1. Problem and Premise

Reputation, endorsement, and credential architectures move a quantity between parties: a rating, a score, a standing quantity, or a credential computed about the party relied upon and consumed by a party that never observed the conduct behind it. Conventional delegation and capability-attenuation architectures have a shape of their own. The cost of relying falls upon the relying party, the party relied upon is protected, and an attenuation upon delegation narrows only the capability passed downward, so being adopted costs the adopted nothing.

Section 9 of U.S. Provisional Application No. 64/117,812 inverts that direction. The metered consequence falls upon the followed agent, being the party such an architecture protects, and in the units of that agent's own capacity to dispatch. The exposure the deferring agent obtains is earned rather than assumed, written only upon a counter-signature that has been paid for and conferring admissibility alone until an artifact arrives and exercises it. A design that attenuates only downward, or that treats adoption as costless to the adopted, teaches away from this.

What passes between the two agents is therefore not an opinion about conduct. It is a way of recognizing the parties recorded as having corrected the followed agent and as having had those corrections thereafter conformed to, each identified by a fact both agents can evaluate against their own records.

2. Core Primitive: Deference as a Metered Charge on the Followed Agent

The core primitive is the deference record (900): an entry appended by a deferring semantic agent (100) binding at least the identity primitive of the followed agent and the followed determination relied upon, and transmitted to that agent for counter-signature.

The setup is narrow. The deferring agent holds a request to dispatch an action of an action class and finds its dispatch-authority predicate satisfied on recomputation from its memory field (102). It is on its own record authorized to determine the matter for itself, and elects instead to execute under a determination the followed agent recorded in its lineage field (104). An agent whose predicate fails, whose authorization gate (300) stands in the withheld state (310), or which has entered the non-executing cognitive mode (302) does not defer: its holding is a withholding, not a following.

The construction has three properties. The charge falls on the followed agent, which upon counter-signature decrements its own authorization budget (404), the units decremented being identical to the units gating its own dispatch. Nothing evaluative crosses: the deferring agent verifies form and retrievability, not merits. And what it receives is exposure, not credit: responsive to the counter-signature and not otherwise, it enrolls into its own replenishment register (402) the origin-equivalence classes (200) recorded as having corrected the followed agent and as having had those corrections conformed to.

3. Origin-Disjointness and the Retrievability-Form Verification

Two gates run in order, both on the deferring agent's own records.

The first is the origin-disjointness test. Under Section 2.2 the agent tests whether it and the followed agent are assignable to a common origin-equivalence class (200), evidenced by a declared relation type such as a shared dispatch lineage (202), a co-signature (204), or a common introduction path (206). Where they are, the record is appended as an attempted deference record and given no effect: it satisfies no predicate, decrements no budget, enrolls no class, and appends nothing adverse, the declining being a positive abstention.

The second gate operates on a followed determination record: the determination as recorded by the followed agent, identifiers of the conduct evaluation artifact (116) and its asserting party (118), the conduct descriptor, an identifier and successor index of the signed policy object (112) in force at the recorded assertion time, and a tested-entry enumeration naming each entry of that agent's lineage field (104) tested against the artifact and not contradicting it, each by position, recorded time, and continuity hash. Against it the agent performs a retrievability-form verification of three ordered conjuncts:

  • The authority credential and continuity hash field are verified against the continuity history store in the deferring agent's counterparty identity record (114) of the followed agent.
  • The enumeration is verified against a lineage commitment comprising a lineage-extent value N, the count of entries in that lineage field (104) at emission, and a lineage accumulator root over the ordered entries, both bound into the successor hash field. The enumeration must be non-empty, each position p must satisfy one not greater than p and p not greater than N, positions must be pairwise distinct and strictly increasing, and recorded times non-decreasing.
  • The determination must be a member of the closed set of Section 1.8, and the action class carried must match that of the pending dispatch.

The verification is of form and of retrievability and not of merits. The deferring agent does not retrieve the enumerated entries, evaluate their sufficiency, re-derive the determination, or compare it against one its own admission evaluator (120) would have produced, and it produces no determination concerning the followed agent's conduct. Where a conjunct fails, the deference is appended as an attempted deference record naming the failing conjunct, again a positive abstention with nothing adverse appended.

4. Counter-Signature and the Merit-Independent Decrement

With both tests satisfied, the deferring agent appends a deference record (900) carrying the identity primitive of the followed agent, the followed determination record as retrieved, the tested-entry enumeration without alteration, a deference abstention entry naming the determination withheld from its own admission evaluator (120), the scope-partition identifier of the deference, the result of each of the two tests, and a counter-signature window field (912).

The counter-signature window field (912) carries the window declared in the deferring agent's policy object, with the reinstatement window and the successor index of each; absent the field, no window applies and silence is not a declination. The abstention entry falls under the conversion bar (502) of Section 5: no stage converts it into a scalar, a default, or a threshold operand.

The followed agent, receiving it as a first governed observation (608) of a matched pair (600), verifies it against its continuity history store and confirms that the record names a determination in its own lineage field (104). Electing to counter-sign, it appends a deference counter-signature (902) as the second governed observation (610), decrementing its own authorization budget (404) by that append.

The deference decrement schedule declares an amount not less than that declared under Section 4.2 for issuance of a conduct evaluation artifact (116) by that agent, monotonically non-decreasing in a count of counter-signatures appended within a declared window, so the marginal cost of being followed rises with the rate of adoption. Two forms are disclosed, b being the base amount and n the count already appended. In a step form the amount is b for each of the first k counter-signatures and m multiplied by b thereafter, m a declared multiple not less than unity. In a linear form it is b plus n multiplied by a declared increment i, and in one embodiment b is one unit and i one quarter of one unit, so the first counter-signature costs one unit and the fifth costs two. Each form is bounded above by a declared maximum, and the decrement occurs without any adjudication, by either agent or any other party, of whether the deference was well founded.

Where the decrement drives the budget to a declared floor, the followed agent's authorization gate (300) goes to the withheld state (310) for an enumerated set of action classes and the escalation record is emitted to its principal. Being followed spends a replenishable quantity and does not terminate the agent: the budget is replenished under Section 4.5 alone, which counter-signing neither accelerates nor diminishes.

A chained deference record carries the record and counter-signature under which the first agent adopted the determination, a chain depth one greater, and each agent already in the chain. The chaining agent decrements its own budget, the originating agent's is not decremented a second time, and a record exceeding a declared maximum depth or naming the agent applying it is given no effect. Where two followed agents present determinations of differing outcome class upon one conduct descriptor, a deference conflict record is appended and neither is adopted.

5. Declination, Ungoverned Reliance, and Non-Renewal

The followed agent may decline to counter-sign, by express declination or by non-response within the counter-signature window carried in the record as received. A declination is not a denial of the deferring agent: it is recorded as a not-determinable determination (126) identifying the counter-signature as an unavailable input, increments no counter of either agent, and is not treated at any stage as a rejected determination (124) or as an accepted determination (122).

Where the counter-signature is declined or absent, the deferring agent may nonetheless execute under the adopted determination, appending an ungoverned reliance record naming the deference record (900), the absent counter-signature, the window applied, and the execution performed, and writing a reliance-bound narrowing record that narrows its own counterparty scope object held for the followed agent by the categories its policy object declares as narrowed upon ungoverned reliance. That narrowing is not a sanction, and it is released upon a counter-signature within the reinstatement window.

Deference renews nothing on the deferring agent's side. Appending a deference record (900), receiving a counter-signature (902), and executing under an adopted determination each renew no authorization quantity and replenish no authorization budget (404) of the deferring agent, and no entry is written to its replenishment register (402).

6. Corrector-Class Enrollment and the Earned Correction

Responsive to receipt of a counter-signature (902), and not otherwise, the deferring agent retrieves, from records the followed agent presents, a corrector-class enumeration (904) for the action class of the deference.

That enumeration is the set of origin-equivalence classes (200) each recorded both as having corrected the followed agent in that action class and as having had its correction thereafter conformed to, membership running through a two-part conformance filter (906). The first part is correction: an accepted determination (122) produced by the followed agent upon a conduct evaluation artifact (116) implicating the action class and originating from an asserting party (118) assignable to that class. The second is subsequent conformity: within an observation interval, a later conduct record of that agent recorded as compliant with that determination. A class satisfying the first and not the second is excluded, and no rating, score, or endorsement is consumed.

A class is identified not by the followed agent's local class identifier, meaningless outside that agent, but by a class-defining reference: the recorded fact defining the class under Section 2.2 that both agents can evaluate against their own records. It carries a relation-type designator, the defining member of that type, and the identity primitive of each asserting party (118) assigned to the class. Where the defining member is a lineage entry, it is named by position and continuity hash and verified against the presented lineage commitment, so that a class cannot be defined by an entry the committed lineage does not contain.

The deferring agent appends an exposure edge record (910) naming the followed agent, the counter-signature (902), the action class, the enumeration with each class-defining reference, and the failing filter part for each excluded class. It then dispatches to each enumerated class an exposure dispatch record: its own execution records within the action class, in the disclosure categories its counterparty scope object admits, declaring that it admits conduct evaluation artifacts (116) from that class, sent to each identity primitive in the reference, being the address set of the class. Nothing transmitted describes the followed agent's standing, and the deferring agent receives no rating from those parties.

Each class is written into the replenishment register (402), for the action class and scope partition of the deference, as an admissible-but-unexercised entry keyed on that reference. Such an entry has two effects and no third. For admissibility, an artifact from a party assignable to the class reaches the admission evaluator (120) despite no prior encounter history, and is not held in the pre-settlement inert state. For renewal, it counts for nothing under Section 4.5 while unexercised.

Whether a later-arriving asserting party (118) is assignable to an enrolled class is decided by an ordered test over each entry held for the action class and scope partition of the arriving artifact: direct membership, where the party's identity primitive is in the enumeration carried; definition membership, where the defining member is satisfied against the agent's own records under the relation-type designator carried; and, upon plurality, transition of the earliest-written entry alone, each further assignable entry being named without transition, so one artifact exercises at most one class. Each test runs on the agent's own records and the arriving artifact, without reference to a centralized registry and without coordination with a further execution node. Where none is satisfied, a non-assignable-arrival record is appended naming the artifact and the failing test, and nothing is transitioned or replenished.

When an admissible artifact arrives, the agent appends an exercised transition record (908) naming the artifact, the entry transitioned, the exposure edge record (910), the deference record (900), the counter-signature (902), and the determination adopted. By the same append a deference correction linkage record is written, keyed on the adopted determination, so the agent can enumerate, for a determination it adopted, each correction thereafter received through the exposure enrolled upon it. That record is of its own encounters, produces no determination concerning the followed agent, and falls under the conversion bar (502).

The artifact so received is the deferring agent's own corrective encounter, not the followed agent's. Upon the append, and where the class is absent from the replenishment register (402) for the declared window, an entry is written and the deferring agent's authorization budget (404) is replenished under Section 4.5, whether or not the artifact was well founded.

7. Operating Parameters

Each quantity below is declared in a signed policy object (112) and recorded with the operation that applied it. The specification recites forms and relations; the values are policy-declared.

  • Counter-signature window and reinstatement window, carried in the counter-signature window field (912) with the successor index of the object from which each is drawn. Absent the field, no window applies.
  • Deference decrement schedule: a base amount b not less than the Section 4.2 issuance amount; a form, step or linear; the step count k and multiple m not less than unity, or the linear increment i; and a declared maximum.
  • Budget floor and ceiling: the floor at which the authorization gate (300) goes to the withheld state (310), and the authorization budget ceiling (408), excess being discarded.
  • Observation interval for the conformance filter (906): that declared for the action class in the deferring agent's own policy object, otherwise that declared for the followed agent where not longer than a declared maximum.
  • Replenishment window of the replenishment register (402), and the budget replenishment schedule (406) amount, greater than zero and less than the per-increment decrement of Section 4.2.
  • Chain depth maximum, and the narrowing categories applied upon ungoverned reliance.

8. Alternative Embodiments

Section 10 extends this chapter along four lines.

Edges, severance, and a valuation-free standing quantity. An edge is a persisted dyadic relation between the agent and one counterparty, constituted by that counterparty's counterparty identity record (114) and each lineage entry recording an interaction; a severance event is a lineage entry recording a structural change terminating the two parties' capacity to interact. For a payment class and an obligation class the agent computes a severance-survival state as not-tested, survived, or not-survived, structurally and never from message content or declaration. From that pair alone an ordered procedure resolves the edge's reason-type as payment-sustained, obligation-sustained, want-sustained, or not-typeable; the asserting party never declares it. A per-reason-type weighting multiplied by a standing multiplier forms the effective weighting applied to the magnitude by which an accepted determination (122) modifies the scoped integrity vector (106) and the self-esteem aggregate (108), raised to a declared non-zero minimum where the type is not-typeable, so an untested edge still moves state but by less than a severance-tested one. The standing quantity behind that multiplier is ordinal: direction of initiation, latency as a median reciprocation interval, and regularity as a median absolute deviation of successive intervals, each mapped through two declared breakpoints and summed, no cardinal magnitude of benefit being computed or stored.

Correction-exposure decay of authorization. An authorization quantity per action class and scope partition decays under a monotonically non-increasing function parameterized by a declared decay constant, applied on each execution and each successor-epoch advance; at a declared floor the gate goes to the withheld state (310). Renewal turns on a register of origin-equivalence classes from which an admissible implicating artifact arrived within a declared window: a class absent renews the quantity, a class present renews nothing, and all four determination classes renew equally, so renewal is valence independent.

Interventional attribution and portable exoneration. The agent separates counterparty-caused from domain-caused latency by varying its own dispatch, routing each interaction in a matched class through a first closure path upon a receipt the counterparty controls or a second upon one it does not, the paths assigned by alternation or by a seeded selection committed to the lineage field (104) beforehand. The differential is the first latency, normalized against a channel baseline of qualifying first-path durations, less the second. An ordered branch yields not-determinable below a declared minimum volume, attribution to the counterparty above a declared threshold bounded below by the channel's own standard deviation, attribution to the domain with a portable exoneration record where the second latency exceeds a second-path baseline by a declared margin, and otherwise no adverse signal.

Deference extensions. An invariance count is computed by re-resolving each retrieved determination against every object of a policy succession sequence in version order, a determination joining the policy-invariant set only where its outcome class is unchanged. Where the count and the succession depth satisfy declared minima, a standing-determination designation is written into the counterparty identity record (114), substituting inside the dispatch-authority predicate so that the conjunct otherwise requiring an accepted determination (122) from the agent's own admission evaluator (120) may instead be satisfied by a verified counterparty determination, confined to that conjunct and to the designated classes and counterparty. Further variants: a bounded count of dispatches per deference record; an adopted-determination supersession record, after which the record satisfies no predicate though a dispatch already performed is a fault of neither agent; a filter requiring a declared minimum count of complied conduct records; lapse of an unexercised entry at a declared enrollment window; and elective exposure enrollment at an agent's own cost. Expressly, a deference records no corrective encounter and does not lengthen the decay constant, so an action class with no complied corrective encounter remains an untested class.

9. Composition With the Rest of the Architecture

This chapter is the last of nine and draws on most of them. Section 2 supplies the origin-equivalence class (200) on which the origin-disjointness test and the class-defining reference depend. Section 4 supplies the authorization budget (404), the replenishment register (402), and the replenishment procedure, the only path by which either agent's budget rises. Section 5 supplies the conversion bar (502), which keeps abstention entries and linkage-record counts from becoming a scalar. Section 6 supplies the matched pair (600), and Section 1 the closed determination set the third conjunct tests.

The dependency also runs outward: the deference decrement schedule declares an amount not less than that Section 4.2 declares for issuance of a conduct evaluation artifact (116), and the classes enrolled here are written into the register Section 4.5 consults, counting for nothing there until an artifact exercises the entry.

As to the broader persistent semantic agent family, the filing incorporates no application by reference in its entirety; incorporation is per element and for a stated limited purpose only. The flat, invariant, and closed framings of the base architecture are conveniences of exposition and not definitions of the shared primitives, which the commonly owned applications state as optional, alternative, and open.

10. Prior-Art Distinctions

Reputation and feedback systems. Feedback scores and endorsement counts compute a quantity about a party from other parties' observations and publish it for consumption. Here no quantity is produced, retrieved, admitted, or consumed; what crosses is a way of recognizing parties, not an opinion about one.

Centralized trust registries and cross-certification. Certificate authorities and federation directories make a third party the arbiter of who is recognized. The origin-equivalence assignment, the assignability tests, and the retrievability-form verification each run on the agent's own records, and a class cannot be defined by an entry the committed lineage does not contain.

Arbitration and dispute-resolution services. Arbitration determines merits: a tribunal or trust-and-safety function decides whether a party was right. The retrievability-form verification is expressly not that, and the decrement occurs without adjudication of correctness.

Credit scoring and bureau models. Credit bureaus aggregate conduct across creditors into a scalar consumed by parties who never observed it. The conversion bar (502) forbids that accumulation: abstention entries, linkage records, and counts of them become no scalar, default, rating, or threshold operand.

Capability attenuation and delegation. Attenuated-capability constructions narrow what passes downward and leave the delegating party untouched, so the cost of relying falls on the relier. This filing charges the followed agent in the units of its own dispatch capacity.

Staking and slashing. Consensus systems bond value and confiscate it upon a protocol violation. The decrement here is conditioned on no violation: it is the cost of being followed, charged whether the adopted determination was correct or not.

11. Disclosure Scope

This article discloses the subject matter of Section 9 of U.S. Provisional Application No. 64/117,812, "Governed Social Conduct of Persistent Semantic Agents Through Record-Grounded Admission of External Conduct Evaluations and Self-Limiting Refusal," with the additional embodiments of Sections 10.1, 10.2, and 10.7 and the defined terms of Section 11 for that chapter.

It is directed to the deference record (900) and its counter-signature (902); the origin-disjointness precondition and the three-conjunct retrievability-form verification against a lineage commitment; the merit-independent decrement of the followed agent's own authorization budget (404) under a schedule non-decreasing in the rate of adoption; declination, ungoverned reliance, and the reliance-bound narrowing; the two-part conformance filter (906) and the corrector-class enumeration (904) identified by class-defining reference; the exposure edge record (910), the admissible-but-unexercised entry, and the exercised transition record (908); and the alternative embodiments recited above.

It is not directed to any particular signature scheme, hash construction, transport protocol, or policy content, each an interchangeable consumer or producer of the disclosed structures, and it states no latency figure, throughput guarantee, or benchmark. The parameters in section 7 are the policy-declared quantities the specification recites, not claim limitations, and the alternative embodiments do not exhaust the admissible space. Nothing here characterizes any commonly owned application as prior art, and nothing asserts that any product, protocol, or publication infringes any claim of any application.