1. The Gap
A confirmation latency recorded against a counterparty admits two explanations that the recorded latency alone cannot separate. In the first, the counterparty is slow to confirm. In the second, the domain in which the interaction runs is slow, and every party operating inside that domain confirms slowly. An agent that reads the recorded latency as the party's fault charges that party for a slowness belonging to the channel it happens to work in. Watching more of the same latency does not separate the two explanations; only varying the dispatch does.
The interventional attribution machinery in U.S. Provisional Application No. 64/117,812 resolves the question inside one agent's own dealings. Instead of watching passively, the agent varies a property of its own dispatch, routing matched interactions to close on a receipt the counterparty controls or on a receipt it does not, and comparing the resulting latencies. When the comparison attributes the latency to the domain rather than to the party, the agent holds a finding that the party is not at fault.
That finding is local. The agent that ran the test knows the party's slowness was the channel's, but a stranger meeting the same party later holds no such finding and starts from nothing. The gap is how to carry a favorable finding across a trust boundary without minting a portable reputation score, without a central registry, and without letting the carried record become something a party could forge or replay to launder a real problem.
2. Mechanism
The portable exoneration record exists only as a product of the domain-attribution branch. When the interventional attribution attributes an observed latency to a domain rather than to a counterparty, the persistent semantic agent (100) appends an exoneration record identifying the matched class, both closure latencies, and the threshold applied, then transmits it to the counterparty as a governed observation. The counterparty now holds a signed record, authored by an agent it dealt with, showing that a specific body of its slow confirmations was charged to the channel and not to itself.
What the record contains is exactly what the branch produced: a matched-class identifier, two latency figures, and the threshold applied. It carries no conduct descriptor and states no characterization of the party's conduct, since the domain branch writes nothing adverse against the counterparty in the first place.
On the consuming side, a further persistent semantic agent (100) presented the exoneration record by that counterparty does not take it on faith. It admits the record as an input to a first-encounter determination alone, and only after checking three things against its own material. First, that the record is signed by the agent that appended it, fixing authorship to the observing agent so the presenting party cannot stand in for it. Second, that the matched class is identified by the feature tuple recorded for it, so the record points to a determinate body of interactions rather than a bare assertion. Third, that the record's recorded time falls within a declared currency interval, so a finding past its declared currency is not admitted.
Where all three hold, the record is recorded in the counterparty identity record (114) as a presented record, admitted for the first-encounter determination alone. It modifies no value of the receiving agent's scoped integrity vector (106) and increments no counter. The record can color how a stranger reads a party it has never met, and it does nothing beyond that: it moves no integrity quantity and writes no adverse attribute against any party.
Where any part of the verification fails, whether the signature does not check, the feature tuple does not match a recorded matched class, or the recorded time falls outside the currency interval, the receiving agent produces the not-determinable determination (126) and appends nothing adverse. A forged, replayed, or expired record is neither believed nor held against the party that presented it. The failure is inert, not punitive.
The record's evidentiary weight traces back to the controlled comparison that produced it. The exoneration branch fires only when the latency differential between the counterparty-controlled closure path and the path the counterparty does not control fails to exceed the declared threshold, and the latency on the path the counterparty does not control is itself elevated above a baseline computed across the agent's other matched classes, excluding the counterparty under evaluation. The finding therefore rests on the party's slowness showing up even where the party controlled no receipt, which is the mark of a slow channel rather than a slow party.
3. Operating Parameters
The filed specification declares the following structure, and sets no numeric values for it in the operative disclosure; the values are policy-declared.
The record's recorded contents are fixed: the matched class, both closure latencies, and the threshold applied. The matched class is identified by a feature tuple, the constituents of which are the action class, the stakes designation, the urgency designation, and the scope-partition identifier, so the class the record refers to is reconstructable rather than nominal.
The threshold copied into the record is not free. It is bounded below by the standard deviation of the recorded durations of the qualifying closures from which the channel baseline is computed, expressed in the units of the baseline and the threshold. A threshold beneath that bound would attribute ordinary channel variation to a party, so the floor ties the decision to the measured dispersion of the channel itself.
The receiving agent's verification predicate has three declared conditions applied together: signature by the appending agent, identity of the matched class by its recorded feature tuple, and the recorded time falling within a declared currency interval. The currency interval is the only temporal bound on portability, and it is declared, not open-ended.
Upstream of issuance, the domain-attribution branch that produces the record at all is itself gated by declared parameters: a minimum per-path volume below which the evaluation is not-determinable and no exoneration issues, and a declared margin by which the uncontrolled-path latency must exceed its cross-class baseline before the domain, rather than the party, is charged.
A further guard sits alongside the branch. Over the same matched class the agent measures counterparty conduct other than latency, namely the rate of non-closure, of returned exceptions, and of amendment requests, per closure path. Where that measure differs across the two paths by more than a declared manipulation margin, the agent records a manipulation attribute and neither attributes nor exonerates that class, because a difference across paths that were meant to be indistinguishable shows the assignment was in fact distinguishable to the counterparty and the class's latency evidence does not stand.
4. Composition
The record sits on top of the interventional attribution family of Section 10.2 and inherits its discipline. Only one of the branch outcomes produces it. Attribution to the counterparty and the no-adverse-signal outcome produce no exoneration record, and the not-determinable outcome produces none, so a record exists only where a controlled comparison actually cleared the party. The exoneration record is therefore not a thing a party can request; it is emitted by the observing agent when its own procedure reaches the domain branch.
On the consuming side, the record plugs into the first-encounter admission machinery of the same filing. When an agent meets a party for which it holds no counterparty identity record, that party begins in an ephemeral, pre-settlement posture in which no determination of the closed set is produced, no integrity value moves, and no counter increments. The exoneration record is admitted as an input to the first-encounter determination alone, which is exactly consistent with that inert posture: it colors how the stranger reads the newcomer without disturbing the rule that a first encounter moves no adverse state.
The record travels on the shared governed-observation substrate, so it carries the same authorship signature and continuity guarantees that the rest of the filing relies on, and it is recorded in the counterparty identity record (114) alongside the other presented records an agent may hold about a party. Its inert failure mode, producing the not-determinable determination (126) rather than a denial, aligns it with the filing's positive-abstention posture, under which an agent that cannot resolve something withholds rather than condemns.
Thematically the record is an earned-exposure instrument. Standing in this architecture is earned from an agent's own recorded conduct, not conferred by a third party, and the portable exoneration record extends that principle across a boundary: it lets a favorable finding about a party's timing conduct move with the party, while refusing to compress that finding into a cardinal reputation number.
5. Prior-Art Distinction
Portable reputation and rating systems carry a cardinal score or an endorsement from party to party. The exoneration record carries neither. It is a bounded finding that a specific measured latency was charged to a channel, it names no score, and when admitted it moves no quantity in the receiving agent. The distinction is structural: a score is a claim about the party's overall worth, while this record is a claim confined to one attributable measurement.
Verifiable-credential frameworks issue a signed claim about a subject that a relying party accepts on the issuer's authority. Here there is no issuing authority. The author is simply an agent the party previously interacted with, the receiving agent verifies the record entirely on its own material rather than deferring to the author, and the record is admitted for a first-encounter determination alone rather than as a standing attribute. Admission grants no weight beyond that single determination and writes nothing adverse.
Centralized and federated trust registries, and reputation oracles, resolve cross-party trust by consulting a shared service. This mechanism consults no registry, no directory, and no external node. The party carries the record itself, the receiving agent checks it against its own records and a declared currency interval, and nothing is looked up. Portability is achieved by transmission and self-verification, not by a lookup against a common authority.
Observational reputation inference reads latency passively and treats slowness as a party's property. The exoneration record is the output of an interventional design in which the observing agent varied its own dispatch, so the finding rests on a controlled comparison rather than on correlation. None of the foregoing categories carries an intervention-grounded, receiver-verified, expiring, effect-inert record of this kind, and nothing here should be read as asserting that any particular system infringes.
6. Disclosure Scope
The portable exoneration record is disclosed in U.S. Provisional Application No. 64/117,812 at Section 10.2, paragraph [0363], within the interventional-attribution family of paragraphs [0358] through [0363]. What is disclosed is the mechanism described here: the appending of an exoneration record identifying the matched class, both closure latencies, and the threshold applied upon a domain attribution; its transmission to the counterparty as a governed observation; and its admission by a further persistent semantic agent (100), as an input to a first-encounter determination alone, upon verification of authorship signature, feature-tuple identity of the matched class, and recorded time within a declared currency interval, with the admitted record recorded as a presented record in the counterparty identity record (114) that moves no scoped integrity vector (106) and increments no counter, and with a failed verification producing the not-determinable determination (126) and nothing adverse. Specific numeric parameter values and any particular deployment setting are not part of the operative disclosure stated here and are governed by the filed application and its declared policy objects. This article is a technical description of the filed disclosure and is neither a legal opinion nor an assertion that any product or party infringes.
