (E24); tool output treated as data.</t></dd>
<dt>Residual risk</dt>
<dd><t>injection may still cause an in-scope but unwanted operation.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Inject 'ignore prior limits' via tool output; subsequent act beyond the original scope <bcp14>MUST</bcp14> be denied.</t></dd>
</dl>
</section>
<section anchor="t3">
<name>T3: Delegated authority expansion</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Uses a delegation chain, handoff or summarization to obtain a wider scope than the originator held.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Delegation boundaries and context transformations.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Each hop narrows or preserves scope.</t></dd>
<dt>Mitigation</dt>
<dd><t>REQ-02; authority record survives compaction (E15); human capsule minimizes disclosure (E17); aggregate and transitive checks (REQ-08).</t></dd>
<dt>Residual risk</dt>
<dd><t>legitimately broad originating authority.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Chain A to B to C where C asks for broader scope; <bcp14>MUST</bcp14> be denied; summarization removes a prohibition; later effect <bcp14>MUST</bcp14> still be denied.</t></dd>
</dl>
</section>
<section anchor="t4">
<name>T4: TOCTOU changes between approval and execution</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Changes target, configuration, dependency topology or policy after validation.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Time gap between B1 decision and B2/B3 effect.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Effect occurs only under the state that was validated or a state that is at least as restrictive.</t></dd>
<dt>Mitigation</dt>
<dd><t>Short validity; re-derivation at the sink (REQ-03); epoch binding (REQ-04); live-topology revalidation (E28).</t></dd>
<dt>Residual risk</dt>
<dd><t>change after the sink's check but before the effect in systems without atomic commit.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Change a dependency edge after validation; the sink <bcp14>MUST</bcp14> reject or re-validate.</t></dd>
</dl>
</section>
<section anchor="t5">
<name>T5: Authorization replay and stale capabilities</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Copies, replays or retains a capability; restores a replica that lost consumption state.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Capability store and sink-local state.</t></dd>
<dt>Protected invariant</dt>
<dd><t>A capability produces at most its designated effect.</t></dd>
<dt>Mitigation</dt>
<dd><t>Non-bearer binding; consumption record (REQ-05, REQ-13); epoch binding (REQ-06).</t></dd>
<dt>Residual risk</dt>
<dd><t>loss of consumption state without detection; mitigated by fencing and state commitments.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Present the same capability to a failed-over replica; the second use <bcp14>MUST</bcp14> be rejected.</t></dd>
</dl>
</section>
<section anchor="t6">
<name>T6: Changed destinations and effective principals</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Changes the resolved destination, account or principal behind a nominal name.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Name resolution and identity mapping at B2.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Effect binds to the destination and principal validated.</t></dd>
<dt>Mitigation</dt>
<dd><t>REQ-03; bind to resolved identity not nominal name.</t></dd>
<dt>Residual risk</dt>
<dd><t>legitimate re-assignment between sink check and effect.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Reassign an account after validation; the sink <bcp14>MUST</bcp14> reject.</t></dd>
</dl>
</section>
<section anchor="t7">
<name>T7: Validator substitution or compromise</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Replaces or controls the validator, or forges its decisions.</t></dd>
<dt>Trust boundary</dt>
<dd><t>B1.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Only a genuine validator in a genuine domain can issue authority; the sink does not rely solely on validator output.</t></dd>
<dt>Mitigation</dt>
<dd><t>Attestation of the validator (<xref target="RFC9334"/>); sink-side re-derivation; multi-validator quorum for high-consequence classes; bounded capability scope.</t></dd>
<dt>Residual risk</dt>
<dd><t>a compromised validator and a compromised sink together defeat the control.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Substitute a validator with an unattested build; sinks <bcp14>MUST</bcp14> refuse its capabilities.</t></dd>
</dl>
</section>
<section anchor="t8">
<name>T8: Forged, replayed or self-reported effect receipts</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Forges or reuses a receipt, or writes a self-reported 'done'.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Evidence path from destination to validator.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Only evidence bound to the act, phase and nonce and originating outside the source advances an act.</t></dd>
<dt>Mitigation</dt>
<dd><t>REQ-07; single-use receipts; binding to act and phase; destination-origin evidence.</t></dd>
<dt>Residual risk</dt>
<dd><t>a compromised destination.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Replay a prior receipt for a new phase; <bcp14>MUST</bcp14> be rejected.</t></dd>
</dl>
</section>
<section anchor="t9">
<name>T9: Missing, delayed or contradictory outcome evidence</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Suppresses or delays evidence, or induces contradictions among sources.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Evidence collection.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Absence or conflict is never treated as success or as failure.</t></dd>
<dt>Mitigation</dt>
<dd><t>REQ-07, REQ-12; hold and quarantine; reconciliation against authoritative criteria (E14, E19).</t></dd>
<dt>Residual risk</dt>
<dd><t>prolonged unresolved outcome ties up exposure.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Withhold evidence; the act <bcp14>MUST</bcp14> remain HELD and no retry <bcp14>MUST</bcp14> be admitted.</t></dd>
</dl>
</section>
<section anchor="t10">
<name>T10: Confused-deputy and privileged-tool execution</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Induces a privileged tool or service to act on the attacker's behalf.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Privileged tool versus act scope.</t></dd>
<dt>Protected invariant</dt>
<dd><t>A tool's privilege never exceeds the scope of the act on whose behalf it runs.</t></dd>
<dt>Mitigation</dt>
<dd><t>Credential isolation; capability bound to act and tool; REQ-15.</t></dd>
<dt>Residual risk</dt>
<dd><t>tools that cannot be wrapped.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Have a privileged tool accept an act-scoped request that asks for broader privilege; <bcp14>MUST</bcp14> be denied.</t></dd>
</dl>
</section>
<section anchor="t11">
<name>T11: Alternate API, credential, hardware or network bypass routes</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Reaches the target through a route the sink does not mediate.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Route inventory versus enforcement points.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Every covered route reaches the effect only through a sink.</t></dd>
<dt>Mitigation</dt>
<dd><t>REQ-09, REQ-16; route discovery and rejection evidence (E25).</t></dd>
<dt>Residual risk</dt>
<dd><t>undiscovered routes.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Add a credential with direct target access; the coverage test <bcp14>MUST</bcp14> flag it or the sink <bcp14>MUST</bcp14> reject its use.</t></dd>
</dl>
</section>
<section anchor="t12">
<name>T12: Hardware DMA, SmartNIC, DPU and accelerator offload paths</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Uses DMA, residual translations, offloaded rules or accelerator queues to cause an effect outside software checks.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Software authorization versus installed hardware behavior.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Installed behavior is within the authorized effect and prior-domain access paths are closed before reassignment.</t></dd>
<dt>Mitigation</dt>
<dd><t>E6 prior-path closure; E10 equivalence and readback; fast-path epoch.</t></dd>
<dt>Residual risk</dt>
<dd><t>hardware that cannot report installed state.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Install a rule broader than authorized; enablement <bcp14>MUST</bcp14> be refused. Reassign a device with an open DMA mapping; transition <bcp14>MUST</bcp14> be withheld.</t></dd>
</dl>
</section>
<section anchor="t13">
<name>T13: Rollback, failover, checkpoint restoration and stale worker authority</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Restores a checkpoint or fails over so that two workers hold effect authority.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Worker identity and ownership epoch.</t></dd>
<dt>Protected invariant</dt>
<dd><t>At most one worker holds effect-commit authority for a logical computation.</t></dd>
<dt>Mitigation</dt>
<dd><t>E5 source fence, destination admission, effect watermark; REQ-05, REQ-13.</t></dd>
<dt>Residual risk</dt>
<dd><t>loss of fencing state.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Restore a checkpoint while the source still runs; the source's publish <bcp14>MUST</bcp14> be rejected after cutover.</t></dd>
</dl>
</section>
<section anchor="t14">
<name>T14: Cross-protocol semantic loss</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Exploits translation that drops restrictions or changes effect semantics.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Protocol bridges.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Translated operation preserves authority semantics or fails closed.</t></dd>
<dt>Mitigation</dt>
<dd><t>REQ-14, REQ-15; feature and exception matrix; end-target enforcement.</t></dd>
<dt>Residual risk</dt>
<dd><t>unknown unsupported features.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Translate a prohibition into a protocol lacking it; result <bcp14>MUST</bcp14> be denial or declared uncovered.</t></dd>
</dl>
</section>
<section anchor="t15">
<name>T15: Transaction races, duplicate payment attempts and unresolved settlement</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Races concurrent agents or retries to exceed a mandate or duplicate a payment.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Exposure ledger and payment sink.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Reserved plus committed plus unresolved exposure never exceeds the mandate.</t></dd>
<dt>Mitigation</dt>
<dd><t>E33 reservation before admission; E14 retry quarantine; conservative unresolved accounting.</t></dd>
<dt>Residual risk</dt>
<dd><t>counterparty behavior outside the ledger.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Run N concurrent agents; the sum <bcp14>MUST</bcp14> not exceed the mandate; lose an ack; retry <bcp14>MUST</bcp14> be held.</t></dd>
</dl>
</section>
<section anchor="t16">
<name>T16: Compromised secure elements and offline delegated authority</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Extracts keys or clones state from an offline secure element.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Secure element assumptions.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Conflicting transfers become detectable and loss is bounded by allocation limits.</t></dd>
<dt>Mitigation</dt>
<dd><t>E35 bounded child allocations, lineage and recipient evidence, fork reconciliation.</t></dd>
<dt>Residual risk</dt>
<dd><t>undiscovered conflicting transfers; privacy cost of reconciliation.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Clone state and spend twice offline; reconciliation <bcp14>MUST</bcp14> identify the fork and the bounded exposure.</t></dd>
</dl>
</section>
<section anchor="t17">
<name>T17: Denial of service caused by unavailable validators</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Floods or disables validators or epoch sources to stall operations or force unsafe defaults.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Availability of B1.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Unavailability yields HOLD/safe state, never implicit allow.</t></dd>
<dt>Mitigation</dt>
<dd><t>Redundant validators; local pre-validated bounded state for fast paths; bounded hold and safe-state rules; rate limits.</t></dd>
<dt>Residual risk</dt>
<dd><t>availability loss is a real cost; fail-closed trades availability for safety.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Disable the validator; consequential operations <bcp14>MUST</bcp14> be held and no default allow <bcp14>MUST</bcp14> occur.</t></dd>
</dl>
</section>
<section anchor="t18">
<name>T18: Recovery after partial or irreversible physical effects</name>
<dl newline="true">
<dt>Attacker capability</dt>
<dd><t>Causes or exploits a partially completed physical or irreversible effect.</t></dd>
<dt>Trust boundary</dt>
<dd><t>Effect domain.</t></dd>
<dt>Protected invariant</dt>
<dd><t>Partial effect is classified and contained before any continuation.</t></dd>
<dt>Mitigation</dt>
<dd><t>Bounded real effect; hold; authoritative outcome criteria; safe-state; compensation only under current authority (REQ-11, REQ-12).</t></dd>
<dt>Residual risk</dt>
<dd><t>some effects cannot be undone.</t></dd>
<dt>Testable failure condition</dt>
<dd><t>Interrupt a multi-step actuation; further steps <bcp14>MUST</bcp14> be held pending classification.</t></dd>
</dl>
</section>
</section>
<section anchor="security">
<name>Security Considerations</name>
<t>This document follows the guidance of <xref target="RFC3552"/>. The threat analysis in <xref target="threats"/> is part of these Security Considerations. Additional points:</t>
<ul>
<li>The architecture is an additional layer. It does not replace authentication, authorization, attestation, network security (for example TLS 1.3, <xref target="RFC8446"/>), safety engineering or incident response, and is weakest where the sink cannot be placed at the true first effectual boundary.</li>
<li>Fail-closed behavior trades availability for safety. Operators <bcp14>SHOULD</bcp14> define, per consequence class, the safe state and the maximum hold duration.</li>
<li>Compromise of the validator and the sink together defeats the control. Separation of administrative domains, attestation and quorum mitigate but do not eliminate this.</li>
<li>Reconstruction is only as complete as the model of effects. Unmodelled branches are a primary residual risk; route coverage evidence (REQ-09) and dependency graphs help.</li>
<li>Evidence admission is security critical. A sink that accepts self-reported evidence gives the source control of continuation.</li>
<li>No security claim in this document is supported by experimental results. Any implementation needs its own testing, including the testable failure conditions listed above.</li>
</ul>
</section>
<section anchor="privacy">
<name>Privacy Considerations</name>
<t>This section follows <xref target="RFC6973"/>. The architecture creates records that can contain personal or confidential data and can enable correlation:</t>
<ul>
<li>Canonical operation representations and receipts may carry identifiers of users, counterparties, amounts and destinations. Implementations <bcp14>SHOULD</bcp14> minimize the data bound into a capability or receipt and <bcp14>SHOULD</bcp14> avoid including content that is not needed for the decision (data minimization).</li>
<li>Receipts, consumption records and ledgers create long-lived linkable records. Retention, access control and purpose limitation <bcp14>SHOULD</bcp14> be defined, and unlinkability techniques <bcp14>SHOULD</bcp14> be considered where the consequence class allows.</li>
<li>Human-assistance capsules (E17) are intended to limit disclosure to the bounded task and evidence needed.</li>
<li>Audience and presentation profiles (E11, E12, E24) process information about who can perceive output; this is itself personal data.</li>
<li>Offline cash reconciliation (E35) has an inherent tension between detecting forks and preserving transaction privacy; the profile aims to protect appropriate privacy but this document does not resolve the trade-off.</li>
<li>Cross-engine data-policy checks (E30) and cohort reconciliation (E29) apply the data subject's existing consent and suppression state and do not create new lawful bases.</li>
</ul>
</section>
<section anchor="iana">
<name>IANA Considerations</name>
<t>This document has no IANA actions. A separate individual draft by the same author, draft-das-ef-registries, proposes registries for the execution-finality family; nothing in this document depends on that proposal.</t>
</section>
<section anchor="prior-art">
<name>Comparison with Established Standards and Prior Art</name>
<t>This section credits established work and states candidate differences without asserting originality. Patentability arguments (motivation to combine, reasonable expectation of success, technical interdependence) are kept in <xref target="appendix-exam"/> and are not part of the proposed architecture.</t>
<dl newline="true">
<dt>RFC 9334 RATS architecture; RFC 9711 EAT</dt>
<dd><t>Established function: Attestation evidence, verifier appraisal and relying-party policy.</t><t>Overlap and candidate difference: E9 concerns correspondence between remediation, evidence and policy per consequence class, and E10/E6 bind attested or installed state to specific effects. It uses RATS roles and does not replace them.</t></dd>
<dt>OAuth 2.0, token exchange, DPoP, mTLS-bound tokens, RAR</dt>
<dd><t>Established function: Delegated authorization, sender constraint, fine-grained authorization details.</t><t>Overlap and candidate difference: The candidate additions are re-derivation of the effective operation at the resource boundary, binding to epochs and consumption, and route-coverage evidence. A RAR authorization detail could carry part of the scope.</t></dd>
<dt>JWT, CWT, COSE</dt>
<dd><t>Established function: Claim and signature formats.</t><t>Overlap and candidate difference: Candidate carriers only; no format is selected.</t></dd>
<dt>Distributed fencing tokens and generation counters</dt>
<dd><t>Established function: Prevent stale holders from writing after lease change.</t><t>Overlap and candidate difference: E5 and E3 use epoch fencing for effect-commit authority of a computation and for reconciliation generations. The difference to examine is the combination with checkpoint cutover, destination admission and an effect watermark.</t></dd>
<dt>Database transactions and the transactional outbox</dt>
<dd><t>Established function: Atomic state change plus reliable event publication; CDC.</t><t>Overlap and candidate difference: E32 binds downstream effect authority to the source commit and to effect state; E31 addresses transitive stored-program effects. Established patterns are credited; the examined element is the authority binding.</t></dd>
<dt>Service-mesh enforcement and workload identity</dt>
<dd><t>Established function: Per-hop identity and policy.</t><t>Overlap and candidate difference: E36 traces parent authority across subrequests, queues and background workers and enforces at outbound effect paths. Mesh policy is closely related and must be compared.</t></dd>
<dt>Programmable-network policy verification</dt>
<dd><t>Established function: P4Runtime, match-action verification, consistent updates.</t><t>Overlap and candidate difference: E10 compares installed effective behavior with a software-authorized effect and fences stale fast-path authority. Existing verification work overlaps and must be compared.</t></dd>
<dt>3GPP mobility, ATSSS and network slicing</dt>
<dd><t>Established function: Access steering, mobility hysteresis, slice isolation and resource management.</t><t>Overlap and candidate difference: E7 evaluates cumulative connectivity trajectory; E38 evaluates slice lending with conservation and enforceable reclaim. 3GPP mechanisms are the primary comparison.</t></dd>
<dt>Secure elements and offline cash</dt>
<dd><t>Established function: Offline double-spend prevention under intact-hardware assumptions.</t><t>Overlap and candidate difference: E35 considers delegated allocations and post-compromise fork reconciliation. It does not claim that existing designs lack double-spend protection under their own assumptions.</t></dd>
<dt>Zero trust, RBAC, ABAC</dt>
<dd><t>Established function: Principle of continuous verification; role and attribute decisions.</t><t>Overlap and candidate difference: Consistent with and layered over; the added element is verification at the actual effect boundary.</t></dd>
</dl>
</section>
<section anchor="refinements">
<name>Foundational Architecture and Later Consequence-Specific Refinements</name>
<t>This section is non-normative. Per Source C and the applicant's description, the June 2026 mothership (PCT/IB2026/055615) already describes the broad protected execution-finality architecture, including Candidate Acts, non-effective staging, protected evidence and receipts, cryptographic validation, Finality Sinks, and accelerator and network egress control in many domains. The later Indian provisional (202631136555, 15 November 2026) is a specialized disclosure that examines narrower implementations and combinations.</t>
<t>Limit of this comparison: the editor did not have the mothership JSON text. Statements about what the mothership contains are taken from the applicant's comparison and are not independently verified. This document does not claim that any feature below is absent from the original text; a feature not appearing under an exact phrase may still be disclosed expressly or inherently. Distinguishing a disclosure refinement from independently established novelty or priority is a matter for examination against the as-filed text and relevant prior art.</t>
<dl newline="true">
<dt>Accelerator authority cutover (E5)</dt>
<dd><t>Technical predicate: Source fence, destination admission and effect watermark give a single effect-commit owner.</t><t>Actual effect governed: Externally authoritative compute output (publish or model-state commit) during migration or failover.</t><t>Claimed functional interdependence: Checkpoint transfer, fencing and the sink's admission check must cooperate; copying state alone confers no authority.</t></dd>
<dt>Hardware trust-domain access-path closure (E6)</dt>
<dd><t>Technical predicate: Evidence that old-domain DMA, interrupt, debug and queued-I/O paths are closed.</t><t>Actual effect governed: Reassignment of a peripheral, memory region or accelerator to another trust domain.</t><t>Claimed functional interdependence: Successor authority is withheld until path-quiescence evidence from the controller and the translation hardware is joined.</t></dd>
<dt>Cumulative temporal connectivity limits (E7)</dt>
<dd><t>Technical predicate: Cumulative effect of transition history compared with a temporal envelope.</t><t>Actual effect governed: Discretionary access transitions of a device.</t><t>Claimed functional interdependence: A protected transition record and the next transition's validation cooperate.</t></dd>
<dt>Consequence-specific composed numerical uncertainty (E8)</dt>
<dd><t>Technical predicate: Composed error versus decision margin; permit, request higher precision, or block.</t><t>Actual effect governed: Release of a numerical decision to a consequential consumer.</t><t>Claimed functional interdependence: Model/precision state, error composition and the consequence's margin must be evaluated together.</t></dd>
<dt>Remediation, attestation and policy correspondence (E9)</dt>
<dd><t>Technical predicate: Patch state, fresh evidence, reference values and policy match per consequence class.</t><t>Actual effect governed: Admission of workload effects after a security remediation.</t><t>Claimed functional interdependence: Remediation controller, attestation evidence controller and class-admission matrix cooperate.</t></dd>
<dt>Software-to-installed-hardware effect equivalence (E10)</dt>
<dd><t>Technical predicate: Installed fast-path match/action effects are a subset of the permitted effect, with an authority epoch.</t><t>Actual effect governed: Hardware-accelerated processing of authorized flows.</t><t>Claimed functional interdependence: Slow-path witness, offload contract, effective hardware descriptor and barrier cooperate.</t></dd>
<dt>Multimodal route and audience continuity (E11, E12)</dt>
<dd><t>Technical predicate: Audience and route evaluated across concurrent and in-flight output.</t><t>Actual effect governed: Human-perceptible presentation of information.</t><t>Claimed functional interdependence: Presentation controller, route state and queue fencing cooperate.</t></dd>
<dt>Negative-observation admissibility (E12A)</dt>
<dd><t>Technical predicate: 'No event' admitted only with sensing opportunity and coverage.</t><t>Actual effect governed: Actions inferred from absence of an event.</t><t>Claimed functional interdependence: Sensor health, coverage and time-gap evidence feed the admission decision.</t></dd>
<dt>Agent-context authority continuity (E15)</dt>
<dd><t>Technical predicate: Authority register independent of model context binds later calls.</t><t>Actual effect governed: Tool calls after context summarization or handoff.</t><t>Claimed functional interdependence: Register, dispatcher and credential isolation cooperate.</t></dd>
<dt>Verification-oracle independence (E16)</dt>
<dd><t>Technical predicate: Acceptance criteria and oracle isolated from the modifying agent; deployed bytes bound to approval.</t><t>Actual effect governed: Release of agent-produced software.</t><t>Claimed functional interdependence: Oracle isolation and artifact binding cooperate at the promotion gate.</t></dd>
<dt>Cross-protocol tool semantic preservation (E20)</dt>
<dd><t>Technical predicate: Canonical authority contract, feature/exception matrix and tool-location map.</t><t>Actual effect governed: Tool execution across model APIs and hosted or client-executed tools.</t><t>Claimed functional interdependence: Bridge, contract and end-target enforcement cooperate.</t></dd>
<dt>Aggregate work-product validation (E21)</dt>
<dd><t>Technical predicate: Materialized composition validated against cross-branch constraints.</t><t>Actual effect governed: Commit of combined agent work.</t><t>Claimed functional interdependence: Composition snapshot and aggregate validator cooperate at the merge boundary.</t></dd>
<dt>Governance enforcement-path coverage (E25)</dt>
<dd><t>Technical predicate: Route inventory and sink-rejection evidence.</t><t>Actual effect governed: Assurance that governance controls are enforceable at effect routes.</t><t>Claimed functional interdependence: Inventory, contract, probes and sink evidence cooperate.</t></dd>
<dt>Cross-engine data-policy equivalence (E30)</dt>
<dd><t>Technical predicate: Mandatory row, column, masking and recipient obligations survive the actual route.</t><t>Actual effect governed: Disclosure of governed data through alternative engines or shares.</t><t>Claimed functional interdependence: Policy semantics per engine and result-release validation cooperate.</t></dd>
<dt>Source-commit-certified downstream effects (E32)</dt>
<dd><t>Technical predicate: Consumer effect authority bound to a source-commit certificate.</t><t>Actual effect governed: External operations derived from committed transactions.</t><t>Claimed functional interdependence: Source-transaction controller, certificate and consumer sink cooperate.</t></dd>
<dt>Agentic payment exposure (E33)</dt>
<dd><t>Technical predicate: Conservative aggregate exposure with reservation before admission.</t><t>Actual effect governed: Payment obligations from one or more agents.</t><t>Claimed functional interdependence: Mandate, ledger, reservation and capture reconciliation cooperate.</t></dd>
<dt>Offline cash post-compromise reconciliation (E35)</dt>
<dd><t>Technical predicate: Delegated allocations, lineage and recipient evidence for fork attribution and bounded loss.</t><t>Actual effect governed: Offline monetary transfers.</t><t>Claimed functional interdependence: Allocation continuity, transfer evidence and fork reconciliation cooperate.</t></dd>
<dt>Causal radio impairment (E37)</dt>
<dd><t>Technical predicate: Attributable impact evidence bounds expansion of a limited real change.</t><t>Actual effect governed: Progressive deployment of RAN optimization.</t><t>Claimed functional interdependence: Canary change, impact attribution and expansion gate cooperate.</t></dd>
<dt>6G slice-lending conservation and reclaimability (E38)</dt>
<dd><t>Technical predicate: Lending conserves capacity, preserves floors and is reclaimable by a deadline.</t><t>Actual effect governed: Temporary slice resource loans.</t><t>Claimed functional interdependence: Entitlement table, lease state, reclaim state and scheduler enforcement cooperate.</t></dd>
</dl>
</section>
<section anchor="eu-applicability">
<name>European Critical-Infrastructure and Digital-Sovereignty Applicability</name>
<t>This section is non-normative. It explains engineering relevance only. This document is not an EU compliance standard, and nothing here states that the architecture satisfies any legal obligation. Obligations vary by actor, product, sector and application date; readers should consult the legal texts and competent authorities.</t>
<dl newline="true">
<dt>EU AI Act (<xref target="EU-AIA"/>)</dt>
<dd><t>The Regulation addresses risk management, record-keeping, human oversight and robustness for certain AI systems. A mechanism that makes a high-consequence action conditional on current, evidence-backed authority at the point of effect, with records of what was authorized and what occurred, is of engineering relevance to these themes. It is not shown here to satisfy any article.</t></dd>
<dt>NIS2 (<xref target="EU-NIS2"/>)</dt>
<dd><t>Entities in covered sectors manage cybersecurity risk and incident handling. Bounded blast radius, route coverage evidence and fail-closed holds relate to risk-management measures at an engineering level.</t></dd>
<dt>DORA (<xref target="EU-DORA"/>)</dt>
<dd><t>Financial entities manage ICT risk and third-party risk. Exposure ledgers, reconciliation of unresolved outcomes and control over agent-initiated operations relate to operational resilience themes (E14, E33, E34).</t></dd>
<dt>Cyber Resilience Act (<xref target="EU-CRA"/>)</dt>
<dd><t>Products with digital elements require security by design and vulnerability handling. The remediation-attestation correspondence profile (E9) and artifact-promotion gate (E16) are relevant to engineering of update and release flows.</t></dd>
<dt>European digital sovereignty</dt>
<dd><t>Jurisdiction-bound, region-bound authority and evidence held by European operators can be modelled as policy predicates at a sink. The architecture does not itself guarantee data residency or provider independence.</t></dd>
<dt>European 6G research</dt>
<dd><t>AI-native orchestration and network automation research may benefit from management-plane authorization at configuration-effect boundaries (E2, E37, E38). No claim is made about any research programme's work or standards contribution.</t></dd>
<dt>Telecommunications service continuity</dt>
<dd><t>Causal impairment control (E37), slice lending with reclaim (E38) and cumulative connectivity limits (E7) address service continuity at the management and device-policy level.</t></dd>
<dt>Industrial and operational-technology security</dt>
<dd><t>Protected hand-off of setpoints, deployments and remote commands can be gated at the actuation or deployment boundary, with local interlocks retained and explicit time budgets.</t></dd>
<dt>Banking and digital-euro research</dt>
<dd><t>Exposure control for agent payments (E33), beneficiary-net outcomes (E34) and offline post-compromise reconciliation (E35) are research-level extensions. Existing secure-element and scheme controls remain authoritative; the Eurosystem's own research should be consulted (<xref target="ECB-DE"/>).</t></dd>
<dt>Trusted and confidential computing</dt>
<dd><t>Attestation-to-effect linkage (E9), trust-domain transitions (E6) and offload equivalence (E10) relate to confidential computing deployments and require the relevant hardware interfaces.</t></dd>
<dt>Critical energy, transport and communications infrastructure</dt>
<dd><t>Autonomous change and distributed actuation can be controlled by effect-specific gates. The architecture does not claim universal safety or certification.</t></dd>
</dl>
<section anchor="eu-latency">
<name>Low-Latency Loops Versus Slower Consequential Boundaries</name>
<t>Radio scheduling, HARQ, protection relays and similar safety-control loops operate at time scales where per-event authorization is generally infeasible; they rely on pre-validated local state and on certified safety functions. The proposals in this document address the slower boundaries where AI-generated or autonomous decisions become configuration, authorization, settlement or deployment effects. Where both are present, validation occurs when the bounded local state is created, renewed, changed or revoked (<xref target="latency"/>).</t>
</section>
<section anchor="zenodo">
<name>Relationship to the Related European Research Publication</name>
<t>Sangam Das, "EUROPE'S NEXT AI SECURITY CHALLENGE: WHO CONTROLS THE FINAL ACTION? From EU AI Act and NIS2 to 6G, the Digital Euro, Industrial AI and Critical Infrastructure", published 16 November 2026, DOI 10.5281/zenodo.24274259 (<xref target="ZENODO-EU"/>). It is a non-peer-reviewed public research publication by the same author. Its relevance to this document is that it discusses, from a European perspective, who controls the final action in AI and autonomous systems, which is the question this document turns into an architecture. The editor could not retrieve the record when preparing this text, so this description is limited to the title and author-supplied context. Neither this document nor that publication establishes regulatory compliance or technical verification.</t>
</section>
</section>
<section anchor="industry">
<name>Industry Integration Profiles (Proposed)</name>
<t>This section is non-normative and AI-assisted. Each entry separates verified public context from proposed integration. Verified context is a coarse public description that readers must re-verify against current vendor documentation. Proposed integration points are the author's engineering hypotheses and are not adoption, endorsement or infringement statements. No vendor has been shown to implement the proposed combinations.</t>
<section>
<name>AI and cloud: OpenAI Agents SDK</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public SDK with function tools, handoffs and guardrails.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Place the sink in the tool dispatcher: before a function tool with a consequential effect runs, the dispatcher re-derives the operation, checks the authority record held outside the prompt, and rejects calls whose target or scope differs. It matters because handoffs and summarization can otherwise lose prohibitions (E15, E20).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Agent code and model unchanged; real credentials held by the dispatcher or a proxy, not the model process. Adds one local check per consequential call; no latency measurement is claimed.</t></dd>
</dl>
</section>
<section>
<name>AI and cloud: Anthropic Claude tool use and MCP</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public tool-use interfaces; MCP is an open protocol for tool and context servers.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Place enforcement in the MCP client/host dispatch and at the MCP server's end target: the host looks up the canonical authority contract for each tool call, the server's sink verifies a scoped capability against the actual operation and rejects widened scope. It matters for cross-protocol semantic loss (E20) and confused-deputy risk (T10).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>MCP servers continue to expose tools; the capability is an additional parameter or header; downstream credentials stay server-side.</t></dd>
</dl>
</section>
<section>
<name>AI and cloud: Microsoft Azure AI (Foundry agents)</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public managed agent service with tool integration and managed identities.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Insert the sink in the connector that holds the downstream credential: the managed identity is used only by the sink after capability verification, so an agent cannot call the target directly. It matters for alternate-credential bypass (T11).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Existing RBAC and identity unchanged; the agent loses direct access to the target credential.</t></dd>
</dl>
</section>
<section>
<name>AI and cloud: Google Cloud AI (ADK, Vertex agents, A2A)</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public agent framework and agent-to-agent protocol.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>In multi-agent runs, validate the composed work product at the shared commit step (E21), and preserve the authority contract across agent-to-agent hand-offs (E15, E20).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Framework unchanged; commit step gains a validator call.</t></dd>
</dl>
</section>
<section>
<name>AI and cloud: AWS agent services</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public managed agent runtime with gateway and policy features.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Where a gateway already enforces policy, add end-target sink checks for effect classes the gateway cannot see (provider-hosted or direct-credential paths), and gather rejection evidence for route coverage (E25).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Gateway unchanged; adds coverage evidence and an end-target check.</t></dd>
</dl>
</section>
<section>
<name>Networking: Ericsson AI-RAN and automation platforms</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public automation platform with rApps in the non-real-time domain.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Gate the point where an rApp's output becomes an applied configuration: reconstruct multi-cell and slice effect, validate, apply as a limited change, and expand only on attributable impact evidence (E2, E37). It matters because a locally successful change can harm neighbors.</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>rApps and SMO unchanged; adds an admission step; PHY/MAC loops untouched.</t></dd>
</dl>
</section>
<section>
<name>Networking: Nokia MantaRay</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public SMO and RAN automation offering.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Same placement as above for SMO policy and slice control; for slice lending, the slice controller enforces entitlement floors and reclaim deadlines (E38).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Existing controllers unchanged; lease state added.</t></dd>
</dl>
</section>
<section>
<name>Networking: Open RAN, 5G Advanced, 6G slicing, edge/mobile</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>O-RAN defines Non-RT/Near-RT RIC and open interfaces; 3GPP defines slicing.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Admission and effect verification at A1/O1-style management actions; no per-transport-block validation. Cumulative connectivity policy at the device (E7).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Standard interfaces unchanged; new checks sit in management or device policy.</t></dd>
</dl>
</section>
<section>
<name>Enterprise: SAP Business AI and Joule</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public agents operating on SAP business processes.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Insert a late-bound validator at the posting or release boundary: recompute tax, FX, pricing and allocation effects from live state and compare with the approval before the posting commits (E26). It matters because the approved document and the posted effect can differ.</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>ERP logic remains authoritative; the check calls the same computation in preview mode where available.</t></dd>
</dl>
</section>
<section>
<name>Enterprise: Oracle database and ERP</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public database with stored programs, triggers and agent integration.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Close the effect set of a stored program before a privileged mutation: enumerate triggers, definer-rights code and autonomous transactions reachable from the entry point and require authority for each material effect (E31).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Database security remains authoritative; enforcement in a supported hook or proxy.</t></dd>
</dl>
</section>
<section>
<name>Enterprise: Salesforce Agentforce</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public agent platform on CRM data and flows.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Gate customer-facing 'completed' statements on authoritative fulfillment evidence (E27) and validate cohort fan-out (E29).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Flows unchanged; communication dispatch gains an evidence check.</t></dd>
</dl>
</section>
<section>
<name>Enterprise: ServiceNow AI workflows</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public workflow and ITOM features.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Revalidate the live dependency topology at the remediation commit (E28) and quarantine unresolved outcomes (E14).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>CMDB unchanged; commit step adds a check.</t></dd>
</dl>
</section>
<section>
<name>Enterprise: IBM watsonx</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public governance and orchestration products.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Governance dashboards can export route inventory; add sink-rejection evidence per route to turn dashboards into coverage proof (E25).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Existing governance product unchanged.</t></dd>
</dl>
</section>
<section>
<name>Enterprise: Databricks Unity Catalog</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public catalog with row and column controls.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Test that obligations survive each query route, including shared or alternative engines, and validate the actual result released (E30); reconcile activation populations (E29).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Catalog policies unchanged; adds route-level equivalence tests.</t></dd>
</dl>
</section>
<section>
<name>Enterprise: Snowflake Cortex</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public catalog and agent features.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Same as above for cross-engine semantics and result release (E30).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Native governance unchanged.</t></dd>
</dl>
</section>
<section>
<name>Compute: NVIDIA accelerators and BlueField</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public GPU partitioning and DPU offload.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Single-writer output cutover on failover (E5); offload readback versus authorized flow (E10).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Drivers unchanged; requires device interfaces for installed-state readback, which must be verified.</t></dd>
</dl>
</section>
<section>
<name>Compute: AMD Instinct and Pensando</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public accelerators and DPU.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Same placements (E5, E10).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Dependent on device interfaces.</t></dd>
</dl>
</section>
<section>
<name>Compute: Intel confidential computing</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public TDX and attestation services.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Join attestation evidence with patch state and policy per consequence class before admitting effects (E9) and close prior-domain paths on reassignment (E6).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Attestation service unchanged.</t></dd>
</dl>
</section>
<section>
<name>Compute: Arm CCA</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public confidential compute architecture with realm management.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Domain transition controller withholds successor authority until DMA and interrupt paths are closed (E6).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Depends on SMMU and RME interfaces.</t></dd>
</dl>
</section>
<section>
<name>Compute: Qualcomm on-device and Apple/Samsung devices</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public SoCs and mobile platforms with sensor hubs and output routing.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Wake-time revalidation of escrowed intents (E13); audience-bound release across display and casting routes (E11, E12); connectivity trajectory limits (E7).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Requires OS vendor interfaces; no adoption is implied.</t></dd>
</dl>
</section>
<section>
<name>Finance: ECB digital euro research, offline payments</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public Eurosystem research on offline payments with secure hardware.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Research extension: delegated child allocations and post-compromise fork reconciliation (E35).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Existing secure-element controls remain authoritative.</t></dd>
</dl>
</section>
<section>
<name>Finance: Visa, Mastercard, Swift, Adyen, Worldline, tokenized settlement, agentic commerce</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public agent-payment initiatives and payment networks.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Exposure ledger and reservation at authorization-to-capture (E33); beneficiary-net checks and reconciliation at cross-border flows (E34); retry quarantine after lost acknowledgments (E14).</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Network controls remain authoritative; participants control only their own phases.</t></dd>
</dl>
</section>
<section>
<name>Industrial: Siemens, Schneider Electric, industrial edge, digital twins, robotics, energy</name>
<dl newline="true">
<dt>Verified public context (re-verify)</dt>
<dd><t>Public industrial edge platforms and automation suites.</t></dd>
<dt>Proposed enforcement interface and how it works</dt>
<dd><t>Gate deployments and remote commands at the edge runtime and at the actuation boundary; retain local interlocks; use digital-twin results as evidence, not as authority.</t></dd>
<dt>Legacy components, credentials, latency and safety</dt>
<dd><t>Safety functions unchanged; validator unavailability must map to the system's safe state.</t></dd>
</dl>
</section>
</section>
<section anchor="ipr">
<name>IPR Disclosure, Attribution and Non-Endorsement</name>
<t>Foundational reference: PCT/IB2026/055615 <xref target="PCT-055615"/>, filed 4 June 2026, published 16 July 2026 as WO 2026/150382 A2.</t>
<t>Later Indian provisional application: 202631136555 <xref target="IN-136555"/>, dated 15 November 2026 (as reported by the applicant from the filing acknowledgment; the supplied PDF is not a certified registry record).</t>
<t>Public research cross-reference: https://doi.org/10.5281/zenodo.24274259 <xref target="ZENODO-EU"/></t>
<t>Related public disclosures by the same author (informative): Technical Disclosure Commons records <xref target="TDC-12542"/>, <xref target="TDC-12667"/>, <xref target="TDC-12668"/>, <xref target="TDC-12853"/>, <xref target="TDC-12931"/>, <xref target="TDC-12961"/>; Zenodo records <xref target="ZEN-21861642"/>, <xref target="ZEN-22082995"/>, <xref target="ZEN-23040181"/>, <xref target="ZEN-23657985"/>; and the earlier Internet-Drafts cited in Section 1.3, including <xref target="I-D.das-execution-finality-protocol-layer"/>. Titles are as listed in the applicant's index and should be verified.</t>
<t>Attribution: the underlying technical invention is attributed to the human inventor, Sangam Kumar Das. AI-assisted drafting, summarization, indexing, translation or literature analysis does not establish independent inventorship.</t>
<t>Limits: citation of a patent document does not establish patent validity, priority, essentiality or licensing terms. This document contains no licensing commitment and makes no statement about royalties. Contributors are bound by BCP 78 and BCP 79 (<xref target="RFC5378"/>, <xref target="RFC8179"/>). The author will make the IPR disclosure for this document at the appropriate time, through the official IETF process and in accordance with IETF guidelines; text in this document is not a substitute for that formal disclosure. An earlier IPR declaration relating to related DAS Internet-Drafts is listed at <xref target="IETF-IPR-7610"/>; whether it covers this document has not been established. Readers should consult the IETF IPR disclosure database for the current status of any declaration relating to this author's drafts.</t>
<t>Source integrity: the Indian filing text supplied to the editor contains 427 physically present numbered claims (1-273 and 276-429). Claims 274 and 275 are absent and claim 273 ends mid-sentence (at 'current Group Audience'). This document does not reconstruct the missing text. Verify against the official filed copy (<xref target="appendix-lineage"/>).</t>
<t>Non-endorsement: this is an individual technical proposal and not an endorsed IETF standard, EU regulatory approval, telecommunications standard, financial certification or implementation by a named company. Company and product references indicate potential technical integration points and are not evidence of adoption, sponsorship, copying, infringement or commercial affiliation. Because AI-generated material may contain errors, readers should verify all patent numbers, dates, claim quotations, product capabilities, standards references and technical assertions against primary sources.</t>
</section>
</middle>
<back>
<references><name>Normative References</name>
<reference anchor="RFC2119" target="https://www.rfc-editor.org/info/rfc2119"><front><title>Key words for use in RFCs to Indicate Requirement Levels</title><author initials="S" surname="Bradner"/><date year="1997" month="March"/></front><seriesInfo name="RFC" value="2119"/><seriesInfo name="DOI" value="10.17487/RFC2119"/><seriesInfo name="BCP" value="14"/></reference>
<reference anchor="RFC8174" target="https://www.rfc-editor.org/info/rfc8174"><front><title>Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words</title><author initials="B" surname="Leiba"/><date year="2017" month="May"/></front><seriesInfo name="RFC" value="8174"/><seriesInfo name="DOI" value="10.17487/RFC8174"/><seriesInfo name="BCP" value="14"/></reference>
</references>
<references><name>Informative References</name>
<reference anchor="RFC3552" target="https://www.rfc-editor.org/info/rfc3552"><front><title>Guidelines for Writing RFC Text on Security Considerations</title><author initials="E" surname="Rescorla"/><author initials="B" surname="Korver"/><date year="2003" month="July"/></front><seriesInfo name="RFC" value="3552"/><seriesInfo name="DOI" value="10.17487/RFC3552"/><seriesInfo name="BCP" value="72"/></reference>
<reference anchor="RFC6973" target="https://www.rfc-editor.org/info/rfc6973"><front><title>Privacy Considerations for Internet Protocols</title><author initials="A" surname="Cooper"/><author initials="H" surname="Tschofenig"/><author initials="B" surname="Aboba"/><author initials="J" surname="Peterson"/><author initials="J" surname="Morris"/><author initials="M" surname="Hansen"/><author initials="R" surname="Smith"/><date year="2013" month="July"/></front><seriesInfo name="RFC" value="6973"/><seriesInfo name="DOI" value="10.17487/RFC6973"/></reference>
<reference anchor="RFC5378" target="https://www.rfc-editor.org/info/rfc5378"><front><title>Rights Contributors Provide to the IETF Trust</title><author initials="S" surname="Bradner"/><author initials="J" surname="Contreras"/><date year="2008" month="November"/></front><seriesInfo name="RFC" value="5378"/><seriesInfo name="DOI" value="10.17487/RFC5378"/><seriesInfo name="BCP" value="78"/></reference>
<reference anchor="RFC8179" target="https://www.rfc-editor.org/info/rfc8179"><front><title>Intellectual Property Rights in IETF Technology</title><author initials="S" surname="Bradner"/><author initials="J" surname="Contreras"/><date year="2017" month="May"/></front><seriesInfo name="RFC" value="8179"/><seriesInfo name="DOI" value="10.17487/RFC8179"/><seriesInfo name="BCP" value="79"/></reference>
<reference anchor="RFC9334" target="https://www.rfc-editor.org/info/rfc9334"><front><title>Remote ATtestation procedureS (RATS) Architecture</title><author initials="H" surname="Birkholz"/><author initials="D" surname="Thaler"/><author initials="M" surname="Richardson"/><author initials="N" surname="Smith"/><author initials="W" surname="Pan"/><date year="2023" month="January"/></front><seriesInfo name="RFC" value="9334"/><seriesInfo name="DOI" value="10.17487/RFC9334"/></reference>
<reference anchor="RFC9711" target="https://www.rfc-editor.org/info/rfc9711"><front><title>The Entity Attestation Token (EAT)</title><author initials="L" surname="Lundblade"/><author initials="G" surname="Mandyam"/><author initials="J" surname="O'Donoghue"/><author initials="C" surname="Wallace"/><date year="2025" month="April"/></front><seriesInfo name="RFC" value="9711"/><seriesInfo name="DOI" value="10.17487/RFC9711"/></reference>
<reference anchor="RFC6749" target="https://www.rfc-editor.org/info/rfc6749"><front><title>The OAuth 2.0 Authorization Framework</title><author initials="D" surname="Hardt"/><date year="2012" month="October"/></front><seriesInfo name="RFC" value="6749"/><seriesInfo name="DOI" value="10.17487/RFC6749"/></reference>
<reference anchor="RFC8693" target="https://www.rfc-editor.org/info/rfc8693"><front><title>OAuth 2.0 Token Exchange</title><author initials="M" surname="Jones"/><author initials="A" surname="Nadalin"/><author initials="B" surname="Campbell"/><author initials="J" surname="Bradley"/><author initials="C" surname="Mortimore"/><date year="2020" month="January"/></front><seriesInfo name="RFC" value="8693"/><seriesInfo name="DOI" value="10.17487/RFC8693"/></reference>
<reference anchor="RFC9449" target="https://www.rfc-editor.org/info/rfc9449"><front><title>OAuth 2.0 Demonstrating Proof of Possession (DPoP)</title><author initials="D" surname="Fett"/><author initials="B" surname="Campbell"/><author initials="J" surname="Bradley"/><author initials="T" surname="Lodderstedt"/><author initials="M" surname="Jones"/><author initials="D" surname="Waite"/><date year="2023" month="September"/></front><seriesInfo name="RFC" value="9449"/><seriesInfo name="DOI" value="10.17487/RFC9449"/></reference>
<reference anchor="RFC8705" target="https://www.rfc-editor.org/info/rfc8705"><front><title>OAuth 2.0 Mutual-TLS Client Authentication and Certificate-Bound Access Tokens</title><author initials="B" surname="Campbell"/><author initials="J" surname="Bradley"/><author initials="N" surname="Sakimura"/><author initials="T" surname="Lodderstedt"/><date year="2020" month="February"/></front><seriesInfo name="RFC" value="8705"/><seriesInfo name="DOI" value="10.17487/RFC8705"/></reference>
<reference anchor="RFC7800" target="https://www.rfc-editor.org/info/rfc7800"><front><title>Proof-of-Possession Key Semantics for JSON Web Tokens (JWTs)</title><author initials="M" surname="Jones"/><author initials="J" surname="Bradley"/><author initials="H" surname="Tschofenig"/><date year="2016" month="April"/></front><seriesInfo name="RFC" value="7800"/><seriesInfo name="DOI" value="10.17487/RFC7800"/></reference>
<reference anchor="RFC9396" target="https://www.rfc-editor.org/info/rfc9396"><front><title>OAuth 2.0 Rich Authorization Requests</title><author initials="T" surname="Lodderstedt"/><author initials="J" surname="Richer"/><author initials="B" surname="Campbell"/><date year="2023" month="May"/></front><seriesInfo name="RFC" value="9396"/><seriesInfo name="DOI" value="10.17487/RFC9396"/></reference>
<reference anchor="RFC7519" target="https://www.rfc-editor.org/info/rfc7519"><front><title>JSON Web Token (JWT)</title><author initials="M" surname="Jones"/><author initials="J" surname="Bradley"/><author initials="N" surname="Sakimura"/><date year="2015" month="May"/></front><seriesInfo name="RFC" value="7519"/><seriesInfo name="DOI" value="10.17487/RFC7519"/></reference>
<reference anchor="RFC8392" target="https://www.rfc-editor.org/info/rfc8392"><front><title>CBOR Web Token (CWT)</title><author initials="M" surname="Jones"/><author initials="E" surname="Wahlstroem"/><author initials="S" surname="Erdtman"/><author initials="H" surname="Tschofenig"/><date year="2018" month="May"/></front><seriesInfo name="RFC" value="8392"/><seriesInfo name="DOI" value="10.17487/RFC8392"/></reference>
<reference anchor="RFC9052" target="https://www.rfc-editor.org/info/rfc9052"><front><title>CBOR Object Signing and Encryption (COSE): Structures and Process</title><author initials="J" surname="Schaad"/><date year="2022" month="August"/></front><seriesInfo name="RFC" value="9052"/><seriesInfo name="DOI" value="10.17487/RFC9052"/></reference>
<reference anchor="RFC8446" target="https://www.rfc-editor.org/info/rfc8446"><front><title>The Transport Layer Security (TLS) Protocol Version 1.3</title><author initials="E" surname="Rescorla"/><date year="2018" month="August"/></front><seriesInfo name="RFC" value="8446"/><seriesInfo name="DOI" value="10.17487/RFC8446"/></reference>
<reference anchor="NIST-ZTA" target="https://doi.org/10.6028/NIST.SP.800-207"><front><title>Zero Trust Architecture (NIST SP 800-207)</title><author><organization>NIST</organization></author><date year="2020" month="August"/></front></reference>
<reference anchor="NIST-ABAC" target="https://doi.org/10.6028/NIST.SP.800-162"><front><title>Guide to Attribute Based Access Control (ABAC) Definition and Considerations (NIST SP 800-162)</title><author><organization>NIST</organization></author><date year="2014" month="January"/></front></reference>
<reference anchor="OUTBOX" target="https://microservices.io/patterns/data/transactional-outbox.html"><front><title>Pattern: Transactional outbox</title><author><organization>microservices.io</organization></author><date year="2018"/></front></reference>
<reference anchor="TS23501" target="https://www.3gpp.org/ftp/Specs/archive/23_series/23.501/"><front><title>System architecture for the 5G System (5GS) (3GPP TS 23.501)</title><author><organization>3GPP</organization></author><date year="2026"/></front></reference>
<reference anchor="ORAN-ARCH" target="https://www.o-ran.org/specifications"><front><title>O-RAN Architecture Description</title><author><organization>O-RAN Alliance</organization></author><date year="2026"/></front></reference>
<reference anchor="ISO20022" target="https://www.iso20022.org/"><front><title>ISO 20022 Financial services - Universal financial industry message scheme</title><author><organization>ISO</organization></author><date year="2026"/></front></reference>
<reference anchor="EU-AIA" target="https://eur-lex.europa.eu/eli/reg/2024/1689/oj"><front><title>Regulation (EU) 2024/1689 (Artificial Intelligence Act)</title><author><organization>European Union</organization></author><date year="2024"/></front></reference>
<reference anchor="EU-NIS2" target="https://eur-lex.europa.eu/eli/dir/2022/2555/oj"><front><title>Directive (EU) 2022/2555 (NIS2)</title><author><organization>European Union</organization></author><date year="2022"/></front></reference>
<reference anchor="EU-DORA" target="https://eur-lex.europa.eu/eli/reg/2022/2554/oj"><front><title>Regulation (EU) 2022/2554 (Digital Operational Resilience Act)</title><author><organization>European Union</organization></author><date year="2022"/></front></reference>
<reference anchor="EU-CRA" target="https://eur-lex.europa.eu/eli/reg/2024/2847/oj"><front><title>Regulation (EU) 2024/2847 (Cyber Resilience Act)</title><author><organization>European Union</organization></author><date year="2024"/></front></reference>
<reference anchor="ECB-DE" target="https://www.ecb.europa.eu/euro/digital_euro/html/index.en.html"><front><title>Digital euro</title><author><organization>European Central Bank</organization></author><date year="2026"/></front></reference>
<reference anchor="ZENODO-EU" target="https://doi.org/10.5281/zenodo.24274259"><front><title>EUROPE'S NEXT AI SECURITY CHALLENGE: WHO CONTROLS THE FINAL ACTION? From EU AI Act and NIS2 to 6G, the Digital Euro, Industrial AI and Critical Infrastructure</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026" month="November" day="16"/></front><seriesInfo name="DOI" value="10.5281/zenodo.24274259"/><refcontent>Zenodo record https://zenodo.org/records/24274259</refcontent></reference>
<reference anchor="PCT-055615" target="https://patentscope.wipo.int/"><front><title>Systems and Methods for Protected Execution Finality (PCT/IB2026/055615; WO 2026/150382 A2)</title><author fullname="Sangam Kumar Das" initials="S." surname="Das"/><date year="2026" month="July" day="16"/></front><refcontent>Filed 4 June 2026; published 16 July 2026. Verify the exact title and text in WIPO PATENTSCOPE.</refcontent></reference>
<reference anchor="IN-136555" target="https://ipindia.gov.in/"><front><title>Indian Provisional Patent Application 202631136555</title><author fullname="Sangam Kumar Das" initials="S." surname="Das"/><date year="2026" month="November" day="15"/></front><refcontent>As reported by the applicant. Not a certified registry record.</refcontent></reference>
<reference anchor="I-D.das-receipt-gated-iev" target="https://datatracker.ietf.org/doc/draft-das-receipt-gated-iev/"><front><title>Evidence Is the Key, Not the Log: Receipt-Gated Staged Effectuation and Interim Effectuation Validation for AI Agents, Frontier AI Model Providers, and Autonomous Systems</title><author fullname="Sangam Das" initials="S." surname="Das"><organization>Independent Inventor</organization></author><date year="2026" month="October"/></front><seriesInfo name="Internet-Draft" value="draft-das-receipt-gated-iev"/></reference>
<reference anchor="I-D.das-execution-finality-foundations" target="https://datatracker.ietf.org/doc/draft-das-execution-finality-foundations/"><front><title>draft-das-execution-finality-foundations (full title not verified by the editor)</title><author fullname="Sangam Das" initials="S." surname="Das"><organization>Independent Inventor</organization></author><date year="2026"/></front><seriesInfo name="Internet-Draft" value="draft-das-execution-finality-foundations"/></reference>
<reference anchor="I-D.das-agent-execution-finality" target="https://datatracker.ietf.org/doc/draft-das-agent-execution-finality/"><front><title>draft-das-agent-execution-finality (full title not verified by the editor)</title><author fullname="Sangam Das" initials="S." surname="Das"><organization>Independent Inventor</organization></author><date year="2026"/></front><seriesInfo name="Internet-Draft" value="draft-das-agent-execution-finality"/></reference>
<reference anchor="I-D.das-interoperable-consequence-authority" target="https://datatracker.ietf.org/doc/draft-das-interoperable-consequence-authority/"><front><title>Secure AI Execution-Finality for iOS, Android, and Cross-Platform AI Agents, Part 1: Foundational Consequential Authority and Common Workflows</title><author fullname="Sangam Das" initials="S." surname="Das"><organization>Independent Inventor</organization></author><date year="2026" month="October"/></front><seriesInfo name="Internet-Draft" value="draft-das-interoperable-consequence-authority"/></reference>
<reference anchor="I-D.das-finality-sink-enforcement" target="https://datatracker.ietf.org/doc/draft-das-finality-sink-enforcement/"><front><title>Secure AI Execution-Finality for iOS, Android, and Cross-Platform AI Agents, Part 4: Finality Sink Enforcement, Actual-Operation Reconstruction, and Anti-Bypass Security</title><author fullname="Sangam Das" initials="S." surname="Das"><organization>Independent Inventor</organization></author><date year="2026" month="October"/></front><seriesInfo name="Internet-Draft" value="draft-das-finality-sink-enforcement"/></reference>
<reference anchor="I-D.das-causal-effect-receipts" target="https://datatracker.ietf.org/doc/draft-das-causal-effect-receipts/"><front><title>Secure AI Execution-Finality for iOS, Android, and Cross-Platform AI Agents, Part 6: Verified Actual-Effect Receipts and Protected Causal Continuation</title><author fullname="Sangam Das" initials="S." surname="Das"><organization>Independent Inventor</organization></author><date year="2026" month="October"/></front><seriesInfo name="Internet-Draft" value="draft-das-causal-effect-receipts"/></reference>
<reference anchor="I-D.das-ef-registries" target="https://datatracker.ietf.org/doc/draft-das-ef-registries/"><front><title>Illustrative Codes Are Not a Namespace: Registries for Execution-Finality Consequence Classes, Sink Types, Act Types, Failure Codes, Profiles, and Media Types</title><author fullname="Sangam Das" initials="S." surname="Das"><organization>Independent Inventor</organization></author><date year="2026"/></front><seriesInfo name="Internet-Draft" value="draft-das-ef-registries"/></reference>
<reference anchor="TDC-12542" target="https://www.tdcommons.org/dpubs_series/12542/"><front><title>Agent Proposes, Boundary Decides (prior-disclosure research challenge)</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Technical Disclosure Commons. Title as listed in the applicant index; verify.</refcontent></reference>
<reference anchor="TDC-12667" target="https://www.tdcommons.org/dpubs_series/12667/"><front><title>Beyond HTTPS and OAuth: action-level execution authorization</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Technical Disclosure Commons. Title as listed in the applicant index; verify.</refcontent></reference>
<reference anchor="TDC-12668" target="https://www.tdcommons.org/dpubs_series/12668/"><front><title>TLS Protects the Channel, Not the Consequence</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Technical Disclosure Commons. Title as listed in the applicant index; verify.</refcontent></reference>
<reference anchor="TDC-12853" target="https://www.tdcommons.org/dpubs_series/12853/"><front><title>Jurisdiction-bound execution finality for sovereign infrastructure</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Technical Disclosure Commons. Title as listed in the applicant index; verify.</refcontent></reference>
<reference anchor="TDC-12931" target="https://www.tdcommons.org/dpubs_series/12931/"><front><title>AI safety for on-device assistants and European interoperability</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Technical Disclosure Commons. Title as listed in the applicant index; verify.</refcontent></reference>
<reference anchor="TDC-12961" target="https://www.tdcommons.org/dpubs_series/12961/"><front><title>6G / AI-RAN / protected network-effect execution</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Technical Disclosure Commons. Title as listed in the applicant index; verify.</refcontent></reference>
<reference anchor="ZEN-21861642" target="https://zenodo.org/records/21861642"><front><title>Execution-finality technical architecture: computation is not authority</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Zenodo. Title as listed in the applicant index; verify.</refcontent></reference>
<reference anchor="ZEN-22082995" target="https://zenodo.org/records/22082995"><front><title>The Internet Solved Communication. It Never Solved Authority</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Zenodo. Title as listed in the applicant index; verify.</refcontent></reference>
<reference anchor="ZEN-23040181" target="https://zenodo.org/records/23040181"><front><title>Earlier DAS disclosures and Meta Muse/Sentinel: independent comparison guide</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Zenodo. Title as listed in the applicant index; verify.</refcontent></reference>
<reference anchor="ZEN-23657985" target="https://zenodo.org/records/23657985"><front><title>Cross-platform iOS/Android AI-agent execution-finality specification</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Zenodo. Title as listed in the applicant index; verify.</refcontent></reference>
<reference anchor="IETF-IPR-7610" target="https://datatracker.ietf.org/ipr/7610/"><front><title>IETF IPR Disclosure 7610 (as cited in the applicant index)</title><author><organization>IETF</organization></author><date year="2026"/></front></reference>
<reference anchor="I-D.das-execution-finality-protocol-layer" target="https://datatracker.ietf.org/doc/draft-das-execution-finality-protocol-layer/"><front><title>The Missing Protocol Layer for the Agentic Internet: Computation Is Not Authority</title><author fullname="Sangam Das" initials="S." surname="Das"><organization>Independent Inventor</organization></author><date year="2026" month="September"/></front><seriesInfo name="Internet-Draft" value="draft-das-execution-finality-protocol-layer"/></reference>
</references>
<section anchor="appendix-emb">
<name>Embodiment Profiles E1 to E38 and E12A</name>
<t>This appendix is non-normative. It uses the claim mapping of the Indian provisional's consolidated claim section as supplied to the editor. For each profile it gives the technical objective, the distinguishing enforcement predicate (taken from the 'wherein' clause of the independent claim), the effectuation boundary, primary failure modes, the claim group, interoperability relevance and a classification. The classification (N narrows an earlier principle; C combines earlier mechanisms; X appears to introduce additional subject matter requiring independent examination) is an editorial reading of the applicant's comparison and is not verified against the mothership text. No independent novelty is asserted merely because a profile contains newly named components. Claim numbers must be verified against the official filed copy: claims 274 and 275 are absent and claim 273 is truncated in the supplied PDF.</t>
<table anchor="tbl-emb">
<name>Embodiment and claim mapping (39 profiles)</name>
<thead><tr><th>ID</th><th>Claims</th><th>Subject</th><th>Class</th></tr></thead>
<tbody>
<tr><td>E1</td><td>1-11</td><td>Protected cryptographic key-lifecycle finality</td><td>HSM and KMS command admission; PKCS#11/KMIP-style management interfaces; key-state fencing.</td></tr>
<tr><td>E2</td><td>12-22</td><td>AI-native radio-network state authorization</td><td>O-RAN SMO/Non-RT RIC policy admission; O1/A1-style configuration paths; vendor automation platforms.</td></tr>
<tr><td>E3</td><td>23-33</td><td>Declarative reconciliation and non-resurrection</td><td>Kubernetes controllers, GitOps revision pinning, admission control, lease/epoch fencing.</td></tr>
<tr><td>E4</td><td>34-44</td><td>Causal downstream-effect closure</td><td>Event meshes, workflow engines, message brokers, saga orchestrators; ERP and CRM event consumers.</td></tr>
<tr><td>E5</td><td>45-55</td><td>Accelerator execution-authority cutover</td><td>Distributed training/inference fleets; GPU partitioning; checkpoint promotion; fencing-token style ownership epochs.</td></tr>
<tr><td>E6</td><td>56-66</td><td>Trust-domain transition and prior-access closure</td><td>Arm CCA/RME with SMMU, Intel TDX, AMD SEV-SNP, SoC domain controllers.</td></tr>
<tr><td>E7</td><td>67-77</td><td>Cumulative connectivity trajectory authorization</td><td>Device connectivity managers; multi-access steering policies; mobility hysteresis and dwell-time management.</td></tr>
<tr><td>E8</td><td>78-88</td><td>Consequence-specific numerical-error reliability</td><td>Mixed-precision and quantized inference stacks; safety-relevant decision gates; accelerator runtimes.</td></tr>
<tr><td>E9</td><td>89-99</td><td>Remediation and current attestation correspondence</td><td>RATS verifier/relying-party roles (RFC 9334); confidential-computing attestation services.</td></tr>
<tr><td>E10</td><td>100-110</td><td>Software-to-hardware offload effect equivalence</td><td>SmartNIC/DPU/IPU offload; P4 pipelines; flow-rule tables; programmable switch fast paths.</td></tr>
<tr><td>E11</td><td>111-121</td><td>Multimodal audience and presentation exposure</td><td>Device output routers; casting and screen-sharing stacks; notification delivery.</td></tr>
<tr><td>E12A</td><td>122-132</td><td>Negative-observation and sensor-coverage evidence</td><td>Home/building automation hubs; physical-security and ambient-sensing platforms.</td></tr>
<tr><td>E12</td><td>133-143</td><td>In-flight media release and route changes</td><td>Streaming and notification queues; casting revocation; secure audio/video output.</td></tr>
<tr><td>E13</td><td>144-154</td><td>Low-power intent escrow and wake-epoch continuity</td><td>Sensor hubs and always-on coprocessors; mobile and embedded SoC power domains.</td></tr>
<tr><td>E14</td><td>155-165</td><td>Uncertain execution outcomes and retry safety</td><td>Idempotency keys, durable workflows, outbox reconciliation, payment status inquiry.</td></tr>
<tr><td>E15</td><td>166-176</td><td>Authority continuity through agent context compaction</td><td>Agent SDKs and runtimes with tool dispatch, memory and handoff features.</td></tr>
<tr><td>E16</td><td>177-187</td><td>Independent verification of autonomous software repairs</td><td>CI/CD pipelines, artifact signing, provenance attestations, protected branches.</td></tr>
<tr><td>E17</td><td>188-198</td><td>Human delegation and minimal-disclosure authority</td><td>Service-desk, approval and case-management workflows with human escalation.</td></tr>
<tr><td>E18</td><td>199-209</td><td>Spoken intent and interruption-epoch validation</td><td>Voice assistants, realtime voice agents, in-car and smart-home voice control.</td></tr>
<tr><td>E19</td><td>210-220</td><td>Cross-source factual evidence consistency</td><td>Retrieval-augmented systems; enterprise search; temporal and snapshot data stores.</td></tr>
<tr><td>E20</td><td>221-231</td><td>Cross-protocol tool-semantic authority</td><td>MCP, vendor agent APIs, agent-to-agent protocols, tool gateways.</td></tr>
<tr><td>E21</td><td>232-242</td><td>Aggregate multi-agent work-product finality</td><td>Multi-agent orchestration; collaborative revision and merge systems.</td></tr>
<tr><td>E22</td><td>243-253</td><td>Document-field spatial provenance</td><td>Document AI, invoice processing, accounts-payable automation.</td></tr>
<tr><td>E23</td><td>254-264</td><td>Visual-target and persistent referent continuity</td><td>Multimodal assistants, robotics, spatial computing, GUI-automation agents.</td></tr>
<tr><td>E24</td><td>265-273*</td><td>Cross-channel communication-origin authority</td><td>Enterprise messaging and collaboration channels, business messaging APIs.</td></tr>
<tr><td>E25</td><td>276-286</td><td>Governance enforcement-coverage reconciliation</td><td>Agent-governance platforms, API/credential inventories, effect-route test harnesses.</td></tr>
<tr><td>E26</td><td>287-297</td><td>Late-bound ERP and accounting effects</td><td>ERP and accounting systems, procure-to-pay, financial close.</td></tr>
<tr><td>E27</td><td>298-308</td><td>Customer commitments versus actual fulfillment</td><td>Customer-service agents, case-management and order-fulfillment systems.</td></tr>
<tr><td>E28</td><td>309-319</td><td>Live-topology IT remediation</td><td>IT operations management, CMDB, cloud change and remediation tooling.</td></tr>
<tr><td>E29</td><td>320-330</td><td>Analytical cohort identity and activation</td><td>Data platforms with activation to marketing, CRM and account actions.</td></tr>
<tr><td>E30</td><td>331-341</td><td>Cross-engine data-policy semantics</td><td>Lakehouse and warehouse catalogs, data-sharing and multi-engine query paths.</td></tr>
<tr><td>E31</td><td>342-352</td><td>Transitive database stored-program effects</td><td>Relational databases with stored programs, triggers and external-call features.</td></tr>
<tr><td>E32</td><td>353-363</td><td>Source-commit-certified downstream effects</td><td>Transactional outbox, change-data-capture, brokers and event buses.</td></tr>
<tr><td>E33</td><td>364-374</td><td>Agentic payment exposure reservation</td><td>Tokenized agent payments, issuer and gateway limit controls, PSP APIs.</td></tr>
<tr><td>E34</td><td>375-385</td><td>Cross-border FX and beneficiary outcomes</td><td>Cross-border payment messaging and tracking; remittance networks.</td></tr>
<tr><td>E35</td><td>386-396</td><td>Offline digital cash post-compromise fork evidence</td><td>Offline digital-currency research and wallets with secure elements.</td></tr>
<tr><td>E36</td><td>397-407</td><td>Edge subrequest and outbound-effect authority</td><td>Edge and serverless runtimes, queues, service bindings, egress gateways.</td></tr>
<tr><td>E37</td><td>408-418</td><td>Causal radio-service impairment control</td><td>RAN automation and rollout management, canary and closed-loop optimization.</td></tr>
<tr><td>E38</td><td>419-429</td><td>6G network-slice capacity lending and reclamation</td><td>Network-slice orchestration and RAN slice controllers.</td></tr>
</tbody>
</table>
<section anchor="emb-e1">
<name>E1 (Claims 1-11): Protected cryptographic key-lifecycle finality</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Gate key creation, activation, rotation, derivation, export, use and destruction on the effective key/partition transition, not on the requesting credential.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Authorization of the originating request, or possession of administrative credentials, does not authorize a materially different or no-longer-authorized lifecycle transition.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>HSM/KMS command-admission point; key-state transition inside the cryptographic module.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Residual key-use authority after rotation or destruction; derived keys outliving a revoked parent; replayed administrative command; split-brain partition state.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 1-11 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>HSM and KMS command admission; PKCS#11/KMIP-style management interfaces; key-state fencing.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e2">
<name>E2 (Claims 12-22): AI-native radio-network state authorization</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Reconstruct the aggregate multi-cell, neighbor and slice state an AI-generated radio configuration would produce and authorize that effective state.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Acceptance of a controller instruction does not authorize a different effective network configuration or an aggregate resource effect exceeding the validated scope.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Configuration-effect point on the management plane (SMO/Non-RT RIC to network element); not PHY/MAC loops.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Neighbor-cell side effects; slice footprint underestimated; stale policy epoch; optimizer request differs from installed parameters.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 12-22 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>O-RAN SMO/Non-RT RIC policy admission; O1/A1-style configuration paths; vendor automation platforms.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e3">
<name>E3 (Claims 23-33): Declarative reconciliation and non-resurrection</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Bind reconciliation authority to a current generation so a controller cannot reinstall a revoked configuration.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Persistence of a desired-state declaration, repetition of a reconciliation loop, or restoration of a controller does not recreate authority after the generation authority is retired.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Reconciliation-admission step before the control-plane change is committed.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Obsolete desired-state snapshot reapplied after rollback; restored controller using retired generation; GitOps revision drift.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 23-33 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Kubernetes controllers, GitOps revision pinning, admission control, lease/epoch fencing.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e4">
<name>E4 (Claims 34-44): Causal downstream-effect closure</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Materialize the causal tree of child effects (workflows, queues, APIs, stored programs) and require current operation-specific authority at each externally effective child.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Authorization of the Root Candidate Act does not confer unrestricted authority on downstream operations generated by causal expansion.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Each downstream sink that can make a derived effect externally effective.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Unmodelled child branch; confused-deputy child using broader credentials; child outliving revocation of root authority.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 34-44 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Event meshes, workflow engines, message brokers, saga orchestrators; ERP and CRM event consumers.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class C: this profile combines earlier mechanisms into an additional implementation arrangement.</t></dd>
</dl>
</section>
<section anchor="emb-e5">
<name>E5 (Claims 45-55): Accelerator execution-authority cutover</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Separate checkpoint possession from authority to publish results during migration or failover; single effect-commit owner via source fence, destination admission and effect watermark.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Possession or restoration of computational state does not confer effect-commit authority; source and destination are prevented from committing conflicting effects.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Protected output publish or model-state commit interface (not every tensor operation).</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Zombie source worker publishing after failover; destination double-commit; stale checkpoint promoted; watermark regression.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 45-55 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Distributed training/inference fleets; GPU partitioning; checkpoint promotion; fencing-token style ownership epochs.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class C: this profile combines earlier mechanisms into an additional implementation arrangement.</t></dd>
</dl>
</section>
<section anchor="emb-e6">
<name>E6 (Claims 56-66): Trust-domain transition and prior-access closure</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Prove old-domain DMA, interrupt, debug and queued-I/O paths are closed before successor authority for a resource is granted.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>A declared ownership change, permission update or domain reassignment does not establish completion while a prior-authority path can still produce a prohibited effect.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Resource-assignment controller and protected DMA-mapping/transition interlock.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Residual IOMMU/SMMU translation; in-flight DMA; interrupt remap not retargeted; debug path left enabled; queued I/O completing in new domain.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 56-66 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Arm CCA/RME with SMMU, Intel TDX, AMD SEV-SNP, SoC domain controllers.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e7">
<name>E7 (Claims 67-77): Cumulative connectivity trajectory authorization</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Evaluate the history of access transitions (Wi-Fi, cellular, satellite) against a cumulative temporal envelope rather than each switch alone.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Independent authorization of individual connectivity operations does not authorize a sequence whose cumulative governed effects exceed the temporal envelope.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Terminal connectivity policy controller, outside high-frequency radio loops.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Oscillating switches each individually permitted; history record tampered or reset; service-continuity obligation violated by sequence.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 67-77 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Device connectivity managers; multi-access steering policies; mobility hysteresis and dwell-time management.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class X: this profile appears to introduce additional technical subject matter requiring independent prior-art and priority-support examination.</t></dd>
</dl>
</section>
<section anchor="emb-e8">
<name>E8 (Claims 78-88): Consequence-specific numerical-error reliability</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Compose quantization, rounding and accumulation error with the decision margin of a specific consequential decision; permit, request higher precision, or block.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Approval of a model, processor, precision or configuration does not authorize a consequential operation when the effective numerical result fails, or cannot be shown to satisfy, its consequence-specific conditions.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Release of a numerical decision to a consequential downstream consumer.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Error bound underestimated; precision downgrade between approval and use; margin computed on a different model build.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 78-88 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Mixed-precision and quantized inference stacks; safety-relevant decision gates; accelerator runtimes.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class X: this profile appears to introduce additional technical subject matter requiring independent prior-art and priority-support examination.</t></dd>
</dl>
</section>
<section anchor="emb-e9">
<name>E9 (Claims 89-99): Remediation and current attestation correspondence</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Match patch state, fresh attestation evidence, reference values and relying-party policy per consequence class, while allowing defined safe continuity.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Installation or reported activation of a remediation does not establish attestation correspondence; earlier attestation acceptance does not authorize an effect that requires a newer condition.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Attested workload admission and effect-specific trust-state fence.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Stale appraisal after patch; reference values lag the patch; grace-period abuse; policy epoch change not reflected in cached verdict.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 89-99 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>RATS verifier/relying-party roles (RFC 9334); confidential-computing attestation services.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class C: this profile combines earlier mechanisms into an additional implementation arrangement.</t></dd>
</dl>
</section>
<section anchor="emb-e10">
<name>E10 (Claims 100-110): Software-to-hardware offload effect equivalence</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Compare the effective installed hardware behavior (wildcards, priority, caches, state) with the effect permitted by the software decision; fence stale fast-path authority.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Installation of an accelerated rule, or software authorization of the corresponding operation, does not authorize repeated hardware effectuation when installed behavior exceeds the permitted scope or current authority.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Offload rule install/readback and protected fast-path enablement barrier.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Wildcard rule broader than the authorized flow; rule precedence changes effect; cached rule surviving revocation; readback differs from installed state.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 100-110 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>SmartNIC/DPU/IPU offload; P4 pipelines; flow-rule tables; programmable switch fast paths.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class X: this profile appears to introduce additional technical subject matter requiring independent prior-art and priority-support examination.</t></dd>
</dl>
</section>
<section anchor="emb-e11">
<name>E11 (Claims 111-121): Multimodal audience and presentation exposure</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Evaluate concurrent audience, modality and residual output routes before information becomes perceptible.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Authorization to generate, access, transmit or present information through one route does not authorize exposure through another route with different audience, modality, privacy or residual-output conditions.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Output routing and endpoint release (display, speaker, casting).</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Cast/mirroring route added; bystander-visible surface; residual buffer on the old route.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 111-121 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Device output routers; casting and screen-sharing stacks; notification delivery.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e12a">
<name>E12A (Claims 122-132): Negative-observation and sensor-coverage evidence</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Admit an absence-of-event inference only if sensing opportunity, health, coverage, buffering and time-gap conditions were met.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Absence of a received event is not sufficient authorization when a mandatory opportunity to observe, or evidence about that opportunity, has not been established.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Protected evidence qualification before physical, access or notification actions.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Sensor offline treated as 'no event'; coverage hole; buffered events delayed past decision; single-source completeness assumption.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 122-132 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Home/building automation hubs; physical-security and ambient-sensing platforms.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e12">
<name>E12 (Claims 133-143): In-flight media release and route changes</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Re-evaluate or fence already queued or streaming segments when audience or route changes.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Authorization of an initial presentation segment does not authorize continued release of unreleased information after a material change in audience authorization or route.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Queue/output sink with epoch-bound release.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Buffered audio or video reaching a newly unauthorized audience; notification queue flushed after privacy-state change.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 133-143 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Streaming and notification queues; casting revocation; secure audio/video output.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e13">
<name>E13 (Claims 144-154): Low-power intent escrow and wake-epoch continuity</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Hold pending intent prepared by an always-on processor and revalidate it against current identity, policy and host state at wake.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Permission for low-power observation or computation does not confer authority to execute an escrowed operation after a power, boot, session or domain change.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Wake boundary and protected intent queue.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Stale escrowed intent executed after user change or policy revocation; wake-epoch not advanced; escrow tampering.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 144-154 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Sensor hubs and always-on coprocessors; mobile and embedded SoC power domains.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e14">
<name>E14 (Claims 155-165): Uncertain execution outcomes and retry safety</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Separate logical-effect identity from attempt identity; quarantine unresolved completion; reconcile against authoritative endpoint evidence before retry or compensation.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Loss of an acknowledgment, timeout, worker failure or originator restart does not establish that an earlier operation failed or authorize an unrestricted duplicate attempt.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Protected retry controller and effect ledger at the consequential endpoint interface.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Blind retry duplicating a committed effect; compensation under stale authority; conflicting replacement operation admitted during quarantine.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 155-165 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Idempotency keys, durable workflows, outbox reconciliation, payment status inquiry.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class X: this profile appears to introduce additional technical subject matter requiring independent prior-art and priority-support examination.</t></dd>
</dl>
</section>
<section anchor="emb-e15">
<name>E15 (Claims 166-176): Authority continuity through agent context compaction</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Keep an authority record independent of the model-visible context so summarization or handoff cannot drop a prohibition.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Omission, modification or reinterpretation of authority information during model-context transformation does not remove an operative prohibition or expand authority.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Tool dispatcher or credential-isolated tool proxy.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Summary omits a 'do not' constraint; handoff widens scope; memory replacement rewrites the task; prompt-injected authority claim.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 166-176 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Agent SDKs and runtimes with tool dispatch, memory and handoff features.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e16">
<name>E16 (Claims 177-187): Independent verification of autonomous software repairs</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Isolate acceptance criteria and verification oracle from the modifying agent; bind deployed bytes to the approved, verified artifact.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>An agent's ability to modify code, generate tests or report success does not confer authority to redefine acceptance criteria or release an artifact lacking independent verification evidence.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>CI/CD artifact-promotion and deployment gate.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Agent edits its own tests; artifact swapped after verification; oracle shares agent credentials.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 177-187 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>CI/CD pipelines, artifact signing, provenance attestations, protected branches.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e17">
<name>E17 (Claims 188-198): Human delegation and minimal-disclosure authority</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Send a bounded task capsule to a human and independently validate the returned work and continuing authority before any external effect.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Transfer of task information to a human does not transfer the agent's full context, stored credentials, general privileges or unrestricted authority.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Human-in-the-loop handoff gateway and action gateway.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Over-disclosure to the human; returned artifact treated as authority; stale authority on re-entry.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 188-198 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Service-desk, approval and case-management workflows with human escalation.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e18">
<name>E18 (Claims 199-209): Spoken intent and interruption-epoch validation</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Bind the final operative spoken intent, after corrections and barge-in, to the actual tool operation through a turn/interruption epoch.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>A provisional transcript, incomplete instruction, outdated interpretation or superseded proposal does not confer authority for the corresponding effect.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Speech-to-tool commit boundary.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Early partial transcript executed; interrupted command still dispatched; ASR correction not reflected.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 199-209 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Voice assistants, realtime voice agents, in-car and smart-home voice control.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e19">
<name>E19 (Claims 210-220): Cross-source factual evidence consistency</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Identify material factual dependencies and reconcile source validity and snapshot coherence at effectuation.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Authenticity of individual retrieved records, or production of a model synthesis, does not establish a coherent and current factual basis for the actual operation.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Evidence-sensitive tool execution gate.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Mutually inconsistent authentic records; stale cache; two 'independent' sources sharing one origin.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 210-220 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Retrieval-augmented systems; enterprise search; temporal and snapshot data stores.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e20">
<name>E20 (Claims 221-231): Cross-protocol tool-semantic authority</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Preserve instruction authority and tool restrictions across model-API conversion, MCP adapters and provider-hosted versus client-executed tools.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Syntactic acceptance of an adapted request does not establish preservation of instruction authority, confidentiality or tool restrictions across protocols.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Protocol bridge plus independent tool/end-target control.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Instruction hierarchy dropped in translation; hosted tool bypasses gateway; unsupported feature silently ignored.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 221-231 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>MCP, vendor agent APIs, agent-to-agent protocols, tool gateways.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e21">
<name>E21 (Claims 232-242): Aggregate multi-agent work-product finality</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Materialize the composition of agent work products and authorize the accepted aggregate or a validated subset.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Individual authorization or local validation of work products does not authorize the consequential state produced by their combination.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Shared merge/commit boundary.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Cross-branch read/write conflict; cumulative limit exceeded by sum; negative constraint violated only after merge.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 232-242 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Multi-agent orchestration; collaborative revision and merge systems.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e22">
<name>E22 (Claims 243-253): Document-field spatial provenance</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Bind an extracted value to its source row/cell, page geometry, payee and business-effect identity before payment or record change.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Recognition of correct strings or field values does not authorize an operation when values are not sufficiently associated with the corresponding labels, records or relationships.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Extraction-to-ERP/payment finality point.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Amount bound to wrong invoice row; payee/IBAN mismatch; table geometry shift; duplicate-page confusion.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 243-253 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Document AI, invoice processing, accounts-payable automation.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e23">
<name>E23 (Claims 254-264): Visual-target and persistent referent continuity</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Verify that the entity currently targeted is the one originally referred to ('this one') before physical, financial or access effects.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>An earlier reference to a screen position, region, label, description or coordinate does not authorize execution against a different entity that later occupies it.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Perception-to-action target verification.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>UI re-layout; moving or occluded object; coordinate reuse; ambiguous referent.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 254-264 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Multimodal assistants, robotics, spatial computing, GUI-automation agents.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e24">
<name>E24 (Claims 265-273*): Cross-channel communication-origin authority</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Separate message transmission from authenticated original-speaker intent and check the current audience before sending protected content. (*Claim 273 truncated in the supplied file; claims 274-275 absent.)</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Receipt, quotation, forwarding or summarization of a message does not confer an alleged earlier sender's authority; earlier authorization to disclose to a group does not authorize disclosure to a changed audience.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Controlled send/recipient-release point.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Quoted instruction treated as command; account reassignment; group membership changed after approval.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 265-273* of the consolidated claim section. The supplied file has nine physically present claims (265-273); claim 273 ends mid-sentence.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Enterprise messaging and collaboration channels, business messaging APIs.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e25">
<name>E25 (Claims 276-286): Governance enforcement-coverage reconciliation</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Verify, route by route, that required governance controls are enforceable at actual effect paths; do not equate logging with prevention.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Registration of an agent, a nominal policy, or compliant sampled traces do not establish that the required control is enforceable across the covered execution routes.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Assurance controller with proof of sink rejection per discovered route.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Undiscovered route; credential bypass; dashboard green while a path is uncontrolled; coverage probe unsafe to run.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 276-286 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Agent-governance platforms, API/credential inventories, effect-route test harnesses.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class X: this profile appears to introduce additional technical subject matter requiring independent prior-art and priority-support examination.</t></dd>
</dl>
</section>
<section anchor="emb-e26">
<name>E26 (Claims 287-297): Late-bound ERP and accounting effects</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Recompute effective posting, tax, FX and approval consequences against live state immediately before posting.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Authorization of an initial instruction, document or simulated transaction does not authorize a later commitment with a materially different financial or operational consequence.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>ERP posting/release boundary.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Tax/FX/pricing change between approval and posting; allocation rule changed; simulated result differs from actual.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 287-297 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>ERP and accounting systems, procure-to-pay, financial close.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e27">
<name>E27 (Claims 298-308): Customer commitments versus actual fulfillment</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Gate outward statements ('refunded', 'shipped') on independently qualified fulfillment evidence, separately from initiating the internal request.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Authorization or submission of the underlying operation does not authorize a statement that it is completed or settled without supporting evidence; approval of the statement does not authorize the operation.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Customer-communication dispatch with authoritative fulfillment check.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>'Done' claimed on request acceptance; refund pending but announced; stale case status.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 298-308 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Customer-service agents, case-management and order-fulfillment systems.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e28">
<name>E28 (Claims 309-319): Live-topology IT remediation</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Reconstruct the live dependency topology at the remediation commit; stale remediation generations cannot resume.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>A previously valid change request does not authorize the same nominal command when the current target, live dependencies or material impact differs from the approved scope.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Protected change/repair controller at the commit point.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>CMDB drift; new dependency edge; criticality changed; resumed stale plan.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 309-319 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>IT operations management, CMDB, cloud change and remediation tooling.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e29">
<name>E29 (Claims 320-330): Analytical cohort identity and activation</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Reconcile the actual downstream target population, suppression and consent per target before fan-out.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Permission to run an analytical query or obtain a policy-filtered population does not authorize an operation against an expanded, duplicated, unsubscribed, suppressed or ineligible target population.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Analytics-to-activation fan-out boundary.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Join or identity-resolution expansion; suppressed identity re-added; duplicate recipient; consent withdrawn after query.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 320-330 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Data platforms with activation to marketing, CRM and account actions.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e30">
<name>E30 (Claims 331-341): Cross-engine data-policy semantics</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Test that mandatory row, column, masking, aggregation and recipient obligations survive the actual query route and result release.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Syntactic compatibility, execution by an alternative engine, or storage-level credentials do not establish preservation of governance semantics.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Query route and effective result-release validation.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Zero-copy share bypassing masking; engine-specific semantic difference; storage-layer credential used directly.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 331-341 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Lakehouse and warehouse catalogs, data-sharing and multi-engine query paths.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class X: this profile appears to introduce additional technical subject matter requiring independent prior-art and priority-support examination.</t></dd>
</dl>
</section>
<section anchor="emb-e31">
<name>E31 (Claims 342-352): Transitive database stored-program effects</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Reconstruct the closure of material effects reachable through triggers, definer-rights code, autonomous transactions and external functions.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Permission to invoke an entry-point operation or stored program does not confer authority for every material effect transitively reachable through it.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Supported database/transaction enforcement point before privileged mutation.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Trigger chain; definer-rights escalation; autonomous transaction side effect; external function call.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 342-352 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Relational databases with stored programs, triggers and external-call features.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class C: this profile combines earlier mechanisms into an additional implementation arrangement.</t></dd>
</dl>
</section>
<section anchor="emb-e32">
<name>E32 (Claims 353-363): Source-commit-certified downstream effects</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Bind downstream release to an authoritative source commit and current policy/effect state; reconcile after uncertain acknowledgments.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Commitment of the source transaction, or presence of a derived event, does not authorize an external operation produced by a downstream consumer or transformed event.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Transactional outbox/CDC and consumer effect boundary.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Uncommitted intent consumed; stale outbox row; unrelated transaction event; duplicate delivery.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 353-363 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Transactional outbox, change-data-capture, brokers and event buses.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class C: this profile combines earlier mechanisms into an additional implementation arrangement.</t></dd>
</dl>
</section>
<section anchor="emb-e33">
<name>E33 (Claims 364-374): Agentic payment exposure reservation</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Conserve aggregate reserved, authorized and captured exposure against the mandate from authorization through capture.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>A payment request cannot consume authority already reserved or committed for another incompatible obligation; unresolved exposure is classified conservatively.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Authorization-to-capture and exposure-reservation boundary.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Concurrent agents exceeding mandate; double-counted representations of one obligation; unresolved exposure treated as released.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 364-374 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Tokenized agent payments, issuer and gateway limit controls, PSP APIs.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e34">
<name>E34 (Claims 375-385): Cross-border FX and beneficiary outcomes</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Distinguish instruction acceptance, settlement stages and independently evidenced beneficiary-net outcome.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Authorization of a nominal originating amount does not authorize an intermediary transformation or final beneficiary-side consequence violating the mandate; a message acknowledgment does not establish beneficiary credit.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Beneficiary-outcome and exception/reconciliation boundary.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>FX or fee drift; intermediary deductions; ack mistaken for credit; routing change.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 375-385 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Cross-border payment messaging and tracking; remittance networks.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e35">
<name>E35 (Claims 386-396): Offline digital cash post-compromise fork evidence</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Preserve delegated spending-authority continuity and reconcile conflicting transfer evidence after protection assumptions fail.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Prevention of unauthorized transfers under intact hardware does not establish that cloning remains impossible after compromise; detection of conflicting evidence does not establish the full amount of undiscovered conflicting transfers.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Offline wallet effectuation point plus later reconciliation.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Secure-element compromise; cloned state; incomplete fork discovery; privacy loss in reconciliation.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 386-396 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Offline digital-currency research and wallets with secure elements.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class X: this profile appears to introduce additional technical subject matter requiring independent prior-art and priority-support examination.</t></dd>
</dl>
</section>
<section anchor="emb-e36">
<name>E36 (Claims 397-407): Edge subrequest and outbound-effect authority</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Preserve parent source/purpose restrictions across subrequests, queues, retries and background workers; enforce at outbound effect paths.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Authentication or authorization of an originating edge request does not authorize each transitive, deferred, cached or independently executed effect the workload generates.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Edge runtime egress and credential-isolated service connector.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Background worker using broader credentials; queued task outliving revocation; cached result reused.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 397-407 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Edge and serverless runtimes, queues, service bindings, egress gateways.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class N: this profile narrows an earlier architectural principle.</t></dd>
</dl>
</section>
<section anchor="emb-e37">
<name>E37 (Claims 408-418): Causal radio-service impairment control</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Correlate attributable impact evidence of a limited real RAN change and expand only within bounded impairment.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>An individually authorized radio-control action is not automatically granted unrestricted continued or expanded deployment because its parameters were valid or telemetry has not yet shown degradation.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Non-real-time RAN rollout gate with bounded canary feedback.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Local metrics miss neighbor-cell harm; confounded attribution; premature expansion.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 408-418 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>RAN automation and rollout management, canary and closed-loop optimization.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class C: this profile combines earlier mechanisms into an additional implementation arrangement.</t></dd>
</dl>
</section>
<section anchor="emb-e38">
<name>E38 (Claims 419-429): 6G network-slice capacity lending and reclamation</name>
<dl newline="true">
<dt>Technical objective</dt>
<dd><t>Account for loaned resources, continuity and enforceable reclaim before activation and further borrowing.</t></dd>
<dt>Distinguishing enforcement predicate</dt>
<dd><t>Temporary non-utilization of a first slice's resource does not establish that it may be lent unless entitlement, conservation and deadline-bounded reclamation remain satisfiable.</t></dd>
<dt>Effectuation boundary</dt>
<dd><t>Slice orchestration and capacity-admission controls; protected scheduler lease state.</t></dd>
<dt>Primary failure modes</dt>
<dd><t>Unreclaimable loan after mobility or reroute; entitlement floor violated; capacity created by lending.</t></dd>
<dt>Associated claim group</dt>
<dd><t>Claims 419-429 of the consolidated claim section.</t></dd>
<dt>Interoperability relevance</dt>
<dd><t>Network-slice orchestration and RAN slice controllers.</t></dd>
<dt>Relation to the June mothership (unverified)</dt>
<dd><t>Editorial class C: this profile combines earlier mechanisms into an additional implementation arrangement.</t></dd>
</dl>
</section>
</section>
<section anchor="appendix-index">
<name>Technical Terminology, Synonym Mapping, and Patent Examiner Search-Phrase Index</name>
<t>This appendix is non-normative. It lists the inventor's terminology and conventional industry equivalents, keyword clusters mapped to sections and embodiments, substantive search phrases and Boolean strings. The index does not imply that any search result anticipates, or is infringed by, any patent claim, and index terms are not claims or verified classification codes.</t>
<section anchor="idx-syn">
<name>Terminology and Synonym Mapping</name>
<table anchor="tbl-syn">
<name>Terminology mapping</name>
<thead><tr><th>Inventor terminology</th><th>Industry search terminology</th><th>Embodiments</th></tr></thead>
<tbody>
<tr><td>Candidate Act</td><td>Proposed operation; deferred operation; staged transaction; pending request</td><td>1-38</td></tr>
<tr><td>Non-Effective State</td><td>Pending non-committed state; staged/uncommitted state</td><td>1-38</td></tr>
<tr><td>Actual Pending Operation</td><td>Reconstructed effective operation; resolved operation; effective-operation derivation</td><td>1-38</td></tr>
<tr><td>Operation Reconstruction</td><td>Effective-operation reconstruction; action canonicalization; effect derivation</td><td>1,2,4,10,31</td></tr>
<tr><td>Finality Sink</td><td>Protected effectuation boundary; execution enforcement point; policy enforcement point at the effect</td><td>1-38</td></tr>
<tr><td>Finality Gate</td><td>Commit gate; release gate; admission gate</td><td>1-38</td></tr>
<tr><td>Source-Authority Envelope</td><td>Originating authorization context; delegated authority envelope</td><td>4,15,17,20,24,36</td></tr>
<tr><td>Protected Enforcement Domain</td><td>Trusted enforcement environment; protected policy domain</td><td>1-38</td></tr>
<tr><td>Independent Validator</td><td>Runtime policy revalidation service; out-of-band authorization verifier</td><td>1-38</td></tr>
<tr><td>Scoped Execution Capability</td><td>Operation-bound execution authorization; non-bearer capability; sender-constrained, act-bound token</td><td>1-38</td></tr>
<tr><td>Consequence-Bound Authorization</td><td>Effect-aware authorization; consequence-specific authorization</td><td>1-38</td></tr>
<tr><td>Authority Continuity</td><td>Delegated authority preservation; authority propagation integrity</td><td>4,15,17,20,36</td></tr>
<tr><td>Protected Policy Epoch</td><td>Policy version fence; policy generation counter</td><td>3,10,25</td></tr>
<tr><td>Revocation Epoch</td><td>Revocation generation; revocation freshness marker</td><td>1,3,9</td></tr>
<tr><td>Evidence-Carrying Receipt</td><td>Authenticated effect evidence; signed execution receipt</td><td>14,27,32,34</td></tr>
<tr><td>Effect Confirmation Receipt</td><td>Authenticated destination effect evidence; destination-origin confirmation</td><td>14,27,34</td></tr>
<tr><td>Interim Effectuation Validation (IEV)</td><td>Receipt-gated continuation validation; staged-effect checkpoint</td><td>14,33</td></tr>
<tr><td>Ledger-Anchored Validation Receipt</td><td>Commit-log-anchored attestation record; transparency-log-anchored receipt</td><td>32,35</td></tr>
<tr><td>Effective Operation Descriptor / Effective Hardware Action Descriptor</td><td>Installed rule semantics; match-action effective behavior</td><td>10</td></tr>
<tr><td>Material Dependency Graph</td><td>Live dependency topology; causal effect graph; transitive effect closure</td><td>4,28,31</td></tr>
<tr><td>Protected State Commitment</td><td>Sealed state digest; authenticated state commitment</td><td>5,9,14</td></tr>
<tr><td>Authoritative Outcome Evidence</td><td>Destination-of-record confirmation; system-of-record outcome</td><td>14,27,34</td></tr>
<tr><td>Stale Authority</td><td>Expired/superseded authority; time-of-check/time-of-use gap</td><td>3,5,13,28</td></tr>
<tr><td>Transitive Effect</td><td>Indirect side effect; derived effect; second-order effect</td><td>4,31,36</td></tr>
<tr><td>Aggregate Consequence</td><td>Cross-agent combined-state effect; cumulative effect</td><td>7,21,33</td></tr>
<tr><td>Bounded Real Effect</td><td>Canary effect; limited trial effect; staged partial effectuation</td><td>37,Receipt-gated IEV</td></tr>
<tr><td>Effectuation-Boundary Enforcement</td><td>Enforcement at the point of effect; last-mile enforcement</td><td>1-38</td></tr>
<tr><td>Control-to-Effect Route Coverage</td><td>Verified enforcement-path completeness; governance coverage</td><td>25</td></tr>
<tr><td>Protected Offload Equivalence</td><td>Software-to-hardware policy correspondence; fast-path equivalence</td><td>10</td></tr>
<tr><td>Aggregate Effect Validation</td><td>Cross-agent combined-state authorization; merge-time policy validation</td><td>21</td></tr>
<tr><td>Execution Finality</td><td>Protected consequential operation commitment; point-of-no-return control</td><td>1-38</td></tr>
</tbody>
</table>
</section>
<section anchor="idx-kw">
<name>Keyword Clusters</name>
<t>140 cluster keywords plus 40 additional keywords and 31 inventor terms listed in the synonym table.</t>
<section>
<name>AI-agent security</name>
<t>Relevant sections and embodiments: Sections 4, 6 (REQ-02, REQ-14, REQ-15), 8.3; Embodiments E15-E21, E24, E36.</t>
<ul>
<li>agentic AI</li>
<li>autonomous execution</li>
<li>tool-call authorization</li>
<li>MCP security</li>
<li>agent delegation</li>
<li>prompt injection</li>
<li>instruction hierarchy</li>
<li>memory compaction</li>
<li>context summarization</li>
<li>cross-model protocol translation</li>
<li>provider-hosted tools</li>
<li>client-executed tools</li>
<li>AI coding agents</li>
<li>agent handoff</li>
<li>confused deputy</li>
<li>credential isolation</li>
<li>tool dispatcher</li>
<li>agent governance</li>
<li>human-in-the-loop</li>
<li>verification oracle</li>
</ul>
</section>
<section>
<name>Cryptographic and trusted computing</name>
<t>Relevant sections and embodiments: Sections 6 (REQ-04 to REQ-06, REQ-10), 8.4; E1, E5, E6, E9.</t>
<ul>
<li>hardware-rooted authorization</li>
<li>non-bearer capability</li>
<li>sender-constrained token</li>
<li>secure execution environment</li>
<li>trusted execution environment</li>
<li>runtime attestation</li>
<li>secure element</li>
<li>hardware security module</li>
<li>key lifecycle</li>
<li>anti-replay</li>
<li>freshness</li>
<li>protected epoch</li>
<li>confidential computing</li>
<li>remote attestation</li>
<li>fencing token</li>
<li>generation counter</li>
<li>key derivation</li>
<li>key destruction</li>
<li>partition state</li>
<li>attestation verifier</li>
</ul>
</section>
<section>
<name>Network and telecommunications</name>
<t>Relevant sections and embodiments: Section 8.7; E2, E7, E37, E38.</t>
<ul>
<li>AI-RAN</li>
<li>O-RAN</li>
<li>RAN Intelligent Controller</li>
<li>rApp</li>
<li>Service Management and Orchestration</li>
<li>6G network slicing</li>
<li>radio impairment</li>
<li>cumulative connectivity transition</li>
<li>slice entitlement</li>
<li>temporary capacity lending</li>
<li>slice reclamation</li>
<li>spectrum consequence</li>
<li>canary rollout</li>
<li>neighbor-cell impact</li>
<li>access steering</li>
<li>multi-access transition</li>
<li>non-real-time control</li>
<li>network-impact policy</li>
<li>configuration drift</li>
<li>service continuity</li>
</ul>
</section>
<section>
<name>Enterprise and database</name>
<t>Relevant sections and embodiments: Sections 8.5, 8.6; E4, E22, E26, E28, E31, E32.</t>
<ul>
<li>transitive database effects</li>
<li>trigger execution</li>
<li>stored-program authorization</li>
<li>definer rights</li>
<li>autonomous transaction</li>
<li>event provenance</li>
<li>transactional outbox</li>
<li>source-commit validation</li>
<li>change data capture</li>
<li>ERP configuration drift</li>
<li>account effect vector</li>
<li>posting reconciliation</li>
<li>purchase-order approval</li>
<li>tax and FX recomputation</li>
<li>document field provenance</li>
<li>configuration management database</li>
<li>live dependency topology</li>
<li>change control</li>
<li>remediation generation</li>
<li>event consumer authorization</li>
</ul>
</section>
<section>
<name>Data governance</name>
<t>Relevant sections and embodiments: Appendix A (E29, E30).</t>
<ul>
<li>row-level security</li>
<li>column masking</li>
<li>heterogeneous query engines</li>
<li>policy semantic equivalence</li>
<li>effective cohort membership</li>
<li>identity resolution</li>
<li>data egress</li>
<li>population activation</li>
<li>suppression list</li>
<li>consent state</li>
<li>zero-copy sharing</li>
<li>catalog policy</li>
<li>aggregation threshold</li>
<li>result-release validation</li>
<li>recipient obligation</li>
<li>data clean room</li>
<li>lineage</li>
<li>policy pushdown</li>
<li>fan-out authorization</li>
<li>opt-out propagation</li>
</ul>
</section>
<section>
<name>Finance</name>
<t>Relevant sections and embodiments: Section 8.8; E14, E33, E34, E35.</t>
<ul>
<li>agentic payments</li>
<li>cumulative authorization exposure</li>
<li>payment reservation</li>
<li>authorization-to-capture</li>
<li>FX beneficiary finality</li>
<li>beneficiary-net amount</li>
<li>offline digital money</li>
<li>offline digital cash</li>
<li>token fork</li>
<li>post-compromise recovery</li>
<li>financial loss reconciliation</li>
<li>double-spend</li>
<li>exposure ledger</li>
<li>payment mandate</li>
<li>settlement evidence</li>
<li>tokenized settlement</li>
<li>payment status inquiry</li>
<li>idempotency key</li>
<li>ISO 20022 tracking</li>
<li>delegated spending authority</li>
</ul>
</section>
<section>
<name>Industrial and semiconductor</name>
<t>Relevant sections and embodiments: Sections 8.4, 18, 19; E5, E6, E8, E10, E13.</t>
<ul>
<li>DMA</li>
<li>SMMU</li>
<li>IOMMU</li>
<li>trust-domain reassignment</li>
<li>accelerator failover</li>
<li>computational effect commitment</li>
<li>numerical uncertainty</li>
<li>mixed precision</li>
<li>quantization error</li>
<li>programmable hardware</li>
<li>DPU forwarding</li>
<li>SmartNIC offload</li>
<li>installed hardware rule</li>
<li>TCAM rule</li>
<li>P4 fast path</li>
<li>interrupt remapping</li>
<li>checkpoint restoration</li>
<li>always-on coprocessor</li>
<li>wake epoch</li>
<li>decision margin</li>
</ul>
</section>
<section>
<name>Additional keywords</name>
<ul>
<li>consequential operation</li>
<li>externally effective operation</li>
<li>point of no return</li>
<li>effect boundary</li>
<li>protected execution finality</li>
<li>non-bypassable enforcement</li>
<li>fail-closed</li>
<li>single-use capability</li>
<li>act-bound authorization</li>
<li>destination binding</li>
<li>effective principal</li>
<li>effective target identity</li>
<li>operation canonicalization</li>
<li>canonical operation digest</li>
<li>stable act identity</li>
<li>delegation depth</li>
<li>authority narrowing</li>
<li>scope attenuation</li>
<li>capability consumption record</li>
<li>crash-consistent consumption</li>
<li>checkpoint restoration</li>
<li>stale worker</li>
<li>split brain</li>
<li>lease epoch</li>
<li>reconciliation loop</li>
<li>non-resurrection</li>
<li>effect watermark</li>
<li>effect ledger</li>
<li>indeterminate outcome</li>
<li>quarantine state</li>
<li>safe state</li>
<li>compensation action</li>
<li>authoritative outcome criterion</li>
<li>outcome reconciliation</li>
<li>receipt-gated continuation</li>
<li>staged effectuation</li>
<li>canary effect</li>
<li>blast-radius bound</li>
<li>aggregate limit</li>
<li>cumulative effect envelope</li>
</ul>
</section>
</section>
<section anchor="idx-phr">
<name>Search Phrases</name>
<t>72 phrases, each usable as an exact-phrase or proximity query.</t>
<ol>
<li>execution finality for autonomous agents</li>
<li>effectuation boundary enforcement</li>
<li>operation reconstruction before authorization</li>
<li>actual pending operation versus requested operation</li>
<li>candidate act held in a non-effective state</li>
<li>authorization to compute is not authorization to effect</li>
<li>consequence-bound authorization of tool calls</li>
<li>independent validator re-derives the effective operation</li>
<li>scoped non-bearer execution capability bound to the act</li>
<li>finality sink verifies capability against the actual operation</li>
<li>time-of-check to time-of-use gap between approval and execution</li>
<li>stale policy epoch rejected at the effect point</li>
<li>revocation epoch checked at the sink</li>
<li>consumed capability atomic replay protection at the sink</li>
<li>destination-origin effect confirmation receipt</li>
<li>self-reported completion is not effect evidence</li>
<li>bounded real effect before continuation authority</li>
<li>interim effectuation validation of earlier effect evidence</li>
<li>timeout is not evidence of non-effect</li>
<li>retry quarantine for unresolved commit outcome</li>
<li>logical effect identity distinct from attempt identity</li>
<li>authority record independent of model context</li>
<li>authority preserved across context compaction</li>
<li>agent handoff must not widen authority</li>
<li>instruction hierarchy preserved across protocol translation</li>
<li>provider-hosted tool execution cannot be intercepted by application gateway</li>
<li>canonical authority contract for MCP tool semantics</li>
<li>verification oracle isolated from the modifying agent</li>
<li>deployed artifact bound to approved tested artifact</li>
<li>minimal-disclosure task capsule for human assistance</li>
<li>spoken intent interruption epoch</li>
<li>cross-source factual coherence at effectuation</li>
<li>aggregate effect of composed multi-agent work products</li>
<li>materialized composition snapshot validation</li>
<li>document field bound to source row and coordinates</li>
<li>visual referent continuity before physical action</li>
<li>quoted or forwarded message not operative authority</li>
<li>current group audience epoch before disclosure</li>
<li>governance enforcement coverage per execution route</li>
<li>logging is not prevention at the effect boundary</li>
<li>sink rejection as coverage evidence</li>
<li>late-bound ERP posting effect reconstruction</li>
<li>customer commitment gated on fulfillment evidence</li>
<li>live dependency graph revalidated at remediation commit</li>
<li>analytical cohort identity through activation fan-out</li>
<li>cross-engine policy semantic equivalence for governed data</li>
<li>transitive stored-program effect closure</li>
<li>source-commit-certified downstream effect</li>
<li>outbox event is not authority without source commit</li>
<li>agentic payment exposure reservation before authorization</li>
<li>authorization-to-capture exposure continuity</li>
<li>beneficiary-net outcome versus nominal payment instruction</li>
<li>offline digital cash post-compromise fork evidence</li>
<li>edge subrequest authority conservation</li>
<li>causal radio-service impairment before rollout expansion</li>
<li>network slice capacity lending with enforceable reclamation</li>
<li>cumulative connectivity trajectory envelope</li>
<li>consequence-specific numerical error composition with decision margin</li>
<li>remediation attested state correspondence to consequence class</li>
<li>software-to-hardware offload effect equivalence</li>
<li>installed fast-path rule readback versus authorized effect</li>
<li>trust-domain reassignment with prior-access path closure</li>
<li>accelerator worker single-writer effect cutover</li>
<li>negative observation requires sensing opportunity evidence</li>
<li>low-power intent escrow revalidated at wake</li>
<li>in-flight media segment fencing on audience change</li>
<li>key lifecycle transition authorized by effective key state</li>
<li>desired-state controller non-resurrection of revoked generation</li>
<li>causal child-effect graph with per-child authorization</li>
<li>control-to-effect route coverage verification</li>
<li>alternate credential path bypass analysis</li>
<li>fail-closed behavior when validator is unavailable</li>
</ol>
</section>
<section anchor="idx-bool">
<name>Boolean Search Strings</name>
<ul>
<li>"execution finality" AND "autonomous agent" AND "effectuation"</li>
<li>"operation reconstruction" AND "authorization" AND "actual effect"</li>
<li>"AI agent" AND "tool invocation" AND "trusted enforcement"</li>
<li>"MCP" AND "semantic equivalence" AND "tool authorization"</li>
<li>"database trigger" AND "transitive effects" AND "authorization"</li>
<li>"SmartNIC" AND "installed rule" AND "policy equivalence"</li>
<li>"offline digital cash" AND "post-compromise" AND "fork"</li>
<li>"network slicing" AND "capacity lending" AND "reclamation"</li>
<li>"source authority" AND "delegation" AND "final effect"</li>
<li>"attestation" AND "remediation" AND "policy epoch"</li>
<li>("candidate act" OR "proposed operation") AND ("non-effective" OR "uncommitted") AND "finality sink"</li>
<li>("context compaction" OR "context summarization") AND "authority" AND ("prohibition" OR "constraint")</li>
<li>("outbox" OR "change data capture") AND "committed transaction" AND "consumer" AND "authorization"</li>
<li>("GPU" OR "accelerator") AND ("failover" OR "migration") AND ("fencing token" OR "single writer") AND "effect"</li>
<li>("DMA" OR "SMMU") AND "domain reassignment" AND ("quiescence" OR "drain")</li>
<li>("payment" AND "agent") AND ("exposure ledger" OR "reservation") AND "capture"</li>
<li>("route coverage" OR "enforcement coverage") AND "governance" AND ("agent" OR "autonomous")</li>
<li>("negative observation" OR "absence of event") AND "sensor coverage" AND "authorization"</li>
</ul>
</section>
</section>
<section anchor="appendix-lineage">
<name>Patent Lineage and Claim-Integrity Note</name>
<t>This appendix is non-normative.</t>
<dl newline="true">
<dt>Foundational application</dt>
<dd><t>PCT/IB2026/055615, filed 4 June 2026, international publication WO 2026/150382 A2 on 16 July 2026. The inventor describes it as approximately 8,598 pages and more than 300 distinct embodiments; these figures are inventor-reported and unverified.</t></dd>
<dt>Later application</dt>
<dd><t>Indian provisional 202631136555, dated 15 November 2026; approximately 2,405 pages; 39 embodiments (1-38 plus 12A); consolidated claim section. Number and date are from the applicant's filing acknowledgment.</t></dd>
<dt>Filing, publication and priority are different</dt>
<dd><t>Publication on 16 July 2026 (not 16 June) does not by itself establish an earlier priority date for every feature of a later filing. Entitlement requires comparison with the as-filed earlier text.</t></dd>
<dt>Claim integrity</dt>
<dd><t>The supplied merged PDF has 427 physically present numbered claims (1-273, 276-429); 274 and 275 absent; claim 273 ends mid-sentence. The text is neither reconstructed nor supplied here.</t></dd>
<dt>Source status</dt>
<dd><t>The mothership JSON (Technical_Disclosure_1.json) was not available to the editor; its identity with the as-filed text is uncertified.</t></dd>
</dl>
<table anchor="tbl-pct">
<name>PCT records listed in the applicant's examiner index</name>
<thead><tr><th>PCT application</th><th>Filed</th><th>WO publication</th><th>Published</th></tr></thead>
<tbody>
<tr><td>PCT/IB2026/053385</td><td>7 Apr 2026</td><td>WO 2026/115520</td><td>4 Jun 2026</td></tr>
<tr><td>PCT/IB2026/054453</td><td>5 May 2026</td><td>WO 2026/150381</td><td>16 Jul 2026</td></tr>
<tr><td>PCT/IB2026/055615</td><td>4 Jun 2026</td><td>WO 2026/150382</td><td>16 Jul 2026</td></tr>
<tr><td>PCT/IB2026/055760</td><td>7 Jun 2026</td><td>WO 2026/150383</td><td>16 Jul 2026</td></tr>
<tr><td>PCT/IB2026/055870</td><td>10 Jun 2026</td><td>WO 2026/150384</td><td>16 Jul 2026</td></tr>
<tr><td>PCT/IB2026/056058</td><td>13 Jun 2026</td><td>WO 2026/154461</td><td>23 Jul 2026</td></tr>
<tr><td>PCT/IB2026/056809</td><td>1 Jul 2026</td><td>WO 2026/172334</td><td>20 Aug 2026</td></tr>
<tr><td>PCT/IB2026/056941</td><td>6 Jul 2026</td><td>WO 2026/172336</td><td>20 Aug 2026</td></tr>
</tbody>
</table>
<t>The list above is non-exhaustive and is drawn from the applicant's index; it is not a declaration that all listed applications share a valid priority claim. Verify each in WIPO PATENTSCOPE.</t>
</section>
<section anchor="appendix-exam">
<name>Examiner-Oriented Considerations</name>
<t>This appendix is non-normative and holds patentability-oriented considerations that are deliberately kept out of the architecture. It does not assert novelty or inventive step.</t>
<dl newline="true">
<dt>Three separate tests</dt>
<dd><t>(1) Written description and effective date: where is each limitation supported in the as-filed June PCT or in the later filing. (2) Novelty: does one earlier source disclose the combination as properly construed. (3) Inventive step: which closest references, what objective technical problem, whether the skilled person would (not merely could) combine them, and with what reasonable expectation of success.</t></dd>
<dt>Motivation to combine</dt>
<dd><t>For each candidate distinction the examiner should ask what problem the combination solves that its constituents do not. Examples: for E10, why combine software authorization with installed-rule readback; for E32, why bind consumer authority to a source-commit certificate rather than to message authenticity; for E35, why add fork evidence to secure-element prevention.</t></dd>
<dt>Reasonable expectation of success</dt>
<dd><t>The profiles that depend on device or database interfaces (E6, E10, E31) depend on supported readback or hook interfaces. Whether such interfaces exist for a given product is a factual question.</t></dd>
<dt>Technical interdependence</dt>
<dd><t>Each profile claims that its elements cooperate (see <xref target="refinements"/>). An examiner should test whether the elements are a mere aggregation or produce a combined technical effect (for example a single effect-commit owner, or a bounded exposure).</t></dd>
<dt>Closest art categories</dt>
<dd><t>Fencing tokens and lease epochs (E3, E5); transactional outbox and CDC (E32); P4/match-action verification (E10); 3GPP ATSSS and slicing (E7, E38); secure-element offline cash (E35); policy pushdown and row/column security (E30); agent guardrails and MCP gateways (E15, E20, E25).</t></dd>
<dt>Limit</dt>
<dd><t>Stronger detail alone does not establish a new priority date, independent novelty or inventive step.</t></dd>
</dl>
</section>
</back>
</rfc>
