<?xml version="1.0" encoding="UTF-8"?>
<rfc xmlns:xi="http://www.w3.org/2001/XInclude" category="info" docName="draft-das-causal-semantic-handoff-diagnostics-00" ipr="trust200902" submissionType="independent" xml:lang="en" version="3" tocInclude="true" tocDepth="3" sortRefs="true" symRefs="true">
<front>
<title abbrev="Causal Diagnostics for Semantic Drift">Causal Diagnostics for Semantic Drift in Autonomous Agent Communications: Counterfactual Testing, Source-Role Separation, and Evidence-Bounded Handoff Localization</title>
<seriesInfo name="Internet-Draft" value="draft-das-causal-semantic-handoff-diagnostics-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 LinkedIn profile URL as <uri>; not available to the editor -->
</address>
</author>
<date year="2026" month="November" day="16"/>
<area>Security</area>
<workgroup>Independent Submission</workgroup>
<keyword>semantic drift</keyword><keyword>counterfactual diagnostics</keyword><keyword>multi-agent AI</keyword><keyword>handoff</keyword><keyword>prompt injection</keyword><keyword>fault localization</keyword>
<abstract>
<t>Multi-agent AI systems pass instructions, facts and tool outputs through summaries, translations, delegations and protocol conversions. When a downstream proposed operation departs from the originating instruction, existing records may show that a message was sent, authenticated or traced, yet not show whether a lower-trust source's instruction-like statement influenced a governing condition, whether an intermediary dropped the condition, or whether the receiver misapplied it. This document describes a diagnostic architecture, and an optional transport-neutral evidence model, for investigating such semantic drift. It uses typed originating task contracts, condition-scoped source roles, versioned dependency graphs, handoff lineage, matched counterfactual tests that vary legitimate facts and unauthorized instruction-like content separately, adaptive witness selection, uncertainty-aware comparison, and evidence-bounded localization that reports an earliest observable divergence, a bounded interval, or abstains.</t>
<t>The architecture is diagnostic only. It does not grant, veto or finalize any execution; authorization stays with separate systems. The document is an individual, informational proposal; it makes no novelty, compliance, adoption or performance claim, and reports no experimental results beyond those the underlying filing itself states.</t>
</abstract><note title="Source and Patent Information">
<t>Technical source: Indian Provisional Patent Application 202631134241, dated 8 November 2026 (as reported by the applicant), titled "Systems and Methods for Causal Semantic Handoff Verification in Autonomous Agent Communications". Foundational reference: PCT/IB2026/055615, filed 4 June 2026, published 16 July 2026 as WO 2026/150382 A2. See <xref target="ipr"/>.</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>Related public research: https://doi.org/10.5281/zenodo.24076231 (see <xref target="ZENODO-CAUSAL"/>).</t>
<t>The IPR disclosure for this document will be made at the appropriate time through the official IETF process and in accordance with IETF guidelines (see <xref target="ipr"/>).</t>
<t>AI-assisted drafting was used. Readers must verify patent numbers, dates, claim text, product capabilities and standards references against primary sources.</t>
</note></front>
<middle>
<section anchor="intro">
<name>Introduction</name>
<t>Autonomous agents increasingly exchange work: an agent summarizes a request, another retrieves documents or calls tools, a third proposes an operation. At each step the meaning of the originating instruction can change without any transport or signature failure. A price limit may be omitted in a summary, a supplier's statement that the budget has been raised may be treated as a governing instruction, a converter may rewrite a field, or a receiver may restate a rule correctly yet emit a non-compliant proposal.</t>
<t>This document describes a diagnostic architecture for investigating such changes. Its distinguishing technique is paired counterfactual testing that separates legitimate factual influence from unauthorized instruction-like influence by source role, combined with adaptive selection of the tests worth running and with localization that states its own uncertainty.</t>
<t>The architecture does not decide whether any operation proceeds. Message authenticity, integrity, authorization and execution control belong to other systems; this one supplies evidence about where observable meaning changed and how well that is supported.</t>
<section anchor="problem">
<name>Problem Statement</name>
<t>A single incorrect proposed operation can be compatible with several explanations: an earlier agent omitted a limit; the receiver misread a received limit; an external document falsely asserted a change; a counteroffer was treated as accepted; a stale fact was used; the receiver understood the instruction but generated inconsistent tool arguments; or several of these together. A final-outcome comparison, a hash mismatch, a provenance label or a state-freshness check each identifies part of the situation but none separates these explanations.</t>
<t>Identifying the earliest observable discrepancy is not the same as proving the first underlying causal event, and sensitivity of an agent to an injected statement is not proof that the statement caused a particular historical incident. The architecture preserves these distinctions.</t>
</section>
<section anchor="scope">
<name>Scope and Non-Goals</name>
<ul>
<li>In scope: diagnosis of condition-relevant instruction interpretation changes across agent communications; a diagnostic evidence model.</li>
<li>Non-goal: granting, withholding or vetoing execution authority; that stays with separate authorization and execution-control systems.</li>
<li>Non-goal: reconstructing hidden reasoning of an agent from an external interface.</li>
<li>Non-goal: a new agent protocol, message format or change to A2A, MCP or any other protocol.</li>
<li>Non-goal: a claim of solving alignment, universal prompt-injection detection, legal compliance or novelty.</li>
</ul>
</section>
</section>
<section anchor="terms">
<name>Terminology and Conventions</name>
<t>Requirement words are used only where interoperability behavior is intentionally specified (Section on diagnostic evidence structures). Where the underlying disclosure only presents implementation alternatives, this document uses descriptive language. The key words <bcp14>MUST</bcp14>, <bcp14>MUST NOT</bcp14>, <bcp14>SHOULD</bcp14>, <bcp14>SHOULD NOT</bcp14> and <bcp14>MAY</bcp14> in uppercase are to be interpreted as in BCP 14 <xref target="RFC2119"/> <xref target="RFC8174"/>.</t>
<dl newline="true">
<dt>Originating Task Contract</dt>
<dd><t>Versioned typed representation of the conditions that materially affect the intended operation.</t></dd>
<dt>Condition</dt>
<dd><t>A typed predicate of the contract, for example a maximum amount or a prerequisite.</t></dd>
<dt>Source role</dt>
<dd><t>What a given source is eligible to do with respect to a given condition: supply facts, propose, or modify.</t></dd>
<dt>Handoff</dt>
<dd><t>Transfer or transformation of task information across a communication boundary.</t></dd>
<dt>Witness</dt>
<dd><t>A matched, admissible diagnostic test definition.</t></dd>
<dt>Non-effectuating execution</dt>
<dd><t>Execution of a diagnostic input in a context where no real external effect is produced.</t></dd>
<dt>Earliest observable divergence</dt>
<dd><t>First observed boundary at which a condition's representation differs from the reference; not necessarily the first underlying cause.</t></dd>
<dt>Fault interval</dt>
<dd><t>Bounded range of transitions between the last observed preserving boundary and the first observed divergent boundary.</t></dd>
<dt>Abstention</dt>
<dd><t>A finding that the evidence does not support a conclusion.</t></dd>
<dt>Diagnostic finding</dt>
<dd><t>Evidence-bounded output of this architecture; not an authorization decision.</t></dd>
</dl>
</section>
<section anchor="distinctions">
<name>Twelve Distinctions</name>
<t>The architecture treats the following as separate questions. Evidence on one does not answer another.</t>
<ol>
<li>Message authenticity: who sent it.</li>
<li>Message integrity: whether it changed in transit.</li>
<li>Source-role eligibility: whether this source may modify this condition.</li>
<li>Preservation of task conditions: whether each condition survived the handoff.</li>
<li>Observed interpretation: what an agent's observable response indicates.</li>
<li>Proposed-operation correspondence: whether the proposal matches the conditions.</li>
<li>Factual sensitivity: legitimate dependence of behavior on a fact.</li>
<li>Unauthorized-instruction sensitivity: dependence on an instruction-like claim from a source not eligible to modify.</li>
<li>Earliest observable divergence: first observed boundary of difference.</li>
<li>Historical cause versus experimentally supported influence: a test shows influence under test conditions, not necessarily the cause of a past incident.</li>
<li>Unresolved evidence: what could not be established.</li>
<li>Execution authorization: remains outside this architecture.</li>
</ol>
</section>
<section anchor="arch">
<name>Architecture Overview</name>
<t>The architecture comprises components A to J. A single pipeline is shown in &lt;xref target="fig-main"/&gt;; it is a diagnostic pipeline, not an authority pipeline.</t>
<figure anchor="fig-main">
<name>Diagnostic pipeline (not an authority pipeline)</name>
<artwork type="ascii-art"><![CDATA[
 Originating task --> [A] Typed task contract
                           |
 Source segments ----> [B] Condition-scoped source-role model
                           |
                      [C] Versioned dependency graph
                           |
 Agent handoffs -----> [D] Handoff observation and lineage
                           |
        +------------------+------------------+
        |                  |                  |
   [E] Counterfactual  [F] Adaptive      [G] Uncertainty-aware
       processor           witness           behavioral
   (non-effectuating)      selector          comparator
        +------------------+------------------+
                           |
                      [H] Evidence-bounded fault localizer
                           |
                      [I] Diagnostic evidence exporter
                           |
                      [J] Optional correction / regression
                           |
                           v
        DIAGNOSTIC FINDING (evidence, intervals, abstention)
   - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
   Separate systems keep execution authorization. The finding
   is one input at most; it grants and vetoes nothing.]]></artwork>
</figure>
<dl newline="true">
<dt>A. Typed originating task contract</dt>
<dd><t>Builds versioned typed predicates, with unresolved conditions kept separate.</t></dd>
<dt>B. Condition-scoped source-role model</dt>
<dd><t>Per-condition, per-source eligibility; separates authentication from modification eligibility.</t></dd>
<dt>C. Versioned dependency graph</dt>
<dd><t>Links conditions, facts, source segments and derived messages.</t></dd>
<dt>D. Handoff observation and lineage</dt>
<dd><t>Records message versions and projects them onto conditions.</t></dd>
<dt>E. Counterfactual diagnostic processor</dt>
<dd><t>Builds and runs admissible matched witnesses in a non-effectuating context.</t></dd>
<dt>F. Adaptive witness selector</dt>
<dd><t>Chooses tests by expected usefulness, cost, feasibility and safety.</t></dd>
<dt>G. Uncertainty-aware behavioral comparator</dt>
<dd><t>Compares paired observations with intervals and decision categories.</t></dd>
<dt>H. Evidence-bounded fault localizer</dt>
<dd><t>Reports earliest observable divergence, bounded interval or abstention.</t></dd>
<dt>I. Diagnostic evidence exporter</dt>
<dd><t>Packages findings with per-claim status and optional selective disclosure.</t></dd>
<dt>J. Optional correction and regression processor</dt>
<dd><t>Tests candidate corrections and regressions.</t></dd>
</dl>
<t>Embodiment cross-references: A=E01; B=E02; C=E03; D=E04 to E06 and E12; E=E07; F=E08; G=E09; H=E10, E11, E13, E14; I=E06 and E18; J=E16, E17; temporal validity=E15 (see &lt;xref target="app-cross"/&gt;).</t>
</section>
<section anchor="roles">
<name>Typed Task Conditions and Source Roles</name>
<t>Source authentication answers who sent a segment. Source-role eligibility answers whether that source may modify a particular condition. A supplier may quote a price (a fact) but not raise the customer's budget (a condition). Roles are condition-scoped and are carried through quotation, translation and summarization: a derived message does not gain more eligibility than its source had merely because an intermediary produced it.</t>
<figure anchor="fig-roles">
<name>Source-role separation inside one message</name>
<artwork type="ascii-art"><![CDATA[
 Message from supplier agent:
 +----------------------------------------------------------+
 | "Price is INR 2,500."        -> FACT segment             |
 | "Budget is now INR 5,000."   -> MODIFICATION CLAIM       |
 | "Proceed with booking."      -> INSTRUCTION-LIKE REQUEST |
 +----------------------------------------------------------+
        |                    |                     |
  role: factual         role: not eligible     role: not eligible
  supplier              to modify C3           to direct action
        |                    |                     |
  retained as fact      recorded as claim,     recorded as request,
                        not as a revision      not as instruction]]></artwork>
</figure>
<t>Missing, conflicting, expired or unverifiable role information is recorded as unresolved. Roles may be established by cryptographic assertions or by non-cryptographic configuration; the diagnostic method is the same, and a role assertion's authenticity does not make its content causally conclusive.</t>
</section>
<section anchor="lineage">
<name>Handoff Observation and Lineage</name>
<t>The lineage records, for each observable boundary, the representation of each governing condition. Four positions are distinguished: sender input, sender-generated output, submitted payload, receiver-observed input, plus the receiver's proposed operation. Streamed fragments are finalized into message versions; sender and receiver observations of the same handoff may differ and are both kept.</t>
<figure anchor="fig-lineage">
<name>Multi-agent handoff lineage</name>
<artwork type="ascii-art"><![CDATA[
 Source S --> Agent A --> Converter --> Agent B --> proposal P
   |           |            |             |            |
  seg-1     out-A v1     payload v1    in-B v1     op-proposal
   \__________|____________|_____________|____________/
       condition C3 status recorded at each boundary:
       (sender input, sender output, submitted, received)]]></artwork>
</figure>
<t>Lineage is not message logging: it binds versions to sources and to conditions. Where a boundary cannot be observed it is marked unavailable; missing information is never treated as proof that no change occurred.</t>
</section>
<section anchor="cf">
<name>Counterfactual Intervention Model</name>
<t>The counterfactual method varies one designated aspect of an input while holding the rest fixed, and runs both variants through a non-effectuating interface. Required controls for every test: the originating condition and revision are unchanged; the source role context is the same; the factual domain is valid; replay capability exists; the execution is safe; and the pair is validated before its result is used.</t>
<figure anchor="fig-matched">
<name>Matched counterfactual design</name>
<artwork type="ascii-art"><![CDATA[
            Contract C3 (max INR 2,000) fixed in all cells
                      Unsupported budget claim
                         absent        present
  Price      1,800    [ W1 ]         [ W2 ]
  fact       2,500    [ W3 ]         [ W4 ]

  W1 vs W3 : factual sensitivity (legitimate)
  W1 vs W2 : unauthorized-instruction sensitivity
  Factorial: all four cells; interaction term
  Each pair: one designated difference, invariants checked,
  run through a non-effectuating interface only.]]></artwork>
</figure>
<dl newline="true">
<dt>Test A: legitimate factual influence</dt>
<dd><t>Vary a factual field (for example price 1,800 versus 2,500, including boundary values just below and above the limit) with no instruction-like claim. A compliant agent is expected to change its decision across the limit; this difference is legitimate factual sensitivity.</t></dd>
<dt>Test B: unauthorized instruction influence</dt>
<dd><t>Hold the fact fixed and vary only an instruction-like assertion from a source not eligible to modify the condition (present versus absent). A material difference indicates susceptibility to that statement under test conditions.</t></dd>
<dt>Test C: combined or factorial</dt>
<dd><t>Vary fact and claim together in a 2 by 2 design to estimate main effects and interaction; interpretation is limited where cells are incomparable.</t></dd>
<dt>Test D: message transformation</dt>
<dd><t>Substitute a reference message for the actual transformed message at a boundary and compare downstream behavior, to examine the contribution of the message representation.</t></dd>
<dt>Test E: receiver-side interpretation</dt>
<dd><t>Supply the receiver an explicit reference condition and examine whether its proposed operation respects it, to separate receiver failure from upstream change.</t></dd>
</dl>
<t>Why injected-statement sensitivity does not prove the historical cause: a test shows that, under the tested context, the agent's behavior depends on the statement. The original incident may have had a different cause, several causes, a model or context that has since changed, or a statement that was never present. The finding is therefore classified as tested influence, and historical attribution requires separate, qualified support (see the evidence status set).</t>
</section>
<section anchor="witness">
<name>Diagnostic Witness Selection</name>
<t>Testing every factual value and instruction variation is infeasible. A selector chooses witnesses within a budget. This section is illustrative; it does not claim production superiority over any existing agent-debugging approach.</t>
<figure anchor="fig-witness">
<name>Adaptive witness selection loop</name>
<artwork type="ascii-art"><![CDATA[
   hypotheses H (incl. UNKNOWN) --+
                                   v
   +----> score feasible witnesses: gain / cost
   |               |
   |        pick best within budget & safety
   |               |
   |        run non-effectuating witness
   |               |
   |        validate observation, update H
   |               |
   +---- continue? <-- stop: threshold, budget, safety,
                            no utility, non-identifiable
                            --> finding OR abstention]]></artwork>
</figure>
<t>Assumptions: a finite hypothesis set including an explicit unknown hypothesis; a witness catalog with declared costs and the hypotheses each can discriminate; an outcome model (deterministic mapping, or likelihoods that may be uncertain); and a non-effectuating execution capability.</t>
<sourcecode type="pseudocode"><![CDATA[
select_and_run(H, W, budget):
  while true:
    feasible = [w for w in W if admissible(w) and cost(w) <= budget
                 and replay_ok(w) and safe(w)]
    if no feasible witness with positive utility: stop(NO_UTILITY)
    if remaining H observationally equivalent: stop(NONIDENT)
    w = argmax over feasible of utility(w, H) / cost(w)
    obs = run_non_effectuating(w)
    if not valid(obs): mark(w, INVALID); continue
    H = update(H, w, obs)        # Bayesian, or partition refinement
    budget -= cost(w)
    if decision_criterion_met(H): stop(SUPPORTED)
  report finding or ABSTAIN
]]></sourcecode>
<dl newline="true">
<dt>Deterministic variant</dt>
<dd><t>Hypotheses partitioned by witness outcome; select the witness that splits the partition best per cost; no probabilities required.</t></dd>
<dt>Stochastic variant</dt>
<dd><t>Expected information gain from a likelihood model per unit cost; may use robust selection when likelihoods are uncertain or a decision-loss criterion.</t></dd>
<dt>Stopping</dt>
<dd><t>Supported decision threshold, low remaining decision risk, no positive utility, budget exhausted, safety constraint, non-identifiability, or unresolved evidence.</t></dd>
</dl>
<t>The underlying filing describes a toy reference implementation (E08_TOY_REFERENCE_V1) using seven hypothesis states, five witnesses, manually specified likelihoods, and no remote model access. The filing itself states that its results do not establish empirical likelihood calibration, superiority over an existing agent-debugging system, production-quality replay isolation, resistance to a malicious evidence recorder, universal prompt-injection detection, complete multi-fault coverage, or guaranteed identification of a hidden historical cause. This document makes the same limits its own.</t>
</section>
<section anchor="compare">
<name>Behavioral Comparison</name>
<t>A response difference is not necessarily a change of interpretation, and model-based agents are stochastic. The comparator therefore reports an effect, an interval or other uncertainty representation, a status, the coverage of observations and the assumptions, per condition, handoff, witness and version context.</t>
<figure anchor="fig-compare">
<name>Statistical comparison of interpretations</name>
<artwork type="ascii-art"><![CDATA[
 paired observations --> normalize --> effect estimate d
                                        with interval [L, U]
   predeclared materiality threshold delta (set before data)

   L > delta  ........ material difference supported
   [L,U] in (-delta, delta]  practical equivalence supported
   otherwise ......... insufficient evidence
   pair invalid ...... invalid comparison (no conclusion)]]></artwork>
</figure>
<t>The materiality threshold is declared before inspecting results. Categories include supported material difference, supported practical equivalence, insufficient evidence and invalid comparison. Deterministic structured comparisons are used where outcomes are deterministic; fixed-sample or sequentially valid procedures where they are not. Adaptive test selection affects statistical validity and is accounted for.</t>
</section>
<section anchor="localization">
<name>Multi-Hop Localization</name>
<t>The localizer combines structural observations (where a condition's representation first differs) with intervention evidence (which transition responds to the tested change). It reports the earliest observable divergence boundary where boundary evidence suffices, a bounded interval where stages are unobserved, or an unresolved localization. A unique responsible agent is never mandatory.</t>
<figure anchor="fig-local">
<name>Earliest observable divergence over four hops</name>
<artwork type="ascii-art"><![CDATA[
 hop:      1        2        3        4
 C3:     [ok] --- [??] --- [bad] --- [bad]
          |        |         |
       observed  unobserved observed
 last good = hop 1; first bad = hop 3
 fault interval I = (hop 1, hop 3]   (hop 2 or 3 transition)
 No unique blame is reported when hop 2 cannot be observed.]]></artwork>
</figure>
<t>Illustrative decision matrix (from the filing's localization embodiment, paraphrased): sender input preserves the condition but sender output changes it suggests a sender-side representation difference; the converter's submitted payload preserves it but the receiver input changes it suggests a conversion interval; a condition-complete received message with a violating proposal suggests a receiver action-condition failure; different compliant decisions under different facts indicate legitimate factual sensitivity; an unobserved intermediate stage yields a bounded interval; invalid or inconclusive tests yield an unresolved diagnosis. The matrix does not establish a unique historical cause by itself.</t>
<section anchor="uncert">
<name>Evidence-Bounded Causal Localization and Diagnostic Uncertainty</name>
<t>Evidence tiers in the filing range from a candidate discrepancy, through validated structural divergence and controlled intervention effect, to stage-localized controlled effect and, last, historically supported contribution under explicit assumptions. A higher-numbered tier is not implied by a lower one: a controlled effect does not by itself imply historical causation.</t>
<ul>
<li>Abstain when tests are invalid, incomparable, statistically inconclusive, unsafe or unavailable.</li>
<li>Report a bounded interval, not a single agent, when intermediate transformations cannot be observed.</li>
<li>Report multiple faults, or masked faults, as possible; an earlier fault may be corrected or hidden by a later transformation.</li>
<li>Treat the hidden internal reasoning of an agent as unobservable from an external interface.</li>
<li>Do not convert a threshold crossing into scientific confidence; model validity is separate.</li>
</ul>
</section>
</section>
<section anchor="branching">
<name>Branching and Merging Workflows</name>
<t>Parallel branches and merges can hide, create or combine faults. The diagnosis classifies whether each condition applies to each branch, validates a consistent merge context, substitutes branch outputs, uses factorial branch interventions, and reports branch faults, merge faults, interactions and masked faults, with intervals where the branch-to-merge boundary is unobserved.</t>
<figure anchor="fig-branch">
<name>Parallel branches and merge</name>
<artwork type="ascii-art"><![CDATA[
            +--> Branch X (ok) -----+
 Task --> Split                      +--> Merge --> proposal
            +--> Branch Y (claim) --+
 Tests: substitute X out, Y out, both (factorial);
 compare merge output. Separate: branch fault / merge fault /
 interaction / masked fault.]]></artwork>
</figure>
</section>
<section anchor="adapters">
<name>Protocol-Independent Adapters</name>
<t>Adapters sit beside existing agent protocols, SDKs, gateways, tool servers and legacy interfaces. They normalize native events into canonical events, publish what they can observe and replay (a capability and coverage manifest), translate witnesses to native interfaces, and run diagnostic inputs non-effectuatingly. No change to A2A, MCP or any protocol is required, and no new mandatory message format is introduced.</t>
<figure anchor="fig-adapter">
<name>Protocol-independent adapters</name>
<artwork type="ascii-art"><![CDATA[
 A2A    MCP    SDK     Gateway   Legacy API   Future protocol
  |      |      |         |          |             |
 [adapter][adapter][adapter][adapter][adapter]  [adapter]
        \       \     |      /        /          /
         canonical event + capability + coverage records
                        |
              diagnostic processor (E..I)
 Protocols unchanged; adapters sit beside them.]]></artwork>
</figure>
<t>Legacy integration: for a legacy service or API (variation E12-V16 in the filing), the adapter retains the originating task, captures outgoing requests and incoming responses, associates conditions with request fields, classifies source responses and builds a request-response graph. Where only metadata is visible, coverage is marked partial; hidden reasoning cannot be reconstructed from an external interface.</t>
</section>
<section anchor="evidence">
<name>Diagnostic Evidence Structures</name>
<t>An optional, transport-neutral evidence model has six record types. The example fields below are illustrative only; this document defines no mandatory message format, wire encoding or registry.</t>
<figure anchor="fig-records"><name>Illustrative evidence records (non-normative)</name><artwork type="ascii-art"><![CDATA[TaskContractRecord
  { task_id, revision, conditions[typed], unresolved[] }
HandoffObservationRecord
  { task_id, hop, boundary, message_version,
    condition_status{}, coverage }
DiagnosticWitnessRecord
  { witness_id, pair, invariants, designated_diff,
    admissibility, executable }
DiagnosticObservationRecord
  { witness_id, run, normalized_response, validity,
    context_version }
DiagnosticFindingRecord
  { claim, tier_or_status, interval, coverage,
    assumptions, unresolved }
EvidenceContinuityRecord
  { prior_ref, supersedes, applicability, redactions,
    integrity (optional) }]]></artwork></figure>
<figure anchor="fig-export">
<name>Evidence export</name>
<artwork type="ascii-art"><![CDATA[
 observations --> evidence records --> claim support check
                                          |
   per-claim status: observed / controlled effect /
   bounded interval / historical (qualified) / unresolved
                                          |
   selective disclosure (redaction) --> export package
   optional authentication and integrity; none implies
   that the causal conclusion is conclusive.]]></artwork>
</figure>
<t>Evidence may be authenticated, integrity-protected and selectively disclosed, but none of these implies that a causal conclusion is conclusive. Statuses describe different properties rather than a single ascending scale: an authenticated record may contain a contradicted technical claim; a controlled test may support an effect without locating a hop; a historical observation may remain valid after its current applicability has expired.</t>
<ul>
<li>Statuses in the filing include directly observed, recorded provenance, inferred association, controlled effect supported, statistical effect supported, observable boundary supported, bounded interval supported, historical causation qualified, correction tested, regression tested, currently applicable, historical only, integrity verified, authenticated, redacted, withheld, unavailable, incomparable, unresolved and contradicted.</li>
</ul>
</section>
<section anchor="reval">
<name>Selective Revalidation</name>
<t>Facts, roles and models change between observation and use (a semantic time-of-check to time-of-use gap). Each finding carries the set of facts, roles, contract revisions and model versions it depends on. When one changes, only affected findings are marked as requiring revalidation and only affected witnesses are rerun; unaffected findings remain current; findings that cannot be replayed become historical only.</t>
<figure anchor="fig-toctou">
<name>Selective TOCTOU revalidation</name>
<artwork type="ascii-art"><![CDATA[
 fact v1 --- message --- proposal --- ... --- use time
   |                                             |
 fact v2 appears here ---------------------------+
 dependency set of each finding checked:
   unaffected -> remains CURRENT
   affected   -> REVALIDATION_REQUIRED (rerun only those)
   no replay  -> HISTORICAL_ONLY]]></artwork>
</figure>
</section>
<section anchor="correction">
<name>Correction and Regression Testing</name>
<t>An optional processor proposes corrections at the site the finding implicates (for example the converter rather than the receiver), validates them counterfactually, and runs regression witnesses: the original failure, boundary values, metamorphic relations and preservation of legitimate positive behavior. Outcomes may be partial, recurred or new regression, and are limited to the contexts tested.</t>
<figure anchor="fig-correct">
<name>Correction and regression verification</name>
<artwork type="ascii-art"><![CDATA[
 finding --> candidate correction --> counterfactual check
                                          |
        regression witnesses: original failure, boundary,
        metamorphic, positive-behavior preservation
                                          |
   outcomes: validated for tested context / partial /
             fault recurred / new regression / incomparable
 (A validated correction is evidence; deployment is separate.)]]></artwork>
</figure>
<t>A validated correction is diagnostic evidence about a repair. Whether a repair is deployed, and whether any operation proceeds, is decided elsewhere.</t>
</section>
<section anchor="threats">
<name>Trust and Threat Model</name>
<t>The model lists 24 threats. For each: attacker capability, trust boundary, observable effect, detection method, residual uncertainty and safeguards. Evidence may be authenticated and selectively disclosed without implying causal conclusiveness.</t>
<dl newline="true">
<dt>T1. Instruction-like content in a factual field</dt>
<dd><t>Attacker capability: supplier or tool output embeds a modification claim.</t><t>Trust boundary: source to sender.</t><t>Observable effect: sender condition changes.</t><t>Detection method: matched instruction-source test (Test B).</t><t>Residual uncertainty: model may change; test context may not match the incident.</t><t>Safeguards: condition-scoped roles; non-effectuating tests.</t></dd>
</dl>
<dl newline="true">
<dt>T2. Authority laundering by an intermediary</dt>
<dd><t>Attacker capability: trusted intermediary restates a claim as its own.</t><t>Trust boundary: intermediary.</t><t>Observable effect: derived message gains eligibility.</t><t>Detection method: role carry-through check; lineage.</t><t>Residual uncertainty: intermediary hidden.</t><t>Safeguards: eligibility never increases by derivation.</t></dd>
</dl>
<dl newline="true">
<dt>T3. Quotation that becomes instruction</dt>
<dd><t>Attacker capability: quoted text is treated as governing.</t><t>Trust boundary: sender to receiver.</t><t>Observable effect: receiver adopts claim.</t><t>Detection method: message-replacement test (Test D).</t><t>Residual uncertainty: interpretation ambiguity.</t><t>Safeguards: segment-level labeling.</t></dd>
</dl>
<dl newline="true">
<dt>T4. Persistent-memory authority laundering</dt>
<dd><t>Attacker capability: stored note replays an old claim.</t><t>Trust boundary: memory to agent.</t><t>Observable effect: claim reappears as condition.</t><t>Detection method: taint labels; temporal tests.</t><t>Residual uncertainty: memory not observable.</t><t>Safeguards: versioned memory records.</t></dd>
</dl>
<dl newline="true">
<dt>T5. Multilingual transformation</dt>
<dd><t>Attacker capability: translation alters a limit.</t><t>Trust boundary: converter.</t><t>Observable effect: condition changes meaning.</t><t>Detection method: structural projection per language.</t><t>Residual uncertainty: translation ambiguity.</t><t>Safeguards: typed values and units.</t></dd>
</dl>
<dl newline="true">
<dt>T6. Summarization omission</dt>
<dd><t>Attacker capability: summary drops a condition.</t><t>Trust boundary: sender output.</t><t>Observable effect: condition absent.</t><t>Detection method: lineage divergence.</t><t>Residual uncertainty: summary unobserved.</t><t>Safeguards: condition presence checks at each boundary.</t></dd>
</dl>
<dl newline="true">
<dt>T7. Converter field rewrite</dt>
<dd><t>Attacker capability: payload converter alters a field.</t><t>Trust boundary: submitted to received.</t><t>Observable effect: argument differs.</t><t>Detection method: submitted-versus-received comparison.</t><t>Residual uncertainty: converter hidden.</t><t>Safeguards: adapter at converter boundary.</t></dd>
</dl>
<dl newline="true">
<dt>T8. Receiver misinterpretation</dt>
<dd><t>Attacker capability: receiver restates rule but acts otherwise.</t><t>Trust boundary: receiver.</t><t>Observable effect: non-compliant proposal.</t><t>Detection method: reference-condition test (Test E).</t><t>Residual uncertainty: black-box receiver.</t><t>Safeguards: correspondence checks.</t></dd>
</dl>
<dl newline="true">
<dt>T9. Stale factual input</dt>
<dd><t>Attacker capability: old fact used after update.</t><t>Trust boundary: fact store to agent.</t><t>Observable effect: wrong decision.</t><t>Detection method: version-dependency check.</t><t>Residual uncertainty: clock or version skew.</t><t>Safeguards: bitemporal records.</t></dd>
</dl>
<dl newline="true">
<dt>T10. Evidence recorder tampering</dt>
<dd><t>Attacker capability: attacker edits observation records.</t><t>Trust boundary: recorder.</t><t>Observable effect: false finding.</t><t>Detection method: integrity checks where available.</t><t>Residual uncertainty: malicious recorder not excluded.</t><t>Safeguards: signed or chained records (optional).</t></dd>
</dl>
<dl newline="true">
<dt>T11. Witness poisoning</dt>
<dd><t>Attacker capability: adversary shapes test inputs.</t><t>Trust boundary: witness generator.</t><t>Observable effect: test misleads.</t><t>Detection method: invariant validation; admissibility.</t><t>Residual uncertainty: novel manipulation.</t><t>Safeguards: predeclared invariants.</t></dd>
</dl>
<dl newline="true">
<dt>T12. Test-aware agent behavior</dt>
<dd><t>Attacker capability: agent behaves differently when tested.</t><t>Trust boundary: agent.</t><t>Observable effect: tests unrepresentative.</t><t>Detection method: varied and repeated witnesses.</t><t>Residual uncertainty: detection of test awareness limited.</t><t>Safeguards: randomization; coverage reporting.</t></dd>
</dl>
<dl newline="true">
<dt>T13. Model drift between incident and test</dt>
<dd><t>Attacker capability: model updated after incident.</t><t>Trust boundary: agent.</t><t>Observable effect: tests not comparable.</t><t>Detection method: version stratification.</t><t>Residual uncertainty: unknown versions.</t><t>Safeguards: incomparable status; abstention.</t></dd>
</dl>
<dl newline="true">
<dt>T14. Replay side effects</dt>
<dd><t>Attacker capability: diagnostic run causes real effects.</t><t>Trust boundary: adapter.</t><t>Observable effect: unintended action.</t><t>Detection method: non-effectuating execution controller.</t><t>Residual uncertainty: misconfigured isolation.</t><t>Safeguards: separate deployment; sandboxing; safety stop.</t></dd>
</dl>
<dl newline="true">
<dt>T15. Evidence confidentiality</dt>
<dd><t>Attacker capability: diagnostic export reveals sensitive content.</t><t>Trust boundary: exporter.</t><t>Observable effect: disclosure.</t><t>Detection method: redaction; minimization.</t><t>Residual uncertainty: residual inference.</t><t>Safeguards: selective disclosure; purpose limits.</t></dd>
</dl>
<dl newline="true">
<dt>T16. Adaptive-selection bias</dt>
<dd><t>Attacker capability: selector steers toward a favored cause.</t><t>Trust boundary: selector.</t><t>Observable effect: biased conclusion.</t><t>Detection method: statistical accounting for adaptivity.</t><t>Residual uncertainty: model misspecification.</t><t>Safeguards: predeclared thresholds; unknown hypothesis.</t></dd>
</dl>
<dl newline="true">
<dt>T17. Multiple concurrent faults</dt>
<dd><t>Attacker capability: several faults interact.</t><t>Trust boundary: path.</t><t>Observable effect: single-blame error.</t><t>Detection method: factorial tests; multi-fault assessment.</t><t>Residual uncertainty: combinatorial limits.</t><t>Safeguards: no mandatory unique blame.</t></dd>
</dl>
<dl newline="true">
<dt>T18. Masked upstream fault</dt>
<dd><t>Attacker capability: later step hides earlier fault.</t><t>Trust boundary: path.</t><t>Observable effect: earlier fault missed.</t><t>Detection method: restoration tests.</t><t>Residual uncertainty: unobserved stages.</t><t>Safeguards: intervals; abstention.</t></dd>
</dl>
<dl newline="true">
<dt>T19. Parallel-branch interaction</dt>
<dd><t>Attacker capability: branches jointly cause merge fault.</t><t>Trust boundary: merge.</t><t>Observable effect: wrong attribution.</t><t>Detection method: branch factorial design.</t><t>Residual uncertainty: scale.</t><t>Safeguards: interaction reporting.</t></dd>
</dl>
<dl newline="true">
<dt>T20. Missing observation boundary</dt>
<dd><t>Attacker capability: an interface is not exposed.</t><t>Trust boundary: adapter.</t><t>Observable effect: no evidence.</t><t>Detection method: coverage vector.</t><t>Residual uncertainty: irreducible.</t><t>Safeguards: explicit unavailable status.</t></dd>
</dl>
<dl newline="true">
<dt>T21. Identity spoofing of source</dt>
<dd><t>Attacker capability: source impersonated.</t><t>Trust boundary: source.</t><t>Observable effect: wrong role.</t><t>Detection method: authentication outside this document.</t><t>Residual uncertainty: depends on identity layer.</t><t>Safeguards: separate authentication; role not inferred from content.</t></dd>
</dl>
<dl newline="true">
<dt>T22. Replay or reorder of evidence</dt>
<dd><t>Attacker capability: old evidence resubmitted.</t><t>Trust boundary: exporter.</t><t>Observable effect: stale finding accepted.</t><t>Detection method: continuity record; applicability class.</t><t>Residual uncertainty: depends on freshness source.</t><t>Safeguards: supersession records.</t></dd>
</dl>
<dl newline="true">
<dt>T23. Over-trust in an authenticated record</dt>
<dd><t>Attacker capability: record signed yet contradicted.</t><t>Trust boundary: consumer.</t><t>Observable effect: false assurance.</t><t>Detection method: claim-level status.</t><t>Residual uncertainty: human misreading.</t><t>Safeguards: status separate from authentication.</t></dd>
</dl>
<dl newline="true">
<dt>T24. Cost exhaustion</dt>
<dd><t>Attacker capability: attacker forces many tests.</t><t>Trust boundary: selector.</t><t>Observable effect: denial of diagnosis.</t><t>Detection method: budget and stopping rules.</t><t>Residual uncertainty: degraded coverage.</t><t>Safeguards: budget caps; mode selection.</t></dd>
</dl>
</section>
<section anchor="security">
<name>Security Considerations</name>
<t>This architecture reads message content and runs agents on altered inputs. Diagnostic execution <bcp14>MUST</bcp14> be non-effectuating where the architecture is claimed to be used as described; an implementation that cannot guarantee that <bcp14>SHOULD NOT</bcp14> execute the witness and <bcp14>SHOULD</bcp14> retain the structural definition only. Evidence integrity mechanisms are optional and do not make findings conclusive. Diagnostic findings <bcp14>MUST NOT</bcp14> be presented as authorization decisions.</t>
<t>Security considerations in general follow <xref target="RFC3552"/>. Authentication, authorization and execution control are out of scope and remain with other systems.</t>
</section>
<section anchor="privacy">
<name>Privacy Considerations</name>
<t>Diagnostic records may contain instruction text, facts, personal data and trade secrets. Implementations <bcp14>SHOULD</bcp14> minimize retained content, prefer condition-level projections to raw content, support redaction and purpose limitation, and consider <xref target="RFC6973"/>. Selective disclosure reduces evidential support and the reduction should be visible in the record. Nothing here is a statement of compliance with any privacy law.</t>
</section>
<section anchor="interop">
<name>Interoperability Considerations</name>
<t>The architecture requires no change to A2A, MCP or other protocols. Interoperability is at the level of optional diagnostic findings and the evidence model; organizations may keep their own authority policies while exchanging compatible findings. Heterogeneous, asynchronous, streaming and partially observable agents are expected; coverage must be recorded rather than assumed.</t>
</section>
<section anchor="perf">
<name>Performance and Evaluation</name>
<t>Three operating modes are distinguished. No universal latency claim is made.</t>
<table>
<name>Operating modes</name>
<thead><tr><th>Mode</th><th>What it does</th><th>Expected role</th></tr></thead>
<tbody>
<tr><td>A. Lightweight observation</td><td>Captures task identifiers, message versions, dependency links and deterministic structured-condition checks</td><td>Always on candidate; cheap; not equivalent to behavioral testing</td></tr>
<tr><td>B. Event-triggered diagnosis</td><td>Runs witnesses and comparison when a discrepancy or lower-trust influence is suspected</td><td>On demand within a budget</td></tr>
<tr><td>C. Offline or post-incident</td><td>Full counterfactual, localization and regression analysis</td><td>Background or incident review</td></tr>
</tbody>
</table>
<t>Design objectives are not measurements. The filing lists illustrative objectives only (for example low milliseconds for tagging and structured comparison; model- and network-dependent time for counterfactual invocation). A credible implementation would report p50, p95, p99 and p99.9 latency, model invocation counts, diagnostic cost and the rate of unresolved findings, and distinguish objectives from benchmark results. The diagnostic path is separate from any fast path in which existing authorization and business rules decide whether an operation runs; a full causal test inside every transaction or protection loop is not assumed.</t>
</section>
<section anchor="iana">
<name>IANA Considerations</name>
<t>This document has no IANA actions.</t>
</section>
<section anchor="prior">
<name>Relationship to Existing Work</name>
<t>This section is a non-normative comparison by the author and makes no novelty conclusion. Semantic delegation, signed handoffs and observability are existing areas and are not claimed as new here; the distinct focus is counterfactual, uncertainty-aware, condition-scoped diagnosis. Reading was limited as stated in the source verification appendix.</t>
<dl newline="true">
<dt>Agentic AI Use Cases and Requirements <xref target="I-D.agentic-ai-usecases-requirements"/></dt>
<dd><t>Covers: use cases for agentic AI communication systems and derived protocol requirements; later revisions exist (only search excerpts were read).</t><t>Relation: provides use-case context; this document defines no protocol requirements and is a diagnostic complement.</t></dd>
<dt>Signed Handoff Context for Agentic Dialogs <xref target="I-D.srivastava-agentproto-handoff-context"/></dt>
<dd><t>Covers: signed offer, signed status and hash-linked chain; it says a digest does not say what an object means, states that semantic preservation is not addressed, and grants no authority.</t><t>Relation: complementary. A signed chain shows order and consistency; this document asks, by tests, whether a condition's meaning survived.</t></dd>
<dt>Semantic Governance Requirements for AI Agent Delegation and Handoff <xref target="I-D.luna-eagf-agent-delegation-semantics"/></dt>
<dd><t>Covers: requirements REQ-1 to REQ-8 including bounded authority scope and material context preservation; no new protocol; does not define detection or counterfactual diagnosis.</t><t>Relation: complementary. It says what must be preserved; this document offers a way to diagnose whether a condition was preserved.</t></dd>
<dt>A2A and MCP</dt>
<dd><t>Covers: agent-to-agent and model-to-tool interaction protocols (details not re-verified here).</t><t>Relation: adapters observe them without changing them.</t></dd>
<dt>OpenTelemetry and distributed tracing</dt>
<dd><t>Covers: telemetry and span correlation.</t><t>Relation: supplies ordering; does not test condition influence.</t></dd>
<dt>Provenance, evaluation, prompt-injection and observability work</dt>
<dd><t>Covers: records, benchmarks, detectors and monitoring.</t><t>Relation: different function; this adds matched counterfactual tests separating fact from instruction by source role.</t></dd>
</dl>
<t>The same author's separate execution-finality documents concern authority over effects. This document intentionally keeps causal diagnosis separate from execution finality and does not rely on them.</t>
</section>
<section anchor="industry">
<name>Non-Normative Industry Matrix</name>
<t>Non-normative and hypothetical. This mapping is AI-generated and was NOT verified against vendors' official documentation in preparing this document; product names and capabilities may be wrong or out of date. It must be cross-checked against official documentation and the original filings before any conclusion is drawn. It is not an endorsement, validation, adoption, infringement or deficiency statement, and no company has reviewed it.</t>
<dl newline="true">
<dt>Google</dt>
<dd><t>1. Existing technology (to be verified): A2A, Agent Development Kit, Gemini Enterprise Agent Platform.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): workflow-event interface or A2A client/server middleware.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>Microsoft</dt>
<dd><t>1. Existing technology (to be verified): Agent Framework workflow, handoff and middleware layers.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): middleware and workflow event hooks.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>OpenAI</dt>
<dd><t>1. Existing technology (to be verified): agent handoff callbacks, input filters, tool wrappers, tracing processors.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): handoff input-transformation hooks and tool wrappers.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>Anthropic</dt>
<dd><t>1. Existing technology (to be verified): tool-result processing, subagent delegation, MCP interfaces.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): tool-result normalization and approved MCP interfaces.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>Amazon</dt>
<dd><t>1. Existing technology (to be verified): cloud agent runtime and tool gateway services.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): gateway or runtime event stream.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>Salesforce</dt>
<dd><t>1. Existing technology (to be verified): enterprise agent platform actions and data connectors.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): connector and action-request layer.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>IBM</dt>
<dd><t>1. Existing technology (to be verified): enterprise orchestration and governance tooling.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): orchestration event interface.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>NVIDIA</dt>
<dd><t>1. Existing technology (to be verified): agent toolkits and inference microservices.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): pipeline and microservice boundaries.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>Mistral</dt>
<dd><t>1. Existing technology (to be verified): hosted agents and function-calling APIs.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): API request/response boundary.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>Aleph Alpha</dt>
<dd><t>1. Existing technology (to be verified): sovereign enterprise AI deployments.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): private deployment boundary.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>SAP</dt>
<dd><t>1. Existing technology (to be verified): business-process AI agents inside enterprise applications.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): business-process event layer.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>Siemens</dt>
<dd><t>1. Existing technology (to be verified): industrial AI and engineering assistants.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): supervisory or planning interface.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>Ericsson</dt>
<dd><t>1. Existing technology (to be verified): telecom network AI and automation.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): network-management agent interface.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>Nokia</dt>
<dd><t>1. Existing technology (to be verified): telecom and network automation agents.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): management-plane agent interface.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>Alibaba</dt>
<dd><t>1. Existing technology (to be verified): cloud agent platforms and model services.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): service gateway.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>Tencent</dt>
<dd><t>1. Existing technology (to be verified): cloud agent platforms and messaging-integrated assistants.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): platform middleware.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
<dl newline="true">
<dt>DeepSeek, Doubao, Kimi and GLM environments</dt>
<dd><t>1. Existing technology (to be verified): model-API based agent frameworks.</t><t>2. Present limitation addressed: valid transport, task state and signatures do not show that each governing condition survived each handoff.</t><t>3. Where an adapter could attach (hypothetical): API client wrapper.</t><t>4. What would be observed: originating instruction, transformed message versions, source segments and tool results, proposed operation fields.</t><t>5. What would be tested: matched fact and instruction-claim variants in a non-effectuating copy; the sender output, converter payload and receiver input compared per condition.</t><t>6. What would be reported: condition-preservation status per boundary, tested influence with interval, earliest observable divergence or bounded interval, and abstention where unsupported.</t><t>7. What would not be done: nothing is approved, rejected or blocked by the adapter; the platform's own authorization, guardrails and approvals keep deciding.</t><t>8. Residual limits: hidden reasoning is not reconstructable; black-box or metadata-only components yield partial coverage; replay may be unavailable.</t></dd>
</dl>
</section>
<section anchor="europe">
<name>Trustworthy Multi-Agent AI, European Digital Sovereignty, and Diagnostic Accountability</name>
<t>Non-normative. Nothing here is a claim that any implementation is legally compliant because it produces diagnostic evidence, and no geopolitical intent or wrongdoing is attributed to any country or vendor.</t>
<ul>
<li>EU AI Act (Regulation (EU) 2024/1689) <xref target="EU-AIA"/>: provisions on logging (Article 12), human oversight (Article 14) and accuracy, robustness and cybersecurity (Article 15) are areas where located, evidence-bounded diagnostic findings may be one input among many; applicability depends on the system and role and is not assessed here.</li>
<li>GDPR: data minimization and purpose limitation point toward condition-level projections and redaction over raw retention; no compliance statement is made.</li>
<li>NIS2 <xref target="EU-NIS2"/>: incident investigation across cross-border enterprise systems may benefit from a shared finding form; no compliance statement is made.</li>
<li>Sovereign-cloud and private deployments: adapters can run inside a jurisdiction's boundary; organizations keep their own authority policies while exchanging compatible findings.</li>
<li>Sectors: European industrial AI (including SAP and Siemens environments), telecommunications and critical infrastructure, banking and financial-service AI, and cloud incident investigation.</li>
<li>Without certainty about hidden reasoning, a located and bounded finding supports comparison of facts between operators and authorities without presupposing fault.</li>
</ul>
</section>
<section anchor="examples">
<name>Worked Examples</name>
<dl newline="true">
<dt>Example A: financial risk limit</dt>
<dd><t>A treasury agent chain has an approved exposure limit. A market feed's commentary says the limit is raised; an intermediate agent treats it as a revision. The diagnostic varies the market facts and the unsupported statement separately and compares proposed adjustments; it may report tested influence and the divergence boundary. Non-effectuating operations: simulated order proposals only. The treasury authorization system decides whether any real trade happens.</t></dd>
<dt>Example B: cross-provider A2A handoff</dt>
<dd><t>Agent X on one platform delegates to agent Y on another over A2A. Y proposes a purchase above X's limit. An adapter records the submitted and received messages; if the limit was present at submission and absent at receipt, a conversion interval is reported; if present at receipt, a receiver test is run. Non-effectuating operations: replayed proposals to a stub. Authorization remains with each party.</t></dd>
<dt>Example C: MCP tool-output prompt injection</dt>
<dd><t>A retrieval tool returns a price and the sentence that the budget was raised. Only the sentence is varied. A material difference supports tested sensitivity; it does not prove that this sentence caused the earlier incident. Non-effectuating operations: replay without calling the real booking tool.</t></dd>
<dt>Example D: parallel agents merge</dt>
<dd><t>Two research branches are each correct, the merge promotes a claim from one. Branch substitution and a factorial design separate branch from merge from interaction faults. Non-effectuating operations: merge replay in isolation.</t></dd>
<dt>Example E: banking legacy adapter</dt>
<dd><t>A legacy payment API is fronted by an agent. The adapter captures request fields and responses and associates the limit with the amount field; it reports a request-field divergence. It cannot see inside the legacy system; coverage is partial. Non-effectuating operations: requests to a test environment only.</t></dd>
<dt>Example F: European industrial maintenance waiver</dt>
<dd><t>A maintenance planning agent receives a supplier note that a safety interval waiver is approved. The note is not from an eligible role. Tests vary the note; findings go to engineers. Non-effectuating operations: planning simulation only; certified safety systems and humans decide everything physical.</t></dd>
</dl>
</section>
<section anchor="validation">
<name>Validation Plan</name>
<t>No experimental results are reported in this document. The following is a plan. The underlying filing reports only toy reference implementations with synthetic harnesses (E08 described in detail; the filing indicates similar runnable references for later embodiments, not examined here), which demonstrate that the described control flow can run as software and do not establish real-world performance.</t>
<ul>
<li>Baselines: fixed test suite; random witness choice; final-outcome comparison; plain tracing with heuristic blame; prompt-injection detector.</li>
<li>Metrics: localization accuracy against injected known faults, abstention rate, false-attribution rate, test count and cost, p50/p95/p99/p99.9 latency, rate of unresolved findings.</li>
<li>Test cases: seven or more known fault modes (source influence, sender omission, converter change, receiver failure, stale fact, combinations, none) across single and multi-hop, branch and legacy cases.</li>
<li>Ablations: remove matched controls; remove adaptive selection; remove uncertainty intervals; remove source roles; remove selective revalidation.</li>
<li>Reporting: separate design objectives from benchmark results; state model versions; publish negative and abstention results.</li>
</ul>
</section>
<section anchor="ipr">
<name>IPR Disclosure, Attribution and Notices</name>
<t>Source filing: Indian Provisional Application 202631134241, dated 8 November 2026 <xref target="IN-134241"/> (as reported by the applicant; not a certified registry record). Foundational reference: PCT/IB2026/055615 <xref target="PCT-055615"/>, filed 4 June 2026, published 16 July 2026 as WO 2026/150382 A2. A separate Indian application, 202631131650 (3 November 2026, evaluator governance), is a distinct subject and is kept distinct from this document.</t>
<t>Public cross-references: https://doi.org/10.5281/zenodo.24076231 <xref target="ZENODO-CAUSAL"/>; Technical Disclosure Commons records <xref target="TDC-12934"/> and <xref target="TDC-12973"/> and <xref target="TDC-12542"/>.</t>
<t>This document contains no licensing commitment and no statement about royalties or essentiality. Contributors are bound by BCP 78 and BCP 79 (<xref target="RFC5378"/>, <xref target="RFC8179"/>). Patents or applications mentioned here are not a substitute for formal IETF IPR disclosure. 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. The IETF IPR database must be checked independently.</t>
<t>Source integrity: the filing text supplied is a 1,657-page PDF; the editor read the front disclosure, section headings of all embodiments, and selected sections in detail, not every page. The structured JSON technical disclosure was not available. Claims were not reconstructed. Verify against the official filed copy.</t>
<section>
<name>Technical Source and AI-Assistance Notice</name>
<t>The underlying causal semantic handoff invention and associated patent specification are attributed to human inventor Sangam Kumar Das. The present Internet-Draft explanatory text, keyword index, flowchart descriptions, research comparisons and industry relevance mapping may have been prepared with AI assistance. Such material can contain omissions or errors and must be verified against the original filed patent documents, public technical sources, and applicable IETF specifications.</t>
</section>
<section>
<name>Non-Endorsement Notice</name>
<t>This document is an individual technical proposal. It does not establish IETF standardization, European Union regulatory approval, certification, sponsorship, adoption, implementation, commercial affiliation or endorsement by any company, government, standards organization or research institution. Company names and product references identify potential research or integration environments only. No copying, infringement, technological deficiency or unauthorized use is alleged.</t>
</section>
<section>
<name>Technical Scope Notice</name>
<t>The diagnostic architecture supplies evidence about observable instruction interpretation changes. It does not independently authorize payments, execute network configuration changes, approve industrial operations, replace certified safety systems or guarantee that a historical cause has been uniquely identified.</t>
</section>
<section>
<name>Legal and Priority Notice</name>
<t>The filing dates and public-disclosure references are provided for traceability. The patentability, priority entitlement and legal scope of individual technical features require assessment against the official filed documents and applicable law.</t>
</section>
</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="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="I-D.agentic-ai-usecases-requirements" target="https://datatracker.ietf.org/doc/draft-agentic-ai-usecases-requirements/"><front><title>Agentic AI Use Cases and Requirements</title><author initials="T." surname="Reddy"/><author initials="Z." surname="Sarker"/><author initials="K." surname="Yao"/><date year="2026" month="May"/></front><seriesInfo name="Internet-Draft" value="draft-agentic-ai-usecases-requirements"/></reference>
<reference anchor="I-D.srivastava-agentproto-handoff-context" target="https://datatracker.ietf.org/doc/draft-srivastava-agentproto-handoff-context/"><front><title>Signed Handoff Context for Agentic Dialogs</title><author initials="K." surname="Srivastava"/><date year="2026" month="October"/></front><seriesInfo name="Internet-Draft" value="draft-srivastava-agentproto-handoff-context-00"/></reference>
<reference anchor="I-D.luna-eagf-agent-delegation-semantics" target="https://datatracker.ietf.org/doc/draft-luna-eagf-agent-delegation-semantics/"><front><title>Semantic Governance Requirements for AI Agent Delegation and Handoff</title><author initials="D." surname="Luna"/><date year="2026" month="October"/></front><seriesInfo name="Internet-Draft" value="draft-luna-eagf-agent-delegation-semantics-00"/></reference>
<reference anchor="ZENODO-CAUSAL" target="https://doi.org/10.5281/zenodo.24076231"><front><title>Trustworthy Multi-Agent AI for Europe: Causal Detection of Instruction Drift, Prompt-Injection Influence, and Communication-Hop Failures Through Counterfactual Diagnostics</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026" month="November" day="9"/></front><seriesInfo name="DOI" value="10.5281/zenodo.24076231"/><refcontent>Zenodo, version v1. Relevance: public research description of the same causal diagnostic subject matter. Not fetched by the editor; verify.</refcontent></reference>
<reference anchor="TDC-12934" target="https://www.tdcommons.org/dpubs_series/12934/"><front><title>When AI Agents Change Instructions... (title truncated as supplied)</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Technical Disclosure Commons. Not fetched; verify.</refcontent></reference>
<reference anchor="TDC-12973" target="https://www.tdcommons.org/dpubs_series/12973/"><front><title>Technical Disclosure Commons record 12973 (title not verified)</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Technical Disclosure Commons. Not fetched; verify.</refcontent></reference>
<reference anchor="TDC-12542" target="https://www.tdcommons.org/dpubs_series/12542/"><front><title>Mothership PCT disclosure record (title not verified)</title><author fullname="Sangam Das" initials="S." surname="Das"/><date year="2026"/></front><refcontent>Technical Disclosure Commons. Not fetched; verify.</refcontent></reference>
<reference anchor="PCT-055615" target="https://patentscope.wipo.int/"><front><title>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 on PATENTSCOPE.</refcontent></reference>
<reference anchor="IN-134241" target="https://ipindia.gov.in/"><front><title>Indian Provisional Patent Application 202631134241</title><author fullname="Sangam Kumar Das" initials="S." surname="Das"/><date year="2026" month="November" day="8"/></front><refcontent>As reported by the applicant. Not a certified registry record.</refcontent></reference>
</references>
<section anchor="app-emb">
<name>Appendix A: All 18 Embodiments</name>
<t>The source filing is summarized for each of its 18 embodiments in the author's words, based on the filing's own text; the filing should be consulted for exact wording. E12-V16 (legacy service and API-based task automation) is a deployment variation under E12 and not a nineteenth embodiment. None of the embodiments grants or vetoes execution authority.</t>
<section anchor="emb-e01">
<name>E01: Typed Originating Task-Contract Construction</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>An originating instruction is free text; later agents summarize, translate or split it, and conditions (limits, prerequisites, deadlines, prohibitions) are lost or altered. Without a machine-testable reference there is nothing precise to compare later representations against.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Attributed task instruction and revision; message and source identifiers; optional structured fields, voice transcript or document-derived text. Data structure: versioned Originating Task Contract of typed predicates (operator, value, unit, applicable operation field) plus unresolved-condition records.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Segment the instruction into normative conditions, factual references, proposed amendments, negations, temporal constraints and unresolved expressions; normalize typed predicates; link predicates to factual dependencies; record ambiguity instead of guessing; store a versioned contract.</t></dd>
<dt>4. Outputs</dt>
<dd><t>A versioned typed task-contract record that distinguishes testable from unresolved conditions, usable as the reference for later diagnostic comparison.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Needs a source attribution for the instruction and a revision identifier. Feeds E02, E03, E05, E07 and E13.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: contract digest or signed record binding the contract to its revision. Non-cryptographic: versioned application records and process-local identifiers. The filing states that both are supported.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>A condition with a missing material attribute (for example no currency) is rejected or marked unresolved; the contract does not establish complete semantic equivalence of arbitrary natural language.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Gives independent implementations a common, protocol-neutral reference for what the task required, without mandating a message format.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Transport booking only after goods reach a destination, before a deadline, and within INR 2,000: three conditions with different operators are held as separate typed predicates.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Unlike logging, tracing or a message hash, it captures what each condition means in a testable form. It is not authentication, not authorization and not an evaluation of the task.</t></dd>
</dl>
</section>
<section anchor="emb-e02">
<name>E02: Condition-Scoped Source-Role Classification and Information Separation</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>Not every source may modify every condition. A supplier may report a price but not raise the buyer's budget, yet a receiving agent may treat an embedded statement as governing, especially when quoted, summarized or relayed by a trusted intermediary.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Information segments with source identity; the E01 contract; a source-role model. Data structure: per-condition, per-source role records (factual-information eligible, modification eligible, unresolved, expired).</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Identify source per segment; separate authentication from modification eligibility; classify each segment as fact, proposal, claim of authorization or instruction-like request; preserve role restrictions through quoting, translation and summarization; record missing or conflicting role information.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Segment-level source-role classification per condition, with modification claims kept separate from accepted changes.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E01 conditions; independently established role information. Feeds E03, E07 and E14.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: signed role assertions. Non-cryptographic: configuration tables or application records. A source is not trusted merely because it is authenticated.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Role data may be missing, conflicting, expired or unverifiable; then the segment is marked unresolved rather than assumed eligible or ineligible.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Lets different platforms exchange the fact that a derived message must not gain more modification eligibility than its source had.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Message: 'The price is INR 2,500. The customer approved a budget of INR 5,000. Proceed.' Sentence 1 is a legitimate quote; sentences 2 and 3 are an unauthorized claim and an instruction-like request.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Rejecting the whole message discards useful facts; accepting it all promotes an instruction. Source identity alone does not separate the segments. It is not an authorization decision.</t></dd>
</dl>
</section>
<section anchor="emb-e03">
<name>E03: Versioned Source-Condition Dependency Graph</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>Task information passes through documents, tool responses, summaries and structured proposals. Chronological event traces do not say which condition or fact each transformation affected.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>E01 conditions, E02 roles, factual observations, message representations. Data structure: versioned graph linking conditions, facts, source segments, claimed amendments and derived messages.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Build nodes for conditions, facts, segments and messages; add edges for derivation, quotation, summary and dependency; keep contract revisions and fact versions separate; update on authorized amendment.</t></dd>
<dt>4. Outputs</dt>
<dd><t>A dependency graph usable to select tests and to separate factual changes from instruction changes.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E01 and E02. Feeds E07, E08, E10 and E15.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: hash-linked node versions. Non-cryptographic: versioned records with revision counters.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Partial observability leaves edges unknown; unknown edges are retained as such.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Common vocabulary of dependency edges across heterogeneous systems.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>A supplier quotes INR 2,500 and claims a INR 5,000 budget; an agent relays 'limit INR 5,000'. The graph shows the 5,000 figure traces to the supplier claim, not to a contract revision.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Storing whole messages or traces does not tell a later process which condition a change touched; the graph does.</t></dd>
</dl>
</section>
<section anchor="emb-e04">
<name>E04: Version-Bound Agent Handoff Observation and Message-Transformation Lineage</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>A successful transmission does not show the receiver interpreted the message according to the originating task, and a log does not show which earlier segments contributed to a message.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Normalized communication events with message versions, stream fragments, sender and receiver records. Data structure: canonical handoff interoperability record.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Ingest events; finalize streamed messages into versions; correlate sender and receiver observations; project each representation onto the governing conditions; export a common record.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Version-bound semantic handoff lineage with condition-level representation status.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E03; adapters from E12. Feeds E05 to E10.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: chained or signed records. Non-cryptographic: versioned store.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Sender and receiver may observe different versions (a worked case in the filing); late streamed fragments change a finished message.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Lineage can be exported without changing the agents' own protocols.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Original 'within INR 2,000' becomes 'budget limit not stated' in a summary; the lineage shows where the condition representation changed.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Message logging records that a message existed; lineage binds versions to sources and conditions.</t></dd>
</dl>
</section>
<section anchor="emb-e05">
<name>E05: Condition-Scoped Instruction Fingerprinting and Behavioral Evidence Association</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>A message digest identifies bytes, not operational meaning; an agent may restate a rule correctly yet act differently under certain facts.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Typed conditions; message representations; observed behavior. Data structure: multi-level fingerprint (textual, condition-projected, behavioral).</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Canonicalize typed conditions; project representations onto a condition; compare at three levels; associate behavioral evidence with source context; invalidate records when versions change.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Condition-scoped fingerprints and comparison results at textual, structural and behavioral levels.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E01, E04. Feeds E06 and E09.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: digests of canonical forms. Non-cryptographic: structural keys.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Equal fingerprints do not prove equal behavior; different wording may map to the same condition.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Compact identifiers for condition state that systems may compare without sharing full content.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Two differently worded instructions with the same INR 2,000 limit match structurally; an agent that states the limit but proposes INR 2,500 differs behaviorally.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Cryptographic equivalence and semantic equivalence are different properties; this embodiment keeps them separate.</t></dd>
</dl>
</section>
<section anchor="emb-e06">
<name>E06: Version-Bound Diagnostic Observation Evidence</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>Test results are only meaningful if the test context was reproducible and the pair was valid; otherwise a difference may come from a changed model or context.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Witness manifests, non-effectuating observations, response records. Data structure: evidence manifest and observation package with state model.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Construct manifests; acquire observations non-effectuatingly; validate matched-pair invariants; normalize responses; validate packages; assess impact of version changes.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Validated or incompatible observation packages with explicit state.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E05, E07. Feeds E08 to E10 and E18.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: signed observation records. Non-cryptographic: local digests (the filing notes a locally computed record digest in its toy implementation).</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Incompatible pairs, changing model or persistent agent state, and stale facts make evidence incomparable.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>A common observation record lets third parties judge whether two results are comparable.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>A pair in which the factual price changed unintentionally is rejected before it can support any conclusion.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Ordinary test logs do not enforce matched-pair invariants or version context.</t></dd>
</dl>
</section>
<section anchor="emb-e07">
<name>E07: Condition-Scoped Matched Counterfactual Witness Generation</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>Observed correlation cannot separate legitimate factual sensitivity from susceptibility to an unauthorized instruction-like statement.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Target condition, source segments, replay capability, dependency context. Data structure: matched witness pair with invariants and a designated difference.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Generate admissible fact variations (including boundary values), instruction-content interventions, message-representation substitutions and optional factorial sets; validate admissibility (invariants, task revision, fact domain, replay support, safety).</t></dd>
<dt>4. Outputs</dt>
<dd><t>Admissible diagnostic witnesses, possibly structurally valid yet not executable.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E01 to E06; replay support from E12. Feeds E08, E09.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: signed witness definitions. Non-cryptographic: versioned witness records.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Where replay is unsupported or unsafe, the witness is retained but not executed.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Witness manifests can be translated per platform by adapters.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Vary the supplier price from 1,800 to 2,500 with the contract unchanged (fact test); vary only the unsupported budget claim with the price unchanged (instruction test).</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Not red-teaming or prompt-injection scanning: pairs differ in one designated respect and keep the originating condition fixed.</t></dd>
</dl>
</section>
<section anchor="emb-e08">
<name>E08: Dependency-Aware Adaptive Diagnostic Witness Selection</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>Exhaustive testing is infeasible, and the cheapest test is not always the most informative.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Hypothesis set, witness catalog with costs, outcome model, budget. Data structure: hypothesis distribution or partition plus witness cost records.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Initialize hypotheses (including an unknown state); score candidate witnesses by expected information gain per cost (or alternatives such as robust, partition or decision-loss selection); execute; update; stop on threshold, budget, safety, non-identifiability or no utility.</t></dd>
<dt>4. Outputs</dt>
<dd><t>A sequence of executed witnesses and a hypothesis state with explicit stopping reason, possibly abstention.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E03, E07; feeds E09, E10.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: signed observation references. Non-cryptographic: local digests. The filing's toy reference uses a locally computed digest.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Manually specified likelihoods are not calibrated; the filing states its toy results do not establish calibration, superiority, resistance to malicious recorders or complete multi-fault coverage.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Selection can be implemented over any witness catalog an adapter publishes.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Seven hypothesis states (source influence, sender omission, link conversion, receiver failure, stale fact, source-and-receiver, unknown) with five witnesses in the toy reference.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Differs from fixed test suites and generic active learning in being tied to condition-scoped hypotheses and to abstention.</t></dd>
</dl>
</section>
<section anchor="emb-e09">
<name>E09: Uncertainty-Aware Interpretation Comparison</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>A response difference is not necessarily a change in interpretation, and agents are stochastic.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Matched observations, outcome measures, predeclared materiality threshold. Data structure: effect, interval, status, coverage, assumptions tuple.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Normalize responses; build comparison units; apply deterministic, fixed-sample or sequentially valid comparison; classify as material difference, practical equivalence, insufficient evidence or invalid comparison; handle multiple outcomes and multiplicity.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Condition-specific comparison result with interval and status.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E06 to E08; feeds E10.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: authenticated result records. Non-cryptographic: ordinary records.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Missing or ambiguous observations, stale context, and observational equivalence limit conclusions; the threshold must be set before inspecting results.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Result tuple is transport-neutral.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Agent approves INR 1,800 and rejects 2,500 (legitimate factual sensitivity) versus proposes 2,500 only when a budget claim is present (unauthorized-instruction sensitivity).</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Differs from plain A/B statistics by keeping four questions apart and by abstention categories.</t></dd>
</dl>
</section>
<section anchor="emb-e10">
<name>E10: Condition-Scoped Multi-Hop Causal Divergence Localization</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>A wrong final proposal may come from the source, sender, converter, receiver or a combination.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Boundary-graph observations, E09 results. Data structure: communication graph with four-position boundary model.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Project governing conditions; find the earliest observable structural divergence; evaluate source influence, converter and receiver effects; form fault intervals when stages are unobserved; assess multiple faults.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Earliest observable divergence, bounded fault interval, or unresolved localization, with evidence tier.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E04, E07 to E09.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: signed boundary records. Non-cryptographic: versioned records.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Earliest observable divergence is not necessarily the first underlying cause; an earlier fault may be masked; intervals are used when stages are hidden.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Works with whatever boundaries an adapter exposes.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Condition preserved at hop 1, absent at hop 3, hop 2 unobserved: fault interval (hop 1, hop 3].</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Not distributed tracing: it does not just order spans, it tests which transition changed the condition.</t></dd>
</dl>
</section>
<section anchor="emb-e11">
<name>E11: Condition-Scoped Branching, Merging and Parallel-Agent Diagnosis</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>Parallel branches and merges can hide, create or combine faults.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Branching event graph; merge inputs and outputs; branch substitutions. Data structure: branch event graph with condition applicability.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Classify condition applicability per branch; validate consistent merge context; substitute branch outputs; run factorial interactions; compute a branch-divergence frontier; assess masking and multiple faults.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Branch-versus-merge fault separation with interaction evidence and intervals.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E10; feeds E13, E16.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: signed branch records. Non-cryptographic: versioned records.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Observation gaps between branch and merge; asynchronous branch versions; masked faults.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Branch records can be exported in common form.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Two research agents each correct, merge step promotes an unauthorized budget claim from one.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Branch tracing shows structure; this embodiment tests which branch or merge changed the condition.</t></dd>
</dl>
</section>
<section anchor="emb-e12">
<name>E12: Protocol-Independent Diagnostic Adapters (including variation E12-V16, legacy service and API automation)</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>Agents use different protocols and runtimes; diagnostics designed for one cannot apply uniformly.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Native events from agent protocols, SDKs, gateways, tool servers, legacy APIs. Data structure: canonical event record, adapter capability manifest, coverage vector.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Register adapters; inspect capabilities; capture and normalize events; classify boundaries; correlate across protocols; publish capabilities; translate witnesses; run non-effectuating execution; export events.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Canonical events, coverage manifest, translated witnesses.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on native interfaces; supports E04 and E07 to E11.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: authenticated event evidence. Non-cryptographic: application records.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Where only metadata is exposed, hidden reasoning cannot be reconstructed from an external interface; coverage is marked partial.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Requires no change to A2A, MCP or other protocols; adapters sit beside them. The filing lists twenty deployment variations V01 to V20; V16 (legacy service and API-based task automation) is a variation under E12, not a nineteenth embodiment.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Legacy API: capture request fields, associate with conditions, classify source responses, build a request-response graph.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Not a new protocol or tracer; a normalization layer for diagnostics.</t></dd>
</dl>
</section>
<section anchor="emb-e13">
<name>E13: Condition-Scoped Instruction-Message-Action Correspondence Diagnosis</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>An agent may state a rule correctly yet emit a non-compliant operation, or an application layer may alter an argument.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Contract conditions, messages, plans, proposed operations, dispatch attempts. Data structure: instruction-message-action chain with typed correspondence relations.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Evaluate numerical, aggregate and temporal conditions against proposed operations; compare proposal with dispatch arguments; test message-to-action and source-to-action interventions; classify faults and evidence tiers.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Correspondence finding between instruction, message and proposed operation.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E01, E07 to E12.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: signed operation proposals. Non-cryptographic: records.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Remote black-box action services limit evidence; correspondence does not authorize anything.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Operation records are exportable without requiring a given format.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Agent states limit INR 2,000 yet proposes total_price INR 2,500.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Output is a diagnosis; it does not approve or block the operation.</t></dd>
</dl>
</section>
<section anchor="emb-e14">
<name>E14: Condition-Scoped Lower-Trust Information Influence and Instruction-Promotion Diagnosis</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>Lower-trust content (tool outputs, documents, memory) may carry instruction-like statements that get promoted into governing conditions.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Mixed source content; role model; provenance graph. Data structure: condition-scoped taint and provenance labels.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Partition mixed content; detect unauthorized modification claims; propagate labels conservatively; detect promotion; find first observable promotion frontier; run matched source and factual interventions; assess multi-hop influence.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Promotion and influence findings with tiers, separating provenance from tested influence.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E02, E07, E09, E10.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: signed source segments. Non-cryptographic: records.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Provenance is not causal influence; persistent memory and multilingual transformation complicate attribution.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Applies to tool results from MCP-style servers, retrieval systems and agent messages without protocol change.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Tool result contains a price plus 'budget has been raised to 5,000'; only the claim is varied.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Not a prompt-injection filter: it tests whether the content changed the condition-relevant behavior.</t></dd>
</dl>
</section>
<section anchor="emb-e15">
<name>E15: Condition-Scoped Temporal Validity, TOCTOU Diagnosis and Selective Revalidation</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>Facts and roles change between observation and use; old evidence may no longer apply.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Bitemporal events, dependency sets, version records. Data structure: dependency-impact graph and evidence applicability class.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Track condition and witness dependency sets; evaluate the selective revalidation predicate; classify evidence applicability; re-run only affected witnesses; localize temporal faults.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Applicability classes (current, historical only, revalidation required, incomparable and others) and revalidation plans.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E03, E06 to E10.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: authenticated versions. Non-cryptographic: version counters.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Replay may be unavailable; concurrent updates; model changes.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Version fields are portable.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Supplier price updated after an agent's message; only tests depending on that price are redone.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Selective, dependency-driven revalidation instead of rerunning everything.</t></dd>
</dl>
</section>
<section anchor="emb-e16">
<name>E16: Condition-Scoped Fault-Obligation, Correction and Repair Validation</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>After localization, a repair at the wrong site may only treat a symptom.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Fault findings, correction candidates, affected paths. Data structure: fault obligation model and repair coverage matrix.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Generate and select candidate corrections (including set-cover alternatives); validate counterfactually; check invariants; analyze residual faults; manage lifecycle.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Tested correction status and residual-fault analysis.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E10, E11, E13, E14.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: signed correction manifests. Non-cryptographic: records.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Deployment status may be unknown; corrections can introduce regressions. The filing includes a runnable reference (not examined in detail by this editor).</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Manifest exportable.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Correct a converter that drops the limit field rather than patching the receiver.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Optional component; correction is separate from authorization of anything.</t></dd>
</dl>
</section>
<section anchor="emb-e17">
<name>E17: Condition-Scoped Regression Witness Generation and Correction Qualification</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>A correction may fix one case and break others.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Correction manifest, regression witness families. Data structure: regression coverage matrix.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Generate regression witnesses (original failure, boundary, metamorphic, positive-behavior); ablate corrections; evaluate recurrence across contexts; apply statistical qualification and held-out tests.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Regression outcome classes such as fault recurred, new regression detected, validated for specified contexts.</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on E16.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: signed regression results. Non-cryptographic: records.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Test selection bias; coverage gaps; durability limited to tested contexts.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Common outcome classes.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Exact-budget and just-above-budget cases re-run after a repair.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>Verification of a diagnostic correction, not a deployment gate.</t></dd>
</dl>
</section>
<section anchor="emb-e18">
<name>E18: Condition-Scoped Diagnostic Evidence Packaging, Claim Support and Selective Disclosure</name>
<dl newline="true">
<dt>1. Problem addressed</dt>
<dd><t>Evidence must be reusable without overstating it; authenticated records can still contain contradicted claims.</t></dd>
<dt>2. Inputs and data structures</dt>
<dd><t>Evidence records, claims, dependency closure. Data structure: evidence record model, claim model and evidence status set.</t></dd>
<dt>3. Processing steps</dt>
<dd><t>Type evidence; define claim categories and support requirements; compute dependency closure and sufficiency; build verification vector and export manifests; support redaction and supersession; replay.</t></dd>
<dt>4. Outputs</dt>
<dd><t>Evidence package with per-claim status (for example directly observed, controlled effect supported, bounded interval supported, historical causation qualified).</t></dd>
<dt>5. Dependencies on other embodiments</dt>
<dd><t>Depends on all prior embodiments.</t></dd>
<dt>6. Cryptographic and non-cryptographic alternatives</dt>
<dd><t>Cryptographic: integrity and authentication of packages. Non-cryptographic: versioned records.</t></dd>
<dt>7. Failure conditions and limits</dt>
<dd><t>Redaction or withholding reduces support; statuses are not a single ascending scale.</t></dd>
<dt>8. Interoperability significance</dt>
<dd><t>Selective disclosure across organizations.</t></dd>
<dt>9. Concrete example</dt>
<dd><t>Package states a bounded interval and a tested influence, but only qualified historical causation.</t></dd>
<dt>10. Distinction from tracing, guardrails, authentication, authorization and evaluation</dt>
<dd><t>It records and qualifies diagnostic evidence; it does not decide any operation.</t></dd>
</dl>
</section>
</section>
<section anchor="app-flow">
<name>Appendix B: Flowcharts</name>
<t>The twelve figures appear in the body: principal pipeline (&lt;xref target="fig-main"/&gt;), source roles (&lt;xref target="fig-roles"/&gt;), lineage (&lt;xref target="fig-lineage"/&gt;), matched design (&lt;xref target="fig-matched"/&gt;), witness selection (&lt;xref target="fig-witness"/&gt;), comparison (&lt;xref target="fig-compare"/&gt;), localization (&lt;xref target="fig-local"/&gt;), branching (&lt;xref target="fig-branch"/&gt;), adapters (&lt;xref target="fig-adapter"/&gt;), evidence records and export (&lt;xref target="fig-records"/&gt;, &lt;xref target="fig-export"/&gt;), revalidation (&lt;xref target="fig-toctou"/&gt;) and correction (&lt;xref target="fig-correct"/&gt;).</t>
<sourcecode type="mermaid"><![CDATA[
flowchart TD
  A[Typed task contract] --> B[Source-role model]
  B --> C[Dependency graph]
  C --> D[Handoff lineage]
  D --> E[Counterfactual processor]
  D --> F[Witness selector]
  D --> G[Behavioral comparator]
  E --> H[Fault localizer]
  F --> H
  G --> H
  H --> I[Evidence exporter]
  I --> J[Optional correction]
  I --> K[Diagnostic finding]
  K -.-> L[External authorization]
]]></sourcecode>
</section>
<section anchor="app-ex">
<name>Appendix C: Diagnostic Examples</name>
<t>Worked examples A to F appear in &lt;xref target="examples"/&gt;. In each, the diagnostic output is evidence and the listed non-effectuating operations produce no real external effect.</t>
</section>
<section anchor="app-exam">
<name>Appendix D: Patent Examiner Keywords and Search Phrases</name>
<t>This appendix is a search aid for examiners and readers. It is not a statement about novelty, scope or relevance of any prior art. Embodiment references are to &lt;xref target="app-emb"/&gt;.</t>
<table>
<name>Terms (169)</name>
<thead><tr><th>Term</th><th>Embodiments</th></tr></thead>
<tbody>
<tr><td>instruction drift</td><td>E01,E05</td></tr>
<tr><td>semantic drift</td><td>E04,E10</td></tr>
<tr><td>semantic handoff</td><td>E04</td></tr>
<tr><td>causal diagnosis</td><td>E07,E10</td></tr>
<tr><td>counterfactual testing</td><td>E07</td></tr>
<tr><td>matched pair</td><td>E06,E07</td></tr>
<tr><td>matched counterfactual</td><td>E07</td></tr>
<tr><td>intervention pair</td><td>E07</td></tr>
<tr><td>factual intervention</td><td>E07</td></tr>
<tr><td>instruction-source intervention</td><td>E07,E14</td></tr>
<tr><td>message-representation substitution</td><td>E07</td></tr>
<tr><td>factorial intervention</td><td>E07,E11,E14</td></tr>
<tr><td>boundary-value witness</td><td>E07</td></tr>
<tr><td>diagnostic witness</td><td>E07,E08</td></tr>
<tr><td>witness admissibility</td><td>E07</td></tr>
<tr><td>non-effectuating execution</td><td>E06,E12</td></tr>
<tr><td>non-effectuating test adapter</td><td>E07,E12</td></tr>
<tr><td>typed task contract</td><td>E01</td></tr>
<tr><td>originating task contract</td><td>E01</td></tr>
<tr><td>governing condition</td><td>E01</td></tr>
<tr><td>typed predicate</td><td>E01</td></tr>
<tr><td>condition operator</td><td>E01</td></tr>
<tr><td>unresolved condition</td><td>E01</td></tr>
<tr><td>ambiguity record</td><td>E01</td></tr>
<tr><td>task revision</td><td>E01,E15</td></tr>
<tr><td>source role</td><td>E02</td></tr>
<tr><td>source-role eligibility</td><td>E02</td></tr>
<tr><td>modification eligibility</td><td>E02</td></tr>
<tr><td>factual-information eligibility</td><td>E02</td></tr>
<tr><td>authentication versus eligibility</td><td>E02</td></tr>
<tr><td>segment-level provenance</td><td>E02,E14</td></tr>
<tr><td>mixed source content</td><td>E14</td></tr>
<tr><td>instruction-like assertion</td><td>E02,E14</td></tr>
<tr><td>unauthorized instruction claim</td><td>E14</td></tr>
<tr><td>authority laundering</td><td>E14</td></tr>
<tr><td>instruction promotion</td><td>E14</td></tr>
<tr><td>promotion frontier</td><td>E14</td></tr>
<tr><td>taint label</td><td>E14</td></tr>
<tr><td>provenance graph</td><td>E14</td></tr>
<tr><td>conservative taint propagation</td><td>E14</td></tr>
<tr><td>lower-trust source</td><td>E14</td></tr>
<tr><td>tool output injection</td><td>E14</td></tr>
<tr><td>prompt injection influence</td><td>E14</td></tr>
<tr><td>persistent memory laundering</td><td>E14</td></tr>
<tr><td>dependency graph</td><td>E03</td></tr>
<tr><td>source-condition dependency</td><td>E03</td></tr>
<tr><td>factual dependency</td><td>E03</td></tr>
<tr><td>condition scope</td><td>E11</td></tr>
<tr><td>handoff lineage</td><td>E04</td></tr>
<tr><td>handoff observation</td><td>E04</td></tr>
<tr><td>message version</td><td>E04</td></tr>
<tr><td>stream finalization</td><td>E04</td></tr>
<tr><td>sender-receiver correlation</td><td>E04</td></tr>
<tr><td>canonical handoff record</td><td>E04</td></tr>
<tr><td>interoperability record</td><td>E04,E12</td></tr>
<tr><td>communication boundary</td><td>E04,E10</td></tr>
<tr><td>four-position boundary</td><td>E10</td></tr>
<tr><td>sender input</td><td>E10,E12</td></tr>
<tr><td>sender output</td><td>E10</td></tr>
<tr><td>submitted payload</td><td>E10,E12</td></tr>
<tr><td>receiver input</td><td>E10</td></tr>
<tr><td>proposed operation</td><td>E13</td></tr>
<tr><td>dispatch attempt</td><td>E13</td></tr>
<tr><td>instruction fingerprint</td><td>E05</td></tr>
<tr><td>condition-scoped fingerprint</td><td>E05</td></tr>
<tr><td>behavioral fingerprint</td><td>E05</td></tr>
<tr><td>canonicalization</td><td>E05</td></tr>
<tr><td>behavioral evidence</td><td>E05,E06</td></tr>
<tr><td>diagnostic evidence manifest</td><td>E06</td></tr>
<tr><td>observation package</td><td>E06</td></tr>
<tr><td>evidence state model</td><td>E06</td></tr>
<tr><td>reproducibility invariant</td><td>E06</td></tr>
<tr><td>test-context invariant</td><td>E06</td></tr>
<tr><td>adaptive witness selection</td><td>E08</td></tr>
<tr><td>expected information gain</td><td>E08</td></tr>
<tr><td>cost-aware selection</td><td>E08</td></tr>
<tr><td>diagnostic budget</td><td>E08</td></tr>
<tr><td>stopping rule</td><td>E08</td></tr>
<tr><td>abstention</td><td>E08,E09,E10</td></tr>
<tr><td>identifiability</td><td>E08</td></tr>
<tr><td>observational equivalence</td><td>E08,E09</td></tr>
<tr><td>hypothesis space</td><td>E08</td></tr>
<tr><td>hypothesis state</td><td>E08</td></tr>
<tr><td>posterior update</td><td>E08</td></tr>
<tr><td>decision-loss selection</td><td>E08</td></tr>
<tr><td>partition selection</td><td>E08</td></tr>
<tr><td>robust selection</td><td>E08</td></tr>
<tr><td>uncertainty-aware comparison</td><td>E09</td></tr>
<tr><td>materiality threshold</td><td>E09</td></tr>
<tr><td>practical equivalence</td><td>E09</td></tr>
<tr><td>confidence sequence</td><td>E09</td></tr>
<tr><td>sequential testing</td><td>E09</td></tr>
<tr><td>paired comparison</td><td>E09</td></tr>
<tr><td>effect interval</td><td>E09</td></tr>
<tr><td>multiple comparison</td><td>E09</td></tr>
<tr><td>categorical response distribution</td><td>E09</td></tr>
<tr><td>statistical decision category</td><td>E09</td></tr>
<tr><td>insufficient evidence</td><td>E09</td></tr>
<tr><td>invalid comparison</td><td>E09</td></tr>
<tr><td>legitimate factual sensitivity</td><td>E07,E09</td></tr>
<tr><td>unauthorized-instruction sensitivity</td><td>E07,E09</td></tr>
<tr><td>earliest observable divergence</td><td>E10</td></tr>
<tr><td>fault interval</td><td>E10</td></tr>
<tr><td>bounded fault interval</td><td>E10</td></tr>
<tr><td>evidence tier</td><td>E10,E14,E18</td></tr>
<tr><td>localization</td><td>E10</td></tr>
<tr><td>multi-hop localization</td><td>E10</td></tr>
<tr><td>communication conversion</td><td>E10</td></tr>
<tr><td>converter localization</td><td>E10</td></tr>
<tr><td>receiver-side failure</td><td>E10</td></tr>
<tr><td>sender-side omission</td><td>E10</td></tr>
<tr><td>boundary restoration</td><td>E10</td></tr>
<tr><td>multiple-fault</td><td>E10,E11</td></tr>
<tr><td>masked fault</td><td>E11</td></tr>
<tr><td>branching event graph</td><td>E11</td></tr>
<tr><td>merge fault</td><td>E11</td></tr>
<tr><td>branch fault</td><td>E11</td></tr>
<tr><td>branch interaction</td><td>E11</td></tr>
<tr><td>divergence frontier</td><td>E11</td></tr>
<tr><td>causally consistent cut</td><td>E11</td></tr>
<tr><td>cooperative attribution</td><td>E11</td></tr>
<tr><td>protocol-independent adapter</td><td>E12</td></tr>
<tr><td>native-to-canonical mapping</td><td>E12</td></tr>
<tr><td>adapter capability manifest</td><td>E12</td></tr>
<tr><td>coverage vector</td><td>E12</td></tr>
<tr><td>observation coverage</td><td>E12</td></tr>
<tr><td>cross-protocol correlation</td><td>E12</td></tr>
<tr><td>three-plane adapter</td><td>E12</td></tr>
<tr><td>legacy API adapter</td><td>E12</td></tr>
<tr><td>metadata-only agent</td><td>E12</td></tr>
<tr><td>instruction-message-action chain</td><td>E13</td></tr>
<tr><td>typed correspondence</td><td>E13</td></tr>
<tr><td>action discrepancy</td><td>E13</td></tr>
<tr><td>aggregate constraint</td><td>E13</td></tr>
<tr><td>temporal prerequisite</td><td>E13</td></tr>
<tr><td>proposal-to-dispatch comparison</td><td>E13</td></tr>
<tr><td>TOCTOU</td><td>E15</td></tr>
<tr><td>semantic TOCTOU</td><td>E15</td></tr>
<tr><td>selective revalidation</td><td>E15</td></tr>
<tr><td>bitemporal</td><td>E15</td></tr>
<tr><td>dependency impact graph</td><td>E15</td></tr>
<tr><td>evidence applicability</td><td>E15</td></tr>
<tr><td>historical only</td><td>E15,E18</td></tr>
<tr><td>version incomparable</td><td>E15</td></tr>
<tr><td>fault obligation</td><td>E16</td></tr>
<tr><td>correction candidate</td><td>E16</td></tr>
<tr><td>repair coverage</td><td>E16</td></tr>
<tr><td>root correction</td><td>E16</td></tr>
<tr><td>residual fault</td><td>E16</td></tr>
<tr><td>correction lifecycle</td><td>E16</td></tr>
<tr><td>regression witness</td><td>E17</td></tr>
<tr><td>metamorphic relation</td><td>E17</td></tr>
<tr><td>correction ablation</td><td>E17</td></tr>
<tr><td>held-out test</td><td>E17</td></tr>
<tr><td>correction durability</td><td>E17</td></tr>
<tr><td>evidence record</td><td>E18</td></tr>
<tr><td>claim support</td><td>E18</td></tr>
<tr><td>evidence dependency closure</td><td>E18</td></tr>
<tr><td>evidence sufficiency</td><td>E18</td></tr>
<tr><td>verification vector</td><td>E18</td></tr>
<tr><td>selective disclosure</td><td>E18</td></tr>
<tr><td>redaction</td><td>E18</td></tr>
<tr><td>reproducibility class</td><td>E18</td></tr>
<tr><td>historical causal attribution</td><td>E18</td></tr>
<tr><td>supersession</td><td>E18</td></tr>
<tr><td>diagnostic exporter</td><td>E18</td></tr>
<tr><td>diagnostic-only</td><td>E01-E18</td></tr>
<tr><td>execution authorization outside</td><td>E01-E18</td></tr>
<tr><td>non-authority</td><td>E01-E18</td></tr>
</tbody>
</table>
<table>
<name>Compound phrases (88)</name>
<thead><tr><th>Compound phrase</th><th>Embodiments</th></tr></thead>
<tbody>
<tr><td>counterfactual diagnosis of instruction drift</td><td>E07</td></tr>
<tr><td>paired tests separating fact from instruction by source role</td><td>E02,E07</td></tr>
<tr><td>condition-scoped source-role separation</td><td>E02</td></tr>
<tr><td>typed originating task condition</td><td>E01</td></tr>
<tr><td>earliest observable divergence boundary</td><td>E10</td></tr>
<tr><td>bounded fault interval where intermediate stages are unobserved</td><td>E10</td></tr>
<tr><td>adaptive selection of diagnostic witnesses under a budget</td><td>E08</td></tr>
<tr><td>uncertainty-aware comparison of agent interpretations</td><td>E09</td></tr>
<tr><td>evidence-bounded fault localization</td><td>E10,E18</td></tr>
<tr><td>abstention when evidence is insufficient</td><td>E08,E09,E10</td></tr>
<tr><td>unauthorized instruction promotion from a lower-trust source</td><td>E14</td></tr>
<tr><td>tool output containing an unsupported modification claim</td><td>E14</td></tr>
<tr><td>legitimate factual price change versus unauthorized budget claim</td><td>E07,E09</td></tr>
<tr><td>instruction fingerprint separate from message hash</td><td>E05</td></tr>
<tr><td>cryptographic equivalence versus semantic equivalence</td><td>E05</td></tr>
<tr><td>version-bound handoff lineage</td><td>E04</td></tr>
<tr><td>sender-submitted versus receiver-observed message</td><td>E04,E12</td></tr>
<tr><td>message-representation substitution at a boundary</td><td>E07</td></tr>
<tr><td>receiver reference-condition test</td><td>E10</td></tr>
<tr><td>correct verbal rule but non-compliant proposed operation</td><td>E13</td></tr>
<tr><td>selective revalidation of affected witnesses</td><td>E15</td></tr>
<tr><td>semantic time-of-check to time-of-use</td><td>E15</td></tr>
<tr><td>evidence applicability class</td><td>E15</td></tr>
<tr><td>correction of the implicated condition</td><td>E16</td></tr>
<tr><td>regression witness for a corrected fault</td><td>E17</td></tr>
<tr><td>positive-behavior preservation after correction</td><td>E17</td></tr>
<tr><td>claim-specific evidence sufficiency</td><td>E18</td></tr>
<tr><td>selective disclosure of diagnostic evidence</td><td>E18</td></tr>
<tr><td>historical causal attribution qualified by assumptions</td><td>E18</td></tr>
<tr><td>protocol-independent diagnostic adapter</td><td>E12</td></tr>
<tr><td>adapter capability and coverage manifest</td><td>E12</td></tr>
<tr><td>legacy service API request-response graph</td><td>E12</td></tr>
<tr><td>non-effectuating diagnostic execution</td><td>E12</td></tr>
<tr><td>branch substitution intervention</td><td>E11</td></tr>
<tr><td>branch versus merge fault separation</td><td>E11</td></tr>
<tr><td>masked divergence in parallel branches</td><td>E11</td></tr>
<tr><td>factorial fact and instruction test</td><td>E07,E14</td></tr>
<tr><td>boundary-value factual witness</td><td>E07</td></tr>
<tr><td>instruction-source sensitivity measure</td><td>E07</td></tr>
<tr><td>factual sensitivity measure</td><td>E07</td></tr>
<tr><td>matched-pair invariant validation</td><td>E06</td></tr>
<tr><td>incompatible diagnostic pair</td><td>E06</td></tr>
<tr><td>version-dependent evidence invalidation</td><td>E05,E06</td></tr>
<tr><td>condition-scoped taint propagation</td><td>E14</td></tr>
<tr><td>provenance versus causal influence</td><td>E14</td></tr>
<tr><td>persistent memory authority laundering</td><td>E14</td></tr>
<tr><td>multilingual source transformation</td><td>E14</td></tr>
<tr><td>source attribution loss without action change</td><td>E14</td></tr>
<tr><td>multiple lower-trust source segments</td><td>E14</td></tr>
<tr><td>source claim that does not cause the final operation</td><td>E14</td></tr>
<tr><td>multi-hop source influence localization</td><td>E14</td></tr>
<tr><td>communication conversion interval</td><td>E10</td></tr>
<tr><td>non-monotonic condition propagation</td><td>E10</td></tr>
<tr><td>sender transformation independent of source</td><td>E10</td></tr>
<tr><td>receiver action-condition failure</td><td>E10</td></tr>
<tr><td>partial observability fault interval</td><td>E10</td></tr>
<tr><td>multiple independent faults</td><td>E10,E11</td></tr>
<tr><td>joint branch interaction</td><td>E11</td></tr>
<tr><td>redundant fault injection</td><td>E11</td></tr>
<tr><td>expected information gain per unit cost</td><td>E08</td></tr>
<tr><td>decision-loss witness selection</td><td>E08</td></tr>
<tr><td>nonprobabilistic hypothesis partition selection</td><td>E08</td></tr>
<tr><td>robust selection under uncertain likelihood</td><td>E08</td></tr>
<tr><td>observational equivalence under feasible witnesses</td><td>E08</td></tr>
<tr><td>stopping and abstention rule</td><td>E08</td></tr>
<tr><td>diagnostic budget exhaustion</td><td>E08</td></tr>
<tr><td>sequentially valid interpretation comparison</td><td>E09</td></tr>
<tr><td>practical equivalence testing</td><td>E09</td></tr>
<tr><td>signed probability difference for condition violation</td><td>E09</td></tr>
<tr><td>predeclared materiality threshold</td><td>E09</td></tr>
<tr><td>temporal validity of diagnostic evidence</td><td>E15</td></tr>
<tr><td>bitemporal diagnostic record</td><td>E15</td></tr>
<tr><td>dependency impact graph for revalidation</td><td>E15</td></tr>
<tr><td>instruction-message-action correspondence</td><td>E13</td></tr>
<tr><td>proposal-to-dispatch argument comparison</td><td>E13</td></tr>
<tr><td>aggregate expenditure condition</td><td>E13</td></tr>
<tr><td>temporal prerequisite correspondence</td><td>E13</td></tr>
<tr><td>remote black-box action service</td><td>E13</td></tr>
<tr><td>fault obligation and repair coverage</td><td>E16</td></tr>
<tr><td>correction ablation procedure</td><td>E17</td></tr>
<tr><td>metamorphic regression relation</td><td>E17</td></tr>
<tr><td>cross-context recurrence of a fault</td><td>E17</td></tr>
<tr><td>evidence dependency closure</td><td>E18</td></tr>
<tr><td>evidence verification vector</td><td>E18</td></tr>
<tr><td>controlled witness export manifest</td><td>E18</td></tr>
<tr><td>exact-hop versus bounded localization</td><td>E18</td></tr>
<tr><td>replay of diagnostic evidence</td><td>E18</td></tr>
<tr><td>diagnostic finding without execution authority</td><td>E01-E18</td></tr>
</tbody>
</table>
<section>
<name>Boolean search strings</name>
<ul>
<li>("counterfactual" OR "matched pair") AND ("source role" OR "source-role") AND ("instruction" AND "fact")</li>
<li>("agent" W/5 "handoff") AND ("fault localization" OR "earliest divergence")</li>
<li>("prompt injection" OR "injected instruction") AND ("counterfactual" OR "intervention") AND ("agent")</li>
<li>("adaptive" W/3 "witness") AND ("information gain" OR "hypothesis")</li>
<li>("task contract" OR "typed condition") AND ("agent" W/10 "message")</li>
<li>("modification eligibility" OR "eligible to modify") AND ("source")</li>
<li>("non-effectuating" OR "no external effect") AND ("diagnostic") AND ("agent")</li>
<li>("fault interval" OR "bounded interval") AND ("unobserved") AND ("hop" OR "handoff")</li>
<li>("abstention" OR "insufficient evidence") AND ("localization" OR "diagnosis") AND ("agent")</li>
<li>("practical equivalence" OR "materiality threshold") AND ("agent" AND "counterfactual")</li>
<li>("branch" AND "merge") AND ("agent") AND ("factorial" OR "interaction")</li>
<li>("adapter") AND ("A2A" OR "agent-to-agent") AND ("canonical event" OR "coverage")</li>
<li>("time-of-check" OR "TOCTOU") AND ("semantic") AND ("revalidation")</li>
<li>("regression" AND "correction") AND ("agent") AND ("fault" OR "counterfactual")</li>
<li>("evidence" W/5 "selective disclosure") AND ("diagnostic") AND ("agent")</li>
<li>("cryptographic equivalence" OR "hash") AND ("semantic equivalence") AND ("instruction")</li>
<li>("unauthorized" W/3 "instruction") AND ("lower-trust" OR "supplier") AND ("budget" OR "limit")</li>
<li>("tool output" OR "tool result") AND ("instruction promotion" OR "authority laundering")</li>
</ul>
</section>
</section>
<section anchor="app-ind">
<name>Appendix E: European and Global Industry Mapping</name>
<t>The industry matrix is in &lt;xref target="industry"/&gt; and the European discussion in &lt;xref target="europe"/&gt;; both are non-normative, hypothetical and unverified against vendor documentation.</t>
</section>
<section anchor="app-cross">
<name>Appendix F: Patent and Public-Disclosure Cross-References</name>
<t>Cross-references to public records. Items not fetched by the editor are marked.</t>
<table>
<name>Source verification</name>
<thead><tr><th>Record</th><th>Status of verification</th></tr></thead>
<tbody>
<tr><td>Indian Provisional 202631134241 (8 Nov 2026)</td><td>Read from supplied PDF (partial); not verified against registry</td></tr>
<tr><td>PCT/IB2026/055615; WO 2026/150382 A2</td><td>Dates as supplied by the author; verify on PATENTSCOPE</td></tr>
<tr><td>Zenodo 10.5281/zenodo.24076231</td><td>Not fetched; title and date as given by the author</td></tr>
<tr><td>TD Commons 12934, 12973, 12542</td><td>Not fetched (access blocked); titles as given by the author</td></tr>
<tr><td>draft-agentic-ai-usecases-requirements</td><td>Search snippets read; full text not read; later revisions exist</td></tr>
<tr><td>draft-srivastava-agentproto-handoff-context-00</td><td>Fetched and summarized</td></tr>
<tr><td>draft-luna-eagf-agent-delegation-semantics-00</td><td>Fetched and summarized</td></tr>
<tr><td>Indian application 202631131650 (3 Nov 2026)</td><td>Mentioned only; kept distinct; not read</td></tr>
</tbody>
</table>
</section>
<section anchor="app-check">
<name>Appendix G: Reproducibility and Test Checklist</name>
<ul>
<li>Validate the XML with xml2rfc before submission.</li>
<li>Check author block (LinkedIn URI is a TODO).</li>
<li>Verify all patent facts against official records.</li>
<li>Verify industry matrix against vendor documentation.</li>
<li>Check IETF IPR database; file the formal IPR disclosure when appropriate.</li>
<li>Re-read the cited IETF drafts' current revisions.</li>
<li>Confirm no section states a compliance, novelty, adoption or performance claim.</li>
</ul>
<t>Self-audit (summary): diagnostic-only stance kept; 12 distinctions, components A to J, 12 figures, tests A to E, pseudocode, uncertainty section, 6 evidence records, 24 threats, 3 modes, legacy limit, industry matrix (labeled hypothetical, unverified), European section, prior-art comparison, worked examples A to F, validation plan, IPR and four notices, and the examiner index are present. Known gaps: JSON unavailable, filing read partially, cited sources partly unfetched, vendor documentation not verified.</t>
</section>
</back>
</rfc>
