1. The Gap

An agent that decides whether an accusation against it is admissible at all holds the gate through which its own accountability passes. In U.S. Provisional Application No. 64/117,812 that gate is the declared value set carried by the signed policy object (112). Before testing whether asserted conduct occurred, the admission evaluator (120) extracts a conduct descriptor from the conduct evaluation artifact (116) and applies a three-part membership test against each value-scope tuple: the action class in the tuple's action-class identifiers, the scope partition in its scope-partition identifiers, and the affected-party class mapping to its empathy-scope designation, an identified counterparty in a counterparty identity record (114) mapping to interpersonal, the agent (100) itself to personal, an unrecorded party to global. A value is implicated only where all three hold; where none is, the evaluator produces the not-applicable determination (128) at no cost.

That ordering puts the declared value set upstream of everything else the architecture asks. The backward route is closed by the retroactive narrowing bar: the admission evaluator (120) resolves an assertion against the signed policy object (112) in force at the recorded assertion time of the artifact, resolving the policy reference field (110) by canonical alias to candidates, retaining one whose validity window encloses that time, discarding one revoked as of that time, and applying an anti-rollback floor in the memory field (102) that maintains a monotonically non-decreasing successor index per alias. A successor admitted after the assertion time bears a later validity window and is not retained, whereby the agent (100) is incapable of narrowing its declared value set after receipt of an artifact so as to render that artifact inapplicable.

Forward motion is another matter. A successor admitted while an assertion is still undetermined does not reach back to that assertion, and it need not be defective in any respect the resolution procedure tests for: it verifies against the principal's signature, carries its own validity window, and stands at or above the anti-rollback floor for its alias. Paragraph [0332] addresses that case. Asking instead whether the contraction was justified runs into what the filing's background records about merits-conditioned regimes, where consequence attaches only after a determination made by an institution other than the penalized party.

2. Mechanism

Identity before comparison. Contraction has to be identifiable before anything can be scored, and the filing resolves identity at the level of the value rather than the object. Each declared value of the signed policy object (112) carries a declared-value identifier assigned by the principal and covered by the object's signature, stable across successor objects admitted under one canonical alias of the policy reference field (110). A successor value corresponds to a value of the immediately preceding object where and only where the two carry an identical identifier. A value whose identifier is carried by no successor value is excluded; a scope element dropped from a corresponding tuple is contracted. Because the identifier is an identity of the value and not a scope element, altering a tuple leaves it unchanged, and both forms are detected by identifier correspondence rather than by comparing full value sets.

Detection. The architecture detects, from the append-only lineage field (104), admission of a successor signed policy object (112) that excludes a declared value, or that drops an action-class, scope-partition, or empathy-scope element from a value-scope tuple corresponding by declared-value identifier.

Pendency, and the window. Detection alone is not scored. The admission must follow an as-yet-undetermined conduct evaluation artifact (116) implicating that value or element, and must occur within a window declared in the policy object in force at the artifact's assertion time rather than in the successor.

Scoring. Responsive to detection, the architecture scores a deflection event and increments the refusal counter (304), without adjudicating whether the contraction was well founded. That counter is the structure into which Section 3 meters refusals: resident in the memory field (102) of the semantic agent (100) and comprising a rate counter (306) and a run counter (308).

Merit independence is defined terminologically in the filing. Merit-independent metering means incrementing of a counter responsive to a determination class or to an issuance, without any determination being performed of whether the refusal, contraction, or issuance was well founded. Contraction sits in that definition beside refusal and issuance. Negative limitations attached to the increment procedure of Section 3.2 and the gate write of Section 3.5 carry the same commitment for the counter receiving the increment: no adjudication by the principal or any other party is solicited, received, or awaited; no increment turns on a finding that an artifact was unfounded, frivolous, abusive, or submitted in bad faith, nor on the absence of one; none turns on review by a reviewing body, arbiter, court, regulator, or out-of-court dispute settlement body, or on forfeiture, posting, or resolution of a stake, bond, deposit, or fee. No field of the refusal counter (304) holds a merits outcome.

The asymmetry. A successor that adds a value or element is scored as no deflection and is given no retroactive effect. The filing states the result of that pairing directly: contraction under a pending assertion is priced, and broadening is neither priced nor retroactive.

3. Declared Conditions

Paragraph [0332] and the substrate provisions it composes on declare the following. No numeric value for the window appears.

  • Detection source. The append-only lineage field (104) of the agent (100) itself.
  • Window placement. The admission occurs within a window declared in the policy object in force at the artifact's assertion time, rather than in the successor being admitted.
  • Pendency. The implicated conduct evaluation artifact (116) is as yet undetermined when the admission occurs.
  • Correspondence rule. A successor value corresponds to a value of the immediately preceding object where and only where the two carry an identical declared-value identifier.
  • Contraction forms. Exclusion of a declared value, or dropping of an action-class, scope-partition, or empathy-scope element from a corresponding value-scope tuple.
  • Counter thresholds. The rate threshold is an integer greater than one, and the run threshold is an integer greater than one and not greater than the rate threshold, each retrieved from the signed policy object (112) in force.
  • Metering window expression. An interval of wall-clock time, a count of conduct evaluation artifacts (116) received, or a count of successor epochs of a dynamic agent hash chain. In the successor-epoch embodiment the window is advanced by no party other than the agent (100) and not by manipulation of a clock available to the hosting execution node.
  • Adjudication. No determination of whether the contraction was well founded is performed at any step.
  • Broadening. Scored as no deflection, and given no retroactive effect.

4. Composition

Within Section 10.1 the paragraph sits alongside value-scope resolution, the declared-value identifier, and the retroactive narrowing bar. The filing states that each mechanism of Section 10 is a distinct embodiment practicable independently or in combination with any other, so the grouping is compositional rather than indivisible.

Downward, the increment lands on a counter inside the metered party. Once both thresholds are satisfied, the refusal counter (304) writes the authorization gate (300) to the withheld state (310) for those action classes recorded in the append-only lineage field (104) as implicated by the conduct descriptors of the artifacts that produced increments within the current window, a class not so recorded not being enumerated. Concurrently the agent (100) enters the non-executing cognitive mode (302) for those classes and the escalation emitter emits an escalation record to the principal. Faculty metered and faculty spent stay distinct: the write withholds execution alone, and the agent continues in the withheld state (310) to receive artifacts, retrieve entries, produce determinations, and append them. Those determinations decrement neither accumulator and do not return the gate to granting. The withheld state (310) persists irrespective of elapsed time, and a restoration controller acts only upon a procedure appended to the lineage field (104).

Sideways, deflection scoring recurs wherever a party withdraws from an exposure it previously accepted. Under the co-signed mutual admission compact of Section 10.9, a successor policy object or compact record excluding a reciprocal value-scope tuple within a declared window following the assertion time of a pending artifact implicating it is scored against the narrowing party alone; the other party is unaffected, and dissolution outside that window or by a co-signed dissolution record is scored against neither. Lapse of the compact validity window inside the window, no successor enumerating the tuple admitted, is scored as an exclusion against the party that declined to admit one. In the adoption path of Section 10.6, an agent abandoning an adopted value within a declared adoption window of a pending artifact is scored on its own refusal counter (304) alone.

One boundary deserves stating plainly. Declaring few values carries no charge: where no declared value is implicated, the admission evaluator (120) produces the not-applicable determination (128) at no cost, applying no increment and not resetting the run accumulator.

5. Prior-Art Distinction

Merits-conditioned accountability regimes. Article 23(2) of Regulation (EU) 2022/2065 directs suspension of the processing of notices from a notifier that frequently submits manifestly unfounded notices, the unfoundedness adjudicated by the platform. Section 623(a)(8)(F) of the Fair Credit Reporting Act permits a furnisher to decline to investigate a dispute reasonably determined frivolous or irrelevant. Model unfair-claims-practices provisions of the National Association of Insurance Commissioners condition consequences for assertive conduct on an institutional finding of unfair practice, and bonded challenge procedures forfeit a stake upon a determination that the challenge failed. The structural distinction: consequence attaches only after a merits determination performed by an institution other than the penalized party, and the faculty suspended is the one exercised.

Policy-enforcement systems that return a decision. Governance systems for autonomous agents evaluate a proposed action against policy and return a decision, systems implementing the XACML standard drawing that decision from permit, deny, not-applicable, and indeterminate. That result is directed at the request: not-applicable and indeterminate are target-match outcomes of a single access decision, returned to the requesting party, producing no persistent state in the deciding entity, imposing no cost, and constraining no subsequent operation of the decider. Here the inapplicability outcome and the policy revision producing it live in persistent state of the deciding party and bear on its own later operation.

Capability attenuation. Schemes of the macaroon family confine a delegate by appending caveats along the forward path of a delegation, each holder able to attenuate but not to expand the delegated capability. They protect the owner of the delegated resource, operate forward only, and attach no consequence to a party that configures itself to exclude what it would otherwise inherit back. The arrangement here runs the other way, scoring a party that contracts what it holds itself to while an assertion is pending.

Auditor-held verdicts, attestation, and external revocation. Accountability protocols of the PeerReview family prove node misbehavior from tamper-evident logs, the verdict held by auditors: witnesses collect evidence, third parties verify the proof, and the accused node makes no governed determination of its own. Remote attestation architectures of the kind described in RFC 9334 locate appraisal of evidence in a verifier and the grant or denial of authority in a relying party, the attested entity holding neither the appraisal function nor any metered faculty. External revocation systems end an agent's ability to act by an act of a controlling authority, the agent's own record playing no part. Deflection scoring requires the opposite arrangement: metered faculty, detecting record, and receiving counter inside one party, the metering complete upon the append.

6. Disclosure Scope

The prospective narrowing bar and deflection scoring are disclosed in U.S. Provisional Application No. 64/117,812, Section 10.1, at paragraph [0332], with value-scope resolution at [0329], the declared-value identifier at [0330], and the retroactive narrowing bar at [0331]. The refusal counter (304), its fields and thresholds, the merit-independence limitations, and the write of the authorization gate (300) are disclosed in Section 3, and the definition of merit-independent metering in the terminology of Section 11. Companion deflection scoring for the co-signed mutual admission compact appears in Section 10.9 and for adopted values in Section 10.6.

Disclosed here is the structure: detection, from the append-only lineage field (104), of an excluding or contracting successor signed policy object (112) by declared-value identifier correspondence; the pendency and window conditions, including placement of the window in the object in force at the assertion time; the scoring of a deflection event; the increment to the contracting party's own refusal counter (304); the absence of any merits adjudication as a condition; and the treatment of a broadening successor.

Disclaimed is everything beyond that. Nothing here states a position on the scope of any claim that may issue, and no claim scope is admitted, narrowed, or surrendered by any characterization above. Nothing here grants a license, express or implied, under any pending or issued patent. Nothing here asserts that any named system, standard, regulation, protocol, or party infringes any claim; the prior-art categories are described to distinguish structure, not to characterize any product. No numeric value for the window is stated because paragraph [0332] declares none, and the counter thresholds are recited as policy-carried integers.