AgentProto Working Group Kayode Oluwole Elekula Internet-Draft Intended status: Informational 16 November 2026 Expires: 20 May 2027 Privacy-Preserving Verification of Linked-Dialog Relationships Across Trust Boundaries draft-elekula-agentproto-linked-dialog-01 Abstract Agentic systems increasingly create, delegate, relay, and transform dialogs across multiple agents, protocols, services, and administrative trust boundaries. In these environments, a verifier may need to determine whether two dialogs are related without requiring unrestricted disclosure of their contents or relying exclusively on the assertions of an intermediary. This document describes properties and a conceptual evaluation model for privacy-preserving verification of linked-dialog relationships across trust boundaries. It distinguishes dialog association from context propagation, relationship verification from authentication and authorization, and verification outcomes from assessments of evidence completeness. The document also considers evidence scope, source positions, temporal references, intermediary continuity, lifecycle changes, identity binding, privacy, and evidence availability. It identifies requirements intended to help future protocol designs represent linked-dialog relationships in a way that can be independently evaluated while limiting unnecessary disclosure. This document does not define a wire format, cryptographic mechanism, identifier format, transport binding, authorization protocol, audit-log structure, or completeness-proof mechanism. It provides terminology, design properties, and interoperability considerations for further discussion. Status of This Memo This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months Elekula Expires 20 May 2027 [Page 1] Internet-Draft Linked-Dialog Verification November 2026 and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at https://www.ietf.org/1id-abstracts.html The list of Internet-Draft Shadow Directories can be accessed at https://www.ietf.org/shadow.html This Internet-Draft will expire on 20 May 2027. Copyright Notice Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved. This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License. Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 5 1.1. Problem Statement . . . . . . . . . . . . . . . . . . . 5 1.2. Goals . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.3. Non-Goals . . . . . . . . . . . . . . . . . . . . . . . 6 1.4. Summary of Contributions . . . . . . . . . . . . . . . . 7 2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . 7 2.1. Dialog . . . . . . . . . . . . . . . . . . . . . . . . . 8 2.2. Linked Dialog . . . . . . . . . . . . . . . . . . . . . 8 2.3. Parent Dialog . . . . . . . . . . . . . . . . . . . . . 8 2.4. Child Dialog . . . . . . . . . . . . . . . . . . . . . . 8 2.5. Relationship Evidence . . . . . . . . . . . . . . . . . 8 2.6. Delegation Provenance . . . . . . . . . . . . . . . . . 9 2.7. Propagation Relationship . . . . . . . . . . . . . . . . 9 2.8. Verifier . . . . . . . . . . . . . . . . . . . . . . . . 9 2.9. Comparison Basis . . . . . . . . . . . . . . . . . . . . 9 2.10. Completeness Within a Declared Boundary . . . . . . . . 10 2.11. Verification Result . . . . . . . . . . . . . . . . . . 10 2.12. Evidence Position . . . . . . . . . . . . . . . . . . . 11 2.13. Temporal Reference . . . . . . . . . . . . . . . . . . 11 2.14. Completeness Appraisal . . . . . . . . . . . . . . . . 12 2.15. Relationship Statement . . . . . . . . . . . . . . . . 12 Elekula Expires 20 May 2027 [Page 2] Internet-Draft Linked-Dialog Verification November 2026 2.16. Evidence Producer . . . . . . . . . . . . . . . . . . . 12 2.17. Evidence Source . . . . . . . . . . . . . . . . . . . . 12 2.18. Ordering Anchor . . . . . . . . . . . . . . . . . . . . 13 3. Conceptual Model . . . . . . . . . . . . . . . . . . . . . . 13 3.1. Relationship Association and Context Propagation . . . . 13 3.2. Relationship Verification and Delegation Provenance . . 13 3.3. Verification and Completeness . . . . . . . . . . . . . 14 3.4. Evidence Position and Temporal Reference . . . . . . . . 14 3.5. Verification Outcomes . . . . . . . . . . . . . . . . . 14 3.6. Historical and Current Claims . . . . . . . . . . . . . 15 4. Privacy-Preserving Design Considerations . . . . . . . . . . 15 4.1. Disclosure Minimization . . . . . . . . . . . . . . . . 15 4.2. Scoped Evidence Disclosure . . . . . . . . . . . . . . . 16 4.3. Correlation Risks . . . . . . . . . . . . . . . . . . . 16 4.4. Selective Disclosure . . . . . . . . . . . . . . . . . . 16 4.5. Privacy and Completeness . . . . . . . . . . . . . . . . 17 4.6. Scope of the Term Privacy-Preserving . . . . . . . . . . 17 5. Proposed Design Properties . . . . . . . . . . . . . . . . . 17 5.1. PDC-1: Scoped Relationship Evidence . . . . . . . . . . 17 5.2. PDC-2: Independently Verifiable Association . . . . . . 18 5.3. PDC-3: Scope and Limitations . . . . . . . . . . . . . . 18 5.4. PDC-4: Cross-Domain and Multi-Intermediary Continuity . 18 5.5. PDC-5: Privacy and Disclosure Minimization . . . . . . . 19 5.6. PDC-6: Verification and Completeness Are Separate . . . 19 5.7. PDC-7: Propagation Relationship Lifecycle Integrity . . 20 5.8. PDC-8: Identity and Principal Binding . . . . . . . . . 20 5.9. PDC-9: Transport-Independent Semantics . . . . . . . . . 20 5.10. PDC-10: Resilience and Evidence Availability . . . . . 21 5.11. PDC-11: Comparison Basis, Evidence Position, and Temporal Reference . . . . . . . . . . . . . . . . . . . 21 5.12. PDC-12: Relationship Statements and Evidence Producers . . . . . . . . . . . . . . . . . . . . . . . 22 5.13. PDC-13: Verifier-Controlled Comparison Basis and Boundary . . . . . . . . . . . . . . . . . . . . . . . . 22 5.14. PDC-14: Cross-Source Ordering . . . . . . . . . . . . . 23 5.15. PDC-15: Relationship Graphs and Composition . . . . . . 23 6. Conceptual Evidence Evaluation Model . . . . . . . . . . . . 24 6.1. Inputs . . . . . . . . . . . . . . . . . . . . . . . . . 24 6.2. Evaluation Steps . . . . . . . . . . . . . . . . . . . . 24 6.3. Result Interpretation . . . . . . . . . . . . . . . . . 25 6.4. Reporting Evidence Positions . . . . . . . . . . . . . . 25 6.5. Reporting Temporal References . . . . . . . . . . . . . 25 6.6. Reporting Conflicting Evidence . . . . . . . . . . . . . 26 7. Verification Result Semantics . . . . . . . . . . . . . . . 26 7.1. Result Scope . . . . . . . . . . . . . . . . . . . . . . 26 7.2. Evidence Availability . . . . . . . . . . . . . . . . . 26 7.3. Temporal Validity . . . . . . . . . . . . . . . . . . . 27 7.4. Negative Findings and Evidence Positions . . . . . . . . 27 8. Interoperability Scenarios . . . . . . . . . . . . . . . . . 28 8.1. Basic Relationship Verification . . . . . . . . . . . . 28 Elekula Expires 20 May 2027 [Page 3] Internet-Draft Linked-Dialog Verification November 2026 8.2. Relationship Substitution . . . . . . . . . . . . . . . 28 8.3. Replay of Relationship Evidence . . . . . . . . . . . . 29 8.4. Expired Evidence . . . . . . . . . . . . . . . . . . . . 29 8.5. Multi-Intermediary Propagation . . . . . . . . . . . . . 29 8.6. Missing Intermediate Evidence . . . . . . . . . . . . . 29 8.7. Conflicting Lifecycle Records . . . . . . . . . . . . . 29 8.8. Revocation Versus Dialog Termination . . . . . . . . . . 30 8.9. Protocol Translation . . . . . . . . . . . . . . . . . . 30 8.10. Privacy-Minimized Evidence . . . . . . . . . . . . . . 30 8.11. Principal and Authorization Separation . . . . . . . . 30 8.12. Unavailable Evidence Source . . . . . . . . . . . . . . 31 8.13. Comparison Basis Consistency . . . . . . . . . . . . . 31 8.14. Local Denial Due to Insufficient Evidence . . . . . . . 31 8.15. Evidence Position Difference . . . . . . . . . . . . . 31 8.16. Historical Validity Versus Current Validity . . . . . . 32 8.17. Timeout During Lifecycle Evaluation . . . . . . . . . . 32 8.18. One-Sided Relationship Claim . . . . . . . . . . . . . 32 8.19. No Ordering Between Sources . . . . . . . . . . . . . . 32 8.20. Multi-Parent Aggregation . . . . . . . . . . . . . . . 33 8.21. Presenter-Selected Boundary . . . . . . . . . . . . . . 33 9. Security Considerations . . . . . . . . . . . . . . . . . . 33 9.1. Forged Relationship Evidence . . . . . . . . . . . . . . 33 9.2. Relationship Substitution . . . . . . . . . . . . . . . 33 9.3. Replay and Context Confusion . . . . . . . . . . . . . . 33 9.4. Unauthorized Delegation . . . . . . . . . . . . . . . . 34 9.5. Lifecycle Manipulation . . . . . . . . . . . . . . . . . 34 9.6. Revocation Ambiguity . . . . . . . . . . . . . . . . . . 34 9.7. Incomplete Evidence Disclosure . . . . . . . . . . . . . 34 9.8. Cross-Domain Trust Confusion . . . . . . . . . . . . . . 34 9.9. Availability Attacks . . . . . . . . . . . . . . . . . . 35 9.10. Evidence Position Manipulation . . . . . . . . . . . . 35 9.11. Conflicting Sources . . . . . . . . . . . . . . . . . . 35 9.12. Correlation and Tracking . . . . . . . . . . . . . . . 36 9.13. Overstated Verification Results . . . . . . . . . . . . 36 9.14. Privacy Leakage Through Metadata . . . . . . . . . . . 36 9.15. Misleading Completeness Claims . . . . . . . . . . . . 36 10. Privacy Considerations . . . . . . . . . . . . . . . . . . 36 11. Relationship to Existing Protocol and Standards Work . . . 37 12. Operational and Interoperability Considerations . . . . . . 38 13. IANA Considerations . . . . . . . . . . . . . . . . . . . . 39 14. Future Work . . . . . . . . . . . . . . . . . . . . . . . . 39 15. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . 39 16. References . . . . . . . . . . . . . . . . . . . . . . . . 40 16.1. Normative References . . . . . . . . . . . . . . . . . 40 16.2. Informative References . . . . . . . . . . . . . . . . 40 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . 41 Author's Address . . . . . . . . . . . . . . . . . . . . . . . . 41 Elekula Expires 20 May 2027 [Page 4] Internet-Draft Linked-Dialog Verification November 2026 1. Introduction Agentic systems increasingly operate through interactions involving multiple agents, tools, services, and execution environments. A task initiated in one dialog may result in one or more child dialogs created by another agent, delegated to an external service, or continued through a different protocol or administrative domain. These interactions create a need to establish whether particular dialogs are related. For example, an initiating agent may create a dialog with a second agent. The second agent may delegate part of the task to a third agent, which may create another dialog with an external service. A later verifier may need to determine whether the resulting dialogs belong to the same interaction chain. The verifier may not have access to all dialog contents. It may also lack direct access to every intermediary, particularly when the dialogs cross organizational or protocol boundaries. A reliable verification model therefore needs to answer a narrower question: What evidence supports the claimed relationship between two or more dialogs, within a stated scope and at a stated point in time? This question is distinct from determining whether: * A participant was authenticated. * A delegation was authorized. * A human approved an action. * The content of a dialog was correct or trustworthy. * Every relevant dialog or intermediary has been disclosed. * A propagation relationship remains active. * A particular action was permitted. A linked-dialog relationship may be established without establishing any of these additional properties. This document proposes privacy-preserving design properties for representing and evaluating linked-dialog relationships across trust boundaries. It aims to make relationship claims independently assessable while limiting the disclosure of dialog content and unrelated operational information. 1.1. Problem Statement A verifier evaluating a linked-dialog claim may encounter: Elekula Expires 20 May 2027 [Page 5] Internet-Draft Linked-Dialog Verification November 2026 * Evidence created by different administrative domains. * Different protocols and identifier namespaces. * Multiple intermediaries. * Partial or unavailable records. * Conflicting lifecycle statements. * Revoked propagation relationships. * Delayed or replicated evidence. * Different source positions at different verification times. * Different interpretations of what a verification result establishes. Without explicit scope and temporal information, two verifiers may produce apparently identical results while having examined materially different evidence. For example, one verifier may evaluate a relationship using a log checkpoint recorded before a revocation. Another may evaluate the same relationship using a later checkpoint containing the revocation. Both may report that the relationship was valid, but their conclusions may refer to different times or states. A verification result should therefore identify not only its conclusion but also the evidence position and temporal reference that bound that conclusion. 1.2. Goals The goals of this document are to: * Define terminology for linked-dialog relationship verification. * Distinguish relationship evidence from delegation and authorization evidence. * Identify properties for independently evaluating relationship claims. * Support cross-domain and multi-intermediary verification. * Minimize unnecessary disclosure of dialog content. * Separate verification outcomes from evidence-completeness assessments. * Distinguish historical validity from current validity. * Make evidence positions and temporal references explicit. * Improve interoperability across protocols and administrative domains. * Identify security and privacy considerations for future protocol work. 1.3. Non-Goals This document does not: * Define a universal dialog identifier. Elekula Expires 20 May 2027 [Page 6] Internet-Draft Linked-Dialog Verification November 2026 * Require a particular cryptographic algorithm. * Specify a signature, token, certificate, or proof format. * Define a global dialog registry. * Require a centralized audit service. * Define a universal trust model. * Establish authorization or delegation policy. * Establish the correctness or safety of agent-generated content. * Require disclosure of complete dialog transcripts. * Define a mandatory transport protocol. * Establish that all intermediaries must be publicly identifiable. * Define a completeness-proof or transparency-log mechanism. These subjects may be addressed by future protocol specifications or related work. 1.4. Summary of Contributions For readers who wish to evaluate the minimal set of primitives this document proposes, its principal contributions are: * A verification result is bound to a claim, a comparison basis, evidence positions, and a temporal reference (PDC-11). * Verification outcomes and completeness appraisals are reported separately (PDC-6). * The lifecycle of a propagation relationship is distinguished from the lifecycle of the dialogs it connects (PDC-7). * Unavailable or missing evidence is reported as such and is treated as neither a positive nor a negative finding (PDC-6 and PDC-10). The remaining properties and the evaluation model support these four contributions. 2. Terminology The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here. The requirements in this document describe proposed properties for designs and implementations. They do not independently impose requirements on any existing protocol. A requirement stated on a relationship mechanism applies to a specification or implementation that claims to provide linked-dialog relationship verification as described in this document. A design that makes no such claim is not affected. Elekula Expires 20 May 2027 [Page 7] Internet-Draft Linked-Dialog Verification November 2026 2.1. Dialog A dialog is a logically related sequence of messages, events, or interactions between one or more participants within an agentic or service environment. A dialog may contain user messages, agent messages, tool interactions, delegated operations, system events, or other protocol-defined activity. A dialog's identity and lifecycle are determined by the system or protocol that defines it. 2.2. Linked Dialog A linked dialog is a dialog for which a relationship to another dialog is asserted or supported by evidence. A linked-dialog relationship may represent parent-child association, continuation, delegation-related propagation, or another explicitly declared relationship. The existence of a linked-dialog relationship does not, by itself, establish that the relationship was authorized. 2.3. Parent Dialog A parent dialog is a dialog identified as the originating or higher-level dialog in a declared parent-child relationship. The term describes the relationship asserted by the relevant evidence. It does not necessarily identify the first interaction in the complete history of a task. A dialog may have more than one parent where work from several dialogs is combined. 2.4. Child Dialog A child dialog is a dialog identified as being created, initiated, or continued as a consequence of activity associated with another dialog. A child dialog may be created by a different agent, service, protocol, or administrative domain. 2.5. Relationship Evidence Relationship evidence is information used to support, challenge, or evaluate a claim that two or more dialogs are related. Elekula Expires 20 May 2027 [Page 8] Internet-Draft Linked-Dialog Verification November 2026 Relationship evidence may include identifiers, signed statements, event references, protocol metadata, source records, or other evidence defined by an implementation. This document does not prescribe a particular evidence format. 2.6. Delegation Provenance Delegation provenance is information describing how an operation or task was transferred, assigned, or delegated between agents or services. Delegation provenance may contribute to a linked-dialog relationship assessment, but it is not equivalent to proof of the relationship itself. 2.7. Propagation Relationship A propagation relationship is a declared relationship through which context, task information, authority claims, or other interaction attributes are carried from one dialog or execution environment into another. A propagation relationship may have its own lifecycle, independent of the lifecycle of the dialogs it connects. 2.8. Verifier A verifier is an entity or process that evaluates evidence associated with a linked-dialog relationship claim. A verifier may be an agent, service, protocol implementation, auditor, or other authorized evaluation component. 2.9. Comparison Basis A comparison basis is the declared set of inputs and criteria used to evaluate a linked-dialog relationship claim. It may include: * Dialog identifiers or references. * Claimed relationship attributes. * Relevant identity or principal bindings. * Trust assumptions and applicable policies. * Lifecycle information. * Evidence sources. * Evidence positions. * Temporal references. * Conflict-resolution rules. Elekula Expires 20 May 2027 [Page 9] Internet-Draft Linked-Dialog Verification November 2026 * Other information materially affecting the evaluation. A comparison basis SHOULD identify the relevant evidence sources and the positions within those sources that were examined. An evidence position may be represented by a source-defined sequence number, log checkpoint, log head, version identifier, or another source-defined reference with equivalent identifying semantics. A timestamp MAY be included as supporting information. A timestamp alone MUST NOT be assumed to establish a complete or ordered source position unless the source's semantics provide that guarantee. Where multiple sources are involved, their positions SHOULD be identified separately. A verifier MUST NOT imply that independently observed source positions constitute one globally consistent snapshot unless the underlying system establishes that property. 2.10. Completeness Within a Declared Boundary Completeness within a declared boundary is an assessment of whether the evidence examined covers the relevant information within an explicitly stated scope. The boundary may be defined by a set of sources, administrative domains, event types, time intervals, intermediary relationships, or other criteria. A completeness assessment MUST identify its applicable boundary and SHOULD identify the evidence positions at which the relevant sources were examined. Completeness at one evidence position does not establish completeness at a later position. Completeness within a declared boundary does not establish universal completeness across undisclosed sources, unknown intermediaries, or systems outside that boundary. 2.11. Verification Result A verification result is the outcome of evaluating a particular linked-dialog relationship claim against a declared comparison basis. A verification result SHOULD identify: * The relationship claim being evaluated. * The comparison basis. * The evidence sources examined. * The evidence positions examined, where available. Elekula Expires 20 May 2027 [Page 10] Internet-Draft Linked-Dialog Verification November 2026 * The temporal reference to which the conclusion applies. * Material assumptions, limitations, and unresolved conflicts. A verification result applies only to the stated claim, comparison basis, evidence positions, and temporal reference. A result that establishes a relationship at one point in time MUST NOT be represented as establishing that the relationship remains valid at all later times. A claim that has not been evaluated has no verification result. The status Not evaluated, described in Section 3.5, is reported in its place. 2.12. Evidence Position An evidence position is a source-defined reference identifying the state or point within an evidence source that was examined by a verifier. Examples include: * A sequence number. * A log checkpoint or log head. * A version identifier. * A source-defined revision. * A timestamp, where the source gives it suitable semantics. * Another source-defined reference that identifies the examined state. An evidence position identifies what was examined in a source. It does not necessarily identify when the underlying event occurred. A source may ingest, replicate, or expose information after the event to which that information relates. Consequently, an evidence position MUST NOT be treated as equivalent to an event-time assertion unless the source semantics support that interpretation. Where multiple sources contribute evidence, each source may have a distinct evidence position. 2.13. Temporal Reference A temporal reference identifies the time or event to which a verification conclusion applies. Examples include: * The time at which a child dialog was created. * The time at which a specified child action occurred. Elekula Expires 20 May 2027 [Page 11] Internet-Draft Linked-Dialog Verification November 2026 * The time at which a verifier evaluated the evidence. * A source-defined event or lifecycle transition. * Another explicitly identified point or interval. A temporal reference is distinct from an evidence position. The evidence position identifies the state of a source that was examined. The temporal reference identifies the time or event about which the verifier is making a statement. A verifier SHOULD identify both when the distinction is material to the result. 2.14. Completeness Appraisal A completeness appraisal is the reported form of the assessment of completeness within a declared boundary defined above. It is a separate assessment describing how fully the available evidence covers that boundary. It is not a substitute for a verification result. A verification result may be satisfied while the completeness appraisal remains limited. Conversely, evidence may be complete within a declared boundary while failing to establish the claimed relationship. 2.15. Relationship Statement A relationship statement is an assertion, produced by a participant in a dialog or by an intermediary, that identifies a linked-dialog relationship between two or more dialogs. A relationship statement is a claim and not proof of the relationship. Its weight depends on who produced it and on what other evidence supports it. 2.16. Evidence Producer An evidence producer is an entity that creates relationship evidence or a relationship statement. Examples include the participant on the parent side of a relationship, the participant on the child side, an intermediary, and the operator of an evidence source. 2.17. Evidence Source An evidence source is a system that records or exposes evidence about dialogs and relationships, such as a log, a registry, or a participant's own records. A verifier examines an evidence source at an evidence position. Elekula Expires 20 May 2027 [Page 12] Internet-Draft Linked-Dialog Verification November 2026 2.18. Ordering Anchor An ordering anchor is a reference that allows a verifier to establish that one event or evidence position precedes, follows, or is concurrent with another, including across different evidence sources. Examples include a reference carried in one record that names the position or digest of a record in another source, a time attestation whose semantics are defined by its issuer, and a single source whose sequence covers both events. 3. Conceptual Model 3.1. Relationship Association and Context Propagation Dialog association and context propagation are related but distinct concepts. Dialog association describes a claimed relationship between dialogs. Context propagation describes the transfer or reuse of information from one dialog or execution environment to another. A system may propagate context without providing sufficient evidence to independently verify the relationship between the originating and resulting dialogs. Similarly, a relationship may be independently verifiable without disclosing the full context that was propagated. A design SHOULD represent these concepts separately. 3.2. Relationship Verification and Delegation Provenance A parent-child relationship may arise through delegation, but the existence of delegation provenance does not automatically establish every property of the resulting dialog relationship. For example: * A delegation record may establish that an agent transferred a task. * A separate relationship assertion may identify a child dialog created during that task. * An authorization record may establish whether the delegation was permitted. * An identity record may establish which principal controlled the delegating agent. These records answer different questions. Elekula Expires 20 May 2027 [Page 13] Internet-Draft Linked-Dialog Verification November 2026 A verifier SHOULD avoid treating evidence of one property as conclusive evidence of another. 3.3. Verification and Completeness Verification and completeness are separate evaluation dimensions. Verification asks whether the available evidence establishes the specific relationship claim under the declared comparison basis. Completeness asks whether the evidence covers the declared boundary sufficiently for the intended assessment. A verifier may be able to verify one relationship without having complete visibility into the entire interaction history. Likewise, a complete record within a declared boundary may show that a claimed relationship is not supported. 3.4. Evidence Position and Temporal Reference Evidence position and temporal reference address different dimensions of an evaluation. Consider a child dialog created at time T1. A propagation relationship is revoked at T2. A verifier examines a source at T3. The verifier may examine a log position that includes the revocation while evaluating whether the propagation relationship was valid at T1. The evidence position is the source state examined at T3. The temporal reference is T1. If the verifier instead evaluates whether the propagation relationship is currently valid at T3, the temporal reference changes. The two evaluations may produce different results without contradiction. A verification result SHOULD make this distinction explicit whenever lifecycle changes could affect the interpretation of the claim. 3.5. Verification Outcomes This document uses four conceptual verification outcomes: * Satisfied: The evidence establishes the specified relationship claim under the declared comparison basis and temporal reference. Elekula Expires 20 May 2027 [Page 14] Internet-Draft Linked-Dialog Verification November 2026 * Unsatisfied: The evidence affirmatively establishes that a required condition of the specified claim is not satisfied under the declared comparison basis and temporal reference. * Indeterminate: The available evidence is insufficient, unavailable, inconsistent, or otherwise unable to support a reliable satisfied or unsatisfied conclusion. * Not evaluated: No evaluation of the claim was performed. This is a status of the request, reported in place of a verification result, and not an outcome of evaluating the claim. The absence of evidence MUST NOT automatically be treated as an unsatisfied finding. Where evidence necessary to evaluate a condition is missing, unavailable, expired, or inaccessible, the result will generally be indeterminate unless other evidence affirmatively establishes that the condition is not satisfied. A local decision to deny an operation because evidence is insufficient is distinct from an unsatisfied verification finding. A system may deny an operation while reporting that the relationship verification outcome is indeterminate. 3.6. Historical and Current Claims A relationship claim may concern historical validity or current validity. A historical claim asks whether the relationship was valid at a specified earlier time or event. A current claim asks whether the relationship is valid at the time identified by the verifier's evaluation. These claims MUST NOT be conflated. A relationship that was valid when a child action occurred may subsequently be revoked. That later revocation does not, by itself, establish that the relationship was invalid when the action occurred. Likewise, evidence that a relationship was valid in the past does not establish that it remains valid now. 4. Privacy-Preserving Design Considerations 4.1. Disclosure Minimization Relationship verification should disclose only the information necessary to evaluate the stated claim. Elekula Expires 20 May 2027 [Page 15] Internet-Draft Linked-Dialog Verification November 2026 A verifier may need to establish that two dialogs are related without receiving: * Full message transcripts. * Unrelated user information. * Internal reasoning or private agent state. * Unrelated tool outputs. * Complete task histories. * Sensitive information from other administrative domains. Implementations SHOULD support evidence representations that allow relationship claims to be evaluated without requiring unrestricted access to dialog contents. 4.2. Scoped Evidence Disclosure Evidence SHOULD be scoped to the claim being evaluated. A disclosure may identify the relevant dialogs, relationship attributes, evidence sources, and applicable temporal reference without exposing unrelated interactions. Where a verifier needs additional evidence, the request SHOULD identify the specific unresolved condition. 4.3. Correlation Risks Dialog identifiers, relationship references, and evidence metadata may enable correlation across services or administrative domains. Implementations SHOULD consider whether identifiers are globally stable, linkable, or reusable across contexts. Where feasible, systems SHOULD avoid unnecessary exposure of persistent identifiers and SHOULD consider scoped or audience-specific references. Privacy mechanisms MUST NOT be treated as substitutes for evidence integrity or reliable relationship evaluation. 4.4. Selective Disclosure Selective disclosure may reduce exposure of unrelated information. However, a verifier should be able to determine what claim was evaluated and whether the disclosed evidence is sufficient for that claim. A disclosure mechanism SHOULD make material limitations visible rather than allowing a partial presentation to appear complete. Elekula Expires 20 May 2027 [Page 16] Internet-Draft Linked-Dialog Verification November 2026 4.5. Privacy and Completeness Privacy minimization may intentionally limit the evidence available to a verifier. A system SHOULD distinguish between: * Evidence intentionally withheld under a privacy policy. * Evidence that does not exist. * Evidence that exists but is unavailable. * Evidence that has expired. * Evidence that was not examined. * Evidence that falls outside the declared boundary. These conditions may have different implications for completeness and verification. 4.6. Scope of the Term Privacy-Preserving In this document, privacy-preserving describes a design objective: limiting the disclosure of dialog content and unrelated operational information to what is needed to evaluate a stated claim. It does not claim anonymity or unlinkability. Relationship verification requires that some party can resolve the dialogs involved, and identifiers, evidence positions, timestamps, and checkpoints can themselves be correlated. A mechanism SHOULD state, for each item disclosed to a verifier, which party learns it and what correlation it enables. Techniques for reducing disclosure, such as audience-scoped references, committed or salted identifiers, and selective disclosure of relationship statements, are outside the scope of this document. A mechanism that relies on such a technique SHOULD state what the technique protects and what it does not. 5. Proposed Design Properties The following properties describe capabilities that future linked-dialog relationship mechanisms should consider. They are design requirements for this problem space, not a claim that any existing protocol currently implements them. 5.1. PDC-1: Scoped Relationship Evidence A linked-dialog relationship mechanism SHOULD provide evidence that identifies the scope of the relationship being asserted. The evidence SHOULD make it possible to distinguish among different Elekula Expires 20 May 2027 [Page 17] Internet-Draft Linked-Dialog Verification November 2026 relationship types, including parent-child association, continuation, delegation-related propagation, and other explicitly declared relationships. The evidence SHOULD identify the dialogs or dialog references to which the claim applies. Evidence for one relationship type MUST NOT automatically be interpreted as evidence for another relationship type. 5.2. PDC-2: Independently Verifiable Association A linked-dialog relationship mechanism SHOULD support independent evaluation by a verifier that is not required to trust the asserting intermediary's conclusion alone. The verifier SHOULD be able to evaluate the relationship using the available evidence and declared comparison basis. Independent verification does not necessarily require disclosure of the full dialog contents or every internal implementation detail. 5.3. PDC-3: Scope and Limitations A relationship assertion SHOULD identify the scope of the claim and its material limitations. A verifier SHOULD be able to determine what the evidence establishes and what it does not establish. A verified relationship MUST NOT, by itself, be interpreted as proof of: * Authorization. * Delegation legitimacy. * Human approval. * Content correctness. * Participant trustworthiness. * Complete historical provenance. * Absence of undisclosed intermediaries. * Current validity beyond the temporal reference of the result. 5.4. PDC-4: Cross-Domain and Multi-Intermediary Continuity A relationship mechanism SHOULD support evaluation of linked-dialog chains that cross multiple agents, protocols, services, or administrative domains. Where a relationship depends on multiple intermediaries, the evidence SHOULD make material gaps or unsupported transitions visible. Elekula Expires 20 May 2027 [Page 18] Internet-Draft Linked-Dialog Verification November 2026 A verifier SHOULD NOT infer that an unobserved intermediary or transition is valid merely because the first and final dialogs are known. A mechanism MAY allow privacy-preserving representations of intermediate relationships where full disclosure is unnecessary. 5.5. PDC-5: Privacy and Disclosure Minimization A relationship mechanism SHOULD minimize the disclosure of dialog content and unrelated operational information. A verifier SHOULD be able to evaluate the declared relationship claim using scoped evidence wherever feasible. Evidence disclosure SHOULD be proportionate to the verification purpose. Implementations SHOULD consider identifier correlation, cross-domain tracking, sensitive metadata exposure, and unintended disclosure of user or agent information. 5.6. PDC-6: Verification and Completeness Are Separate A relationship mechanism MUST distinguish the outcome of relationship verification from an appraisal of evidence completeness. A verification result SHOULD use explicit outcomes, such as: * Satisfied. * Unsatisfied. * Indeterminate. * Not evaluated. A completeness appraisal SHOULD identify the declared boundary to which it applies. Missing, unavailable, expired, or incomplete evidence MUST NOT, by itself, be treated as proof that the historical relationship does not exist. Where necessary evidence is unavailable and no independent evidence resolves the claim, the verifier SHOULD report an indeterminate result. A local system MAY deny an operation because evidence is insufficient. Such a denial MUST NOT be represented as an unsatisfied verification finding unless the evidence affirmatively establishes that the specified relationship condition is not satisfied. Elekula Expires 20 May 2027 [Page 19] Internet-Draft Linked-Dialog Verification November 2026 5.7. PDC-7: Propagation Relationship Lifecycle Integrity A relationship mechanism SHOULD distinguish the lifecycle of a dialog from the lifecycle of a propagation relationship. Revocation or termination of a propagation relationship MUST NOT, by itself, be interpreted as terminating the dialogs involved or erasing the historical fact that a linked-dialog relationship existed. A verification result concerning lifecycle validity SHOULD identify whether it evaluates: * Validity at the time of a specified child-dialog action or event; or * Validity at the time of verification. These are distinct claims. Evidence that a propagation relationship was valid when a child action occurred does not establish that the relationship remains valid at verification time. Likewise, a later revocation does not, by itself, establish that the relationship was invalid when the earlier action occurred. Where evidence cannot establish the relationship's state at the requested temporal reference, the verifier SHOULD report the limitation and use an indeterminate outcome where appropriate. 5.8. PDC-8: Identity and Principal Binding A relationship mechanism SHOULD make relevant identity and principal bindings explicit. Where an assertion identifies an agent, service, or administrative domain, the verifier SHOULD be able to determine what identity claim is being made and what evidence supports that claim. Identity binding MUST NOT be treated as equivalent to authorization. A verified relationship between dialogs does not independently establish that the associated principal was authorized to perform a particular action. 5.9. PDC-9: Transport-Independent Semantics The semantic meaning of a linked-dialog relationship SHOULD be independent of the transport used to carry its evidence. A relationship mechanism MAY be transported through different Elekula Expires 20 May 2027 [Page 20] Internet-Draft Linked-Dialog Verification November 2026 protocols or communication systems. Transport-specific properties MUST NOT silently change the meaning of a relationship claim. Protocol translation SHOULD preserve material relationship attributes, scope, evidence limitations, and temporal semantics. 5.10. PDC-10: Resilience and Evidence Availability A relationship mechanism SHOULD account for unavailable evidence sources, network partitions, timeouts, and inaccessible intermediaries. A verifier MUST NOT treat a timeout, network partition, or unavailable intermediary, by itself, as evidence establishing a positive verification result. Where unavailable information is necessary to evaluate the requested claim, the verifier SHOULD report an indeterminate result or an explicit availability limitation. Unavailability alone MUST NOT be treated as proof that a propagation relationship is active, revoked, or terminated. A verifier MAY still establish a historical relationship using previously obtained evidence, provided that the result identifies the relevant evidence position and temporal reference and does not imply an unsupported current-state conclusion. 5.11. PDC-11: Comparison Basis, Evidence Position, and Temporal Reference A verification result SHOULD identify the comparison basis used to evaluate the relationship claim. Where applicable, the result SHOULD identify: * The relevant evidence sources. * The evidence position examined for each source. * The temporal reference to which the conclusion applies. * The criteria used to compare relationship attributes. * Material conflicts between evidence sources. * The basis on which conflicts were resolved, if they were resolved. Evidence positions MAY use source-defined sequence numbers, log checkpoints, log heads, versions, or other source-defined references. A timestamp SHOULD NOT be treated as a reliable source position unless the source semantics support that interpretation. Elekula Expires 20 May 2027 [Page 21] Internet-Draft Linked-Dialog Verification November 2026 Where different sources provide inconsistent lifecycle records or incomparable positions, the verifier SHOULD disclose the conflict. A verifier MUST NOT silently present a result as globally consistent when the evidence does not establish a consistent comparison basis. Where no reliable comparison basis can be established, the verifier SHOULD report an indeterminate result. 5.12. PDC-12: Relationship Statements and Evidence Producers A relationship mechanism SHOULD identify who produced each relationship statement and in what role, for example the participant on the parent side, the participant on the child side, or an intermediary. A relationship statement produced by only one side of a claimed relationship SHOULD be identified as one-sided. A verifier SHOULD NOT treat a one-sided statement as equivalent to a relationship supported by statements or other evidence from both sides. Where a claim depends on a participant that the verifier cannot reach or does not trust, the result SHOULD state which side's evidence is missing. Support from both sides is not required for every claim. Where a profile accepts a one-sided statement, it SHOULD state the additional evidence that makes the statement acceptable and the limitation that remains. The conclusion of the producer of a statement alone MUST NOT be treated as sufficient to establish the relationship. 5.13. PDC-13: Verifier-Controlled Comparison Basis and Boundary The comparison basis and the completeness boundary used to evaluate a claim SHOULD be selected or approved by the verifier under its own policy, and not solely by the party presenting the evidence. Where a presenter proposes a comparison basis or boundary, the verifier SHOULD evaluate whether it covers the evidence sources material to the claim, including sources that could record revocation or other contrary lifecycle events. A completeness appraisal whose boundary was chosen by the presenter SHOULD be identified as such. A boundary that omits an evidence source known to be material to the claim MUST NOT be reported as complete for that claim. Elekula Expires 20 May 2027 [Page 22] Internet-Draft Linked-Dialog Verification November 2026 5.14. PDC-14: Cross-Source Ordering Where a verification result depends on the order of events or evidence positions recorded in different evidence sources, the mechanism SHOULD provide an ordering anchor, or the verifier SHOULD state that no ordering between the sources could be established. A verifier MUST NOT infer ordering between sources from timestamps alone unless the semantics of those timestamps provide that guarantee, as described for the comparison basis. Where no ordering anchor exists, a historical claim that depends on cross-source ordering SHOULD be reported as indeterminate, and the result SHOULD state the missing ordering. A mechanism MAY record, in the record that creates or continues a dialog, a reference to the evidence position or digest of the originating record in the other source. This is an example of an ordering anchor and is not a required construction. 5.15. PDC-15: Relationship Graphs and Composition A relationship mechanism SHOULD NOT assume that dialogs form a single chain. A dialog MAY have more than one parent, and a parent MAY have more than one child. A claim that spans several relationships SHOULD state whether it requires every relationship along a specified path or the existence of any supported path. For a claim over a specified path, the verification result SHOULD identify the outcome for each relationship the claim depends on, together with the evidence positions and temporal reference used for each. The composed outcome SHOULD be determined as follows: * Satisfied only if every required relationship is Satisfied under compatible temporal references. * Unsatisfied if at least one required relationship is Unsatisfied. * Not evaluated if none of the required relationships was evaluated. * Indeterminate in all other cases. For a claim over any supported path, the verifier SHOULD report the paths examined and the outcome for each. The claim is Satisfied if at least one examined path is Satisfied. It SHOULD NOT be reported as Unsatisfied unless the completeness appraisal for the declared boundary supports the conclusion that no other path exists within that boundary; otherwise it SHOULD be reported as Indeterminate. Where the relationships in a composed claim were evaluated at different temporal references, the result SHOULD state each of them Elekula Expires 20 May 2027 [Page 23] Internet-Draft Linked-Dialog Verification November 2026 and SHOULD NOT present them as a single temporal reference unless their compatibility is established. 6. Conceptual Evidence Evaluation Model This section describes a conceptual process. It does not prescribe a specific implementation or wire format. 6.1. Inputs A verifier may receive: * A linked-dialog relationship claim. * References to the dialogs involved. * Relationship evidence. * Identity or principal assertions. * Delegation provenance, where relevant. * Lifecycle evidence. * Trust information. * Evidence-source references. * Evidence positions. * Relationship statements and the roles of their producers. * Ordering anchors between evidence sources, where available. * A requested temporal reference. * A declared comparison basis. * A declared completeness boundary. Not every evaluation requires every input. The verifier SHOULD identify which inputs are material to the claim being evaluated. 6.2. Evaluation Steps A conceptual evaluation may proceed as follows: 1. Identify the relationship claim. 2. Identify the dialogs and relationship type involved. 3. Establish the declared comparison basis. 4. Identify the evidence sources and their relevant positions. 5. Identify the producer and role of each relationship statement. 6. Identify the temporal reference for the requested conclusion. 7. Establish whether an ordering anchor relates the evidence sources where the conclusion depends on cross-source ordering. 8. Evaluate the relationship evidence. 9. Evaluate relevant identity and principal bindings. 10. Examine lifecycle evidence where the claim depends on lifecycle state. 11. Identify missing, unavailable, expired, or conflicting evidence. 12. Determine the verification outcome. Elekula Expires 20 May 2027 [Page 24] Internet-Draft Linked-Dialog Verification November 2026 13. Separately assess completeness within the declared boundary. 14. Produce a result that identifies its scope, evidence positions, temporal reference, and material limitations. 6.3. Result Interpretation A verification result should be interpreted only within its declared scope. A satisfied result establishes the specified claim under the comparison basis and temporal reference used. It does not automatically establish: * That no other dialogs exist. * That every intermediary has been identified. * That all relevant evidence has been disclosed. * That a relationship remains valid at a later time. * That a principal was authorized. * That an action was safe or correct. An unsatisfied result should identify the condition that was affirmatively not met. An indeterminate result should identify the material uncertainty, evidence gap, unavailable source, or conflict that prevented a reliable conclusion. A not-evaluated result indicates that the claim was not assessed. 6.4. Reporting Evidence Positions Where an evidence source provides a reliable position, the verifier SHOULD record that position in the result. Where the source does not provide a suitable position, the verifier SHOULD state that limitation rather than inventing an equivalent value. For multiple evidence sources, the verifier SHOULD identify their positions independently. A collection of source positions MUST NOT be described as a single globally consistent snapshot unless the relevant system provides a basis for that assertion. 6.5. Reporting Temporal References The verifier SHOULD state whether the conclusion concerns: Elekula Expires 20 May 2027 [Page 25] Internet-Draft Linked-Dialog Verification November 2026 * The time a child dialog was created. * The time a child action occurred. * The time of a lifecycle transition. * The time of verification. * Another explicitly identified event or interval. If the requested temporal reference cannot be established from the evidence, the verifier SHOULD report that limitation. 6.6. Reporting Conflicting Evidence Evidence sources may disagree because of: * Delayed replication. * Different lifecycle views. * Clock differences. * Inconsistent administrative records. * Incomplete intermediary evidence. * Different evidence positions. * Different interpretations of a relationship. A verifier SHOULD identify conflicts that materially affect the conclusion. A verifier SHOULD state the policy or evidence basis used to resolve a conflict. If a conflict cannot be reliably resolved, the result SHOULD be indeterminate rather than silently selecting one source. 7. Verification Result Semantics 7.1. Result Scope A verification result SHOULD identify the specific relationship claim being evaluated. The result SHOULD identify the comparison basis and, where available, the evidence positions examined. It SHOULD identify the temporal reference when the claim depends on historical or current lifecycle state. The result SHOULD distinguish between the existence of a relationship and the current status of any propagation mechanism associated with that relationship. 7.2. Evidence Availability A verifier may not be able to obtain every relevant record. Elekula Expires 20 May 2027 [Page 26] Internet-Draft Linked-Dialog Verification November 2026 An evidence source may be unavailable because of a timeout, network partition, access restriction, retention policy, or service failure. Unavailability is not equivalent to an affirmative lifecycle statement. A verifier SHOULD identify unavailable sources when they materially affect the evaluation. If the missing information is necessary to determine the requested condition, the verifier SHOULD report an indeterminate outcome. 7.3. Temporal Validity A verification result SHOULD make clear whether it describes validity at the time of a specified action or validity at the time of verification. These statements are not interchangeable. For example, a propagation relationship may have been valid when a child dialog performed an action and may have been revoked before a later verifier examined the evidence. A result concerning the earlier action may be satisfied while a result concerning current validity is unsatisfied or indeterminate. The result SHOULD identify the relevant temporal reference and the evidence position used to support it. Where two verifiers examine different evidence positions, their results may differ without necessarily contradicting each other. A result that does not identify its temporal reference SHOULD NOT be interpreted as establishing both historical and current validity. 7.4. Negative Findings and Evidence Positions A negative finding MUST be bounded by the evidence examined and the comparison basis used. A verifier SHOULD identify: * The relationship condition evaluated. * The evidence sources examined. * The evidence position for each relevant source, where available. * The temporal reference. * The affirmative evidence supporting the negative finding. * Any material limitation or unresolved conflict. Elekula Expires 20 May 2027 [Page 27] Internet-Draft Linked-Dialog Verification November 2026 The absence of a record at a particular evidence position does not automatically establish that the corresponding event never occurred. For example, the absence of a revocation record at one log checkpoint does not prove that no revocation occurred before or after that checkpoint. Similarly, an unavailable lifecycle source does not establish that a propagation relationship remains active or has been revoked. A negative finding SHOULD distinguish between an affirmative record showing that a condition was not met and an inability to locate evidence establishing that it was met. A verifier SHOULD NOT describe a result as a definitive historical negative when the evidence position or temporal reference does not support that conclusion. Where timestamps are used, their ordering and completeness semantics SHOULD be stated or otherwise established by the source. Where multiple sources are used, their positions SHOULD be reported separately unless a consistent combined snapshot is established. 8. Interoperability Scenarios The following scenarios illustrate how the proposed properties may apply across different systems. They are not protocol test vectors and do not prescribe a particular implementation. 8.1. Basic Relationship Verification Agent A creates a parent dialog and initiates a child dialog with Agent B. A verifier receives relationship evidence and evaluates whether the child dialog is associated with the parent dialog. The verifier reports the relationship claim, comparison basis, evidence examined, and verification outcome. 8.2. Relationship Substitution An intermediary attempts to substitute a different child dialog reference into an existing relationship claim. A verifier should be able to identify that the presented evidence does not support the substituted relationship. Elekula Expires 20 May 2027 [Page 28] Internet-Draft Linked-Dialog Verification November 2026 8.3. Replay of Relationship Evidence Previously valid relationship evidence is presented in a different context or at a later time. The verifier should evaluate whether the evidence applies to the requested dialogs, relationship type, comparison basis, and temporal reference. Evidence valid for one claim should not automatically establish another claim. 8.4. Expired Evidence A verifier receives relationship evidence that has expired or is no longer acceptable under the applicable policy. The verifier should distinguish between the current acceptability of the evidence and the historical existence of the relationship. Expired evidence does not necessarily establish that the historical relationship never existed. 8.5. Multi-Intermediary Propagation Agent A creates a dialog with Agent B. Agent B delegates work to Agent C. Agent C creates a further dialog with an external service. A verifier evaluates the claimed relationship chain. The evidence should make material intermediary transitions visible while allowing privacy-preserving disclosure where appropriate. 8.6. Missing Intermediate Evidence A verifier receives evidence for the first and final dialogs but lacks evidence for one intermediate transition. The verifier should not silently infer that the missing transition is valid. The result may be indeterminate if the intermediate transition is necessary to establish the complete claimed chain. 8.7. Conflicting Lifecycle Records Two administrative domains provide conflicting information about the lifecycle of a propagation relationship. The verifier should identify the conflicting records, their sources, Elekula Expires 20 May 2027 [Page 29] Internet-Draft Linked-Dialog Verification November 2026 and their evidence positions. If the conflict cannot be resolved under the declared comparison basis, the result should be indeterminate. 8.8. Revocation Versus Dialog Termination A propagation relationship is revoked while the parent and child dialogs remain stored or otherwise identifiable. A verifier should distinguish revocation of the propagation relationship from termination of either dialog. Revocation should not erase the historical existence of the linked-dialog relationship. 8.9. Protocol Translation A relationship is established in one protocol and represented in another protocol after an intermediary translates the relevant information. The translation should preserve the material semantics of the relationship claim, including its scope, temporal reference, and limitations. A verifier should not assume that equivalent-looking identifiers or metadata necessarily carry equivalent semantics. 8.10. Privacy-Minimized Evidence A verifier receives a scoped relationship assertion without receiving the full dialog transcript. The verifier evaluates whether the disclosed evidence is sufficient for the declared relationship claim. The result should identify any material limitations arising from the restricted disclosure. 8.11. Principal and Authorization Separation A relationship assertion identifies the agents associated with two dialogs. The verifier determines that the relationship is supported but has no evidence establishing whether the relevant principal authorized the action. The relationship result should not be interpreted as an authorization Elekula Expires 20 May 2027 [Page 30] Internet-Draft Linked-Dialog Verification November 2026 result. 8.12. Unavailable Evidence Source A verifier attempts to contact an intermediary that is temporarily unavailable. The verifier should not treat the timeout as evidence that the relationship is valid, invalid, active, or revoked. If the unavailable source is necessary to establish the requested claim, the verifier should report an indeterminate outcome or an explicit limitation. 8.13. Comparison Basis Consistency Two verifiers evaluate the same relationship claim using different evidence sources or positions. Their results should identify those differences so that the results can be compared meaningfully. Identical outcome labels do not necessarily mean that the verifiers established identical facts. 8.14. Local Denial Due to Insufficient Evidence A local system refuses to proceed because it cannot obtain sufficient relationship evidence. The system may deny the operation as a matter of local policy. That decision should not be confused with an unsatisfied verification finding unless the evidence affirmatively establishes that the relationship condition is not met. 8.15. Evidence Position Difference Two verifiers examine the same source at different positions. The earlier position does not contain a lifecycle transition, while the later position does. Each verifier should identify the position examined. The results should not be presented as equivalent merely because they use the same outcome label. If the later position includes a revocation record, a verifier evaluating current validity may reach a different conclusion from one evaluating an earlier historical event. Elekula Expires 20 May 2027 [Page 31] Internet-Draft Linked-Dialog Verification November 2026 8.16. Historical Validity Versus Current Validity A propagation relationship is valid when a child action occurs and is revoked before a later verification. A verifier evaluating validity at the time of the child action may establish that the relationship was valid at that time. A verifier evaluating current validity may establish that it is no longer valid. The results are not contradictory if their temporal references are explicit. 8.17. Timeout During Lifecycle Evaluation A verifier has evidence that a relationship existed historically but cannot contact the lifecycle source needed to determine its current status. The verifier may report the historical relationship as satisfied if the available evidence supports that conclusion. The verifier should separately report current validity as indeterminate if the unavailable lifecycle information is necessary to evaluate the current-state claim. The timeout itself does not establish a positive or negative lifecycle state. 8.18. One-Sided Relationship Claim A child-side participant presents a statement claiming a prominent dialog as its parent. No statement or other evidence from the claimed parent side is available. The verifier should identify the statement as one-sided. Unless the governing profile defines additional evidence that makes such a statement acceptable, the verifier should not report the relationship as satisfied on that statement alone, and the result may be indeterminate. 8.19. No Ordering Between Sources The creation of a child dialog is recorded in one domain's evidence source, and a revocation of the propagation relationship is recorded in another domain's evidence source. The sources share no ordering anchor. A verifier asked whether the relationship was valid when the child Elekula Expires 20 May 2027 [Page 32] Internet-Draft Linked-Dialog Verification November 2026 dialog was created cannot determine whether the revocation preceded the creation. The result should be indeterminate and should state that no ordering between the sources was established. 8.20. Multi-Parent Aggregation Dialog D is created by combining work from dialogs B and C, both of which descend from dialog A. A verifier evaluates the claim that D is related to A. The result should state whether the claim requires a specified path or any supported path, identify the outcome of each relationship on each path examined, and apply the composition rules for that claim type. 8.21. Presenter-Selected Boundary A presenter submits evidence with a declared boundary that excludes the evidence source that would record a revocation. The verifier should identify that the boundary was selected by the presenter, assess whether it covers the sources material to the claim, and should not report the claim as complete with respect to the omitted source. 9. Security Considerations 9.1. Forged Relationship Evidence An attacker may attempt to create or alter evidence to make unrelated dialogs appear connected. Implementations should consider evidence integrity, source authentication, and appropriate validation mechanisms. This document does not prescribe a particular cryptographic solution. 9.2. Relationship Substitution An attacker may attempt to substitute dialog references, relationship attributes, or intermediary identities. A verifier should evaluate the specific claim and evidence bindings rather than relying on superficial identifier similarity. 9.3. Replay and Context Confusion Previously valid evidence may be replayed in another context. Implementations should consider binding evidence to the relevant Elekula Expires 20 May 2027 [Page 33] Internet-Draft Linked-Dialog Verification November 2026 relationship claim, scope, and temporal conditions. 9.4. Unauthorized Delegation A relationship may be genuine even if the delegation that produced it was unauthorized. Relationship verification MUST NOT be treated as proof of authorization. Implementations requiring authorization decisions should evaluate authorization evidence separately. 9.5. Lifecycle Manipulation An attacker may attempt to suppress, forge, delay, or reorder lifecycle records. Implementations should consider the integrity and ordering semantics of lifecycle evidence. A verifier should identify the evidence positions examined and disclose material uncertainty about record ordering. 9.6. Revocation Ambiguity A revocation record may be interpreted incorrectly as proof that a historical relationship never existed. Implementations should distinguish the historical existence of a relationship from the current validity of its propagation mechanism. 9.7. Incomplete Evidence Disclosure An asserting intermediary may disclose only favorable evidence while withholding records that materially change the interpretation of the relationship. A verifier should assess completeness within a declared boundary and should avoid presenting a limited evidence review as a complete historical assessment. 9.8. Cross-Domain Trust Confusion Different administrative domains may use different identity systems, lifecycle policies, and trust assumptions. A verifier should identify material trust assumptions and avoid treating one domain's assertion as universally authoritative. Elekula Expires 20 May 2027 [Page 34] Internet-Draft Linked-Dialog Verification November 2026 9.9. Availability Attacks An attacker may disrupt access to an evidence source or intermediary. A verifier MUST NOT treat a timeout, network partition, or unavailable intermediary, by itself, as evidence establishing a positive verification result. The absence of a response is not an affirmative lifecycle statement. Where a current-state conclusion depends on an unavailable source, the verifier should report the resulting limitation. Previously obtained evidence may continue to support a historical finding, provided the result identifies its evidence position and temporal reference and does not imply an unsupported current-state conclusion. An attacker able to block access to an evidence source can force indeterminate results and thereby cause operations to be denied under local policy. Deployments that require availability SHOULD consider redundant evidence sources and cached evidence with identified positions, while preserving the distinctions required by PDC-10. 9.10. Evidence Position Manipulation An attacker may attempt to provide stale, misleading, or selectively chosen evidence positions. Implementations should consider whether evidence positions are authenticated, comparable, and meaningful under the source's semantics. A verifier should not assume that a timestamp or version label establishes a reliable ordering unless the source provides that guarantee. 9.11. Conflicting Sources An attacker or malfunctioning system may provide inconsistent evidence from different sources. A verifier should identify material conflicts and the basis used to resolve them. Where no reliable resolution is available, the verifier should report an indeterminate outcome. Elekula Expires 20 May 2027 [Page 35] Internet-Draft Linked-Dialog Verification November 2026 9.12. Correlation and Tracking Stable dialog identifiers and relationship evidence may permit tracking across services or administrative domains. Implementations should consider identifier scope, retention, audience restrictions, and unnecessary cross-domain correlation. 9.13. Overstated Verification Results A system may present a narrow verification result as proof of broader trustworthiness, authorization, correctness, or complete provenance. Implementations should make the scope and limitations of results explicit. A verifier should not make claims that exceed the evidence, comparison basis, evidence positions, or temporal reference. 9.14. Privacy Leakage Through Metadata Even when message content is withheld, relationship metadata may reveal participants, organizational structure, task timing, or operational patterns. Implementations should evaluate metadata disclosure as part of their privacy threat model. 9.15. Misleading Completeness Claims A completeness appraisal may be misunderstood as proof that all relevant evidence in all systems has been examined. A completeness statement should identify its declared boundary and evidence positions. It should not imply universal completeness unless the underlying system can establish that property. 10. Privacy Considerations Privacy is a central concern in cross-domain relationship verification. The ability to establish that two dialogs are related should not automatically require access to their full contents. Implementations should consider the following: * Minimize disclosure to the evidence required for the stated claim. Elekula Expires 20 May 2027 [Page 36] Internet-Draft Linked-Dialog Verification November 2026 * Avoid exposing unrelated dialog content. * Limit unnecessary use of globally stable identifiers. * Consider the correlation risks of relationship metadata. * Identify the intended audience for disclosed evidence. * Distinguish intentionally withheld information from unavailable or nonexistent evidence. * Make completeness limitations visible. * Avoid retaining relationship evidence longer than necessary for the stated purpose. * Consider whether evidence positions and timestamps reveal sensitive operational patterns. * Ensure that privacy minimization does not cause a partial result to be misrepresented as complete. A privacy-preserving mechanism should make it possible to disclose enough information to evaluate a relationship while limiting unrelated information exposure. 11. Relationship to Existing Protocol and Standards Work This document describes a cross-cutting problem that may be relevant to agent communication protocols [AGENTPROTO] and related work. The concepts may be applicable to systems involving: * Agent-to-agent communication [A2A]. * Model Context Protocol (MCP) interactions [MCP]. * Delegated tool execution. * HTTP-based services [HTTP]. * QUIC-based communication [QUIC]. * WebTransport [WEBTRANSPORT]. * WebRTC [WEBRTC]. * Media over QUIC (MoQ) [MOQ]. * Other systems that create or propagate dialogs across execution or administrative boundaries. These references identify areas where the problem may arise. They do not imply that the protocols listed above currently provide all the properties described in this document. The concepts are intended to complement, rather than replace, protocol-specific identity, authorization, delegation, security, and lifecycle mechanisms. Where a dialog management protocol defines dialog identifiers, lifecycle states, or context propagation, such as the work chartered in the AgentProto Working Group [AGENTPROTO], linked-dialog relationship evidence SHOULD refer to those identifiers and lifecycle states and SHOULD NOT define parallel ones. Elekula Expires 20 May 2027 [Page 37] Internet-Draft Linked-Dialog Verification November 2026 Individual drafts in this area describe typed lineage relationships among dialogs [AGENTPROTO-CONTEXT] and cryptographically verifiable conversation records [VERIFIABLE-CONVERSATIONS]. This document does not depend on any particular relationship representation or record format. It addresses how a claim made using such representations can be evaluated, and what the result does and does not establish. This document does not define audit data structures or their production, validation, or consumption. Future work may consider whether linked-dialog relationship evidence should be specified within an existing protocol, through an extension, or through a protocol-independent mechanism. 12. Operational and Interoperability Considerations Future implementations should consider documenting: * Supported relationship types. * Identifier scope and lifecycle. * Evidence-source semantics. * Evidence-position semantics. * Temporal-reference semantics. * Trust assumptions. * Lifecycle and revocation behavior. * Conflict-resolution rules. * Completeness boundaries. * Privacy and disclosure policies. * Behavior when evidence is unavailable. * The distinction between relationship verification and authorization. Interoperability testing should include cases where: * Evidence is incomplete. * Evidence is unavailable. * Two sources disagree. * A relationship is revoked after a child action. * A verifier examines a later evidence position. * Two verifiers use different temporal references. * A historical claim is satisfied while a current claim is not. * A local system denies an operation despite an indeterminate verification outcome. * A relationship statement is supported from only one side. * Two evidence sources share no ordering anchor. * A dialog has more than one parent. * A presenter selects a boundary that omits a material evidence source. The purpose of such testing should be to ensure that implementations Elekula Expires 20 May 2027 [Page 38] Internet-Draft Linked-Dialog Verification November 2026 do not give materially different meanings to the same result labels. 13. IANA Considerations This document has no IANA actions. 14. Future Work Further work may explore: * A formal model for linked-dialog relationship claims. * Interoperable evidence representations. * Evidence-position and temporal-reference conventions. * Relationship lifecycle and revocation semantics. * Privacy-preserving disclosure mechanisms. * Cross-domain trust and identity binding. * Consistency and conflict-resolution models for multiple evidence sources. * Test vectors for verification outcomes and completeness appraisals. * Integration with existing agent communication protocols. Threat models for evidence substitution, replay, unavailability, and correlation. Any future specification should preserve the distinction between relationship verification, authorization, delegation provenance, and completeness. 15. Conclusion Agentic systems increasingly create relationships between dialogs across agents, services, protocols, and administrative trust boundaries. Verifying these relationships requires more than accepting an intermediary's assertion. A verifier needs a declared claim, an identifiable comparison basis, appropriately scoped evidence, and a clear understanding of what the evidence establishes. Privacy-preserving verification should avoid requiring unrestricted disclosure of dialog contents while still making material limitations visible. This document identifies fifteen proposed design properties for that purpose. It emphasizes that relationship verification is distinct from authorization, that verification and completeness are separate assessments, and that lifecycle changes must be interpreted in relation to the time or event being evaluated. Elekula Expires 20 May 2027 [Page 39] Internet-Draft Linked-Dialog Verification November 2026 It also establishes two important dimensions of a verification result: * The evidence position, identifying the source state examined. * The temporal reference, identifying the time or event to which the conclusion applies. These dimensions allow a verifier to state not only what it concludes, but also the evidential and temporal limits of that conclusion. A reliable linked-dialog verification model should make those limits explicit, preserve privacy, and avoid claims stronger than the available evidence supports. 16. References 16.1. Normative References [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. [RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, May 2017. 16.2. Informative References [AGENTPROTO] IETF, "Agent Communication Protocols (agentproto) Working Group", . [MCP] Model Context Protocol documentation. [A2A] Agent-to-Agent protocol materials. [HTTP] Fielding, R., et al., "HTTP Semantics", RFC 9110, June 2022. [QUIC] Iyengar, J. and M. Thomson, "QUIC: A UDP-Based Multiplexed and Secure Transport", RFC 9000, May 2021. [WEBTRANSPORT] WebTransport protocol specifications and related materials. [WEBRTC] WebRTC specifications and related materials. [MOQ] Media over QUIC working group materials. [AGENTPROTO-CONTEXT] Moussa, H. and A. Akhavain, "Context and Correlation Elekula Expires 20 May 2027 [Page 40] Internet-Draft Linked-Dialog Verification November 2026 Management Architecture for Distributed Agentic Activities", draft-agentproto-context-management-architecture-00, October 2026, Work in Progress. [VERIFIABLE-CONVERSATIONS] Birkholz, H., Heldt, T., and O. Steele, "Verifiable Agent Conversation Records", draft-birkholz-verifiable-agent-conversations-01, August 2026, Work in Progress. Acknowledgements The author thanks Toshikatsu Oga for review comments that helped clarify the treatment of negative findings, evidence availability, and propagation relationship lifecycle semantics. The comments contributed to the refinement of the comparison basis, the distinction between evidence positions and temporal references, and the separation of historical validity from current validity. Author's Address Kayode Oluwole Elekula Lagos Nigeria Email: kayodeelekula@gmail.com Elekula Expires 20 May 2027 [Page 41]