1. The certificate travels, the exposure stays home
A credential is a claim some other party made, at a time now past, about conduct the relying party cannot inspect. What carries weight in human practice, as that arrangement is commonly described, is the structure behind the document: a body that can be petitioned, a register that can be audited, a supervisor who stays answerable after the file goes wrong. The certificate points into that structure without being it.
Autonomous practitioners inherit the pointer alone. An agent asked to act where practice is gated on qualification can present an artifact that verifies cryptographically, establishing that the issuer signed it and nothing about whether that issuer is presently exposed should the underlying judgment turn out wrong.
Cost, on that arrangement, runs one way. An issuer can be cited at scale, in matters it never contemplated, and afterward stand where it stood before, while the party that relied carries the downside. The same transferability that makes a credential convenient is what lets an agent hold standing it was never corrected into.
2. Where attestation stacks and revocation lists stop short
Machine-readable qualification, as publicly described, keeps refining a single shape. Signed attestation formats are presented as addressing the integrity and portability of the claim. Status-list and revocation designs give an issuer a way to publish a negative after the fact. Capability and delegation token schemes hand a narrowed subset of an authority's powers to a delegate. Useful work, and none of it inverts who pays.
Attenuation in that family flows downward only. The filed disclosure characterizes the conventional arrangement as one in which cost falls upon the relying party while the party relied upon is protected, an attenuation upon delegation narrowing only the capability passed down. Insulating the source is sensible where it is a well-capitalized institution and the delegate a short-lived process, and reads differently where the source is itself an autonomous agent whose judgments are adopted at scale by parties it will never inspect.
Revocation answers part of the problem, late. A status change is a broadcast negative published after reliance, and it asks the issuer to conclude something was wrong, a judgment on the merits. Nothing in a status list, as those designs are publicly documented, meters adoption, so an issuer followed heavily and an issuer followed once look alike to it. A transferable score fails differently: a quantity built to be presented is also a quantity that downstream decisions can collapse onto. What none supplies is a mechanism under which being followed costs the followed party in its own capacity to act.
3. Deference records, counter-signature, and the merit-independent decrement
The filed chapter discloses that mechanism. A deferring agent holds a request to dispatch an action of an action class and finds its dispatch-authority predicate satisfied upon recomputation from its memory field: on its own record it is authorized to determine the matter itself. It elects instead to execute under a determination recorded in the lineage field of a second semantic agent, the followed agent. An agent whose predicate fails, whose authorization gate stands in the withheld state, or which has entered the non-executing cognitive mode does not defer at all; its holding is a withholding.
Before any deference record is given effect, the deferring agent applies the origin-disjointness test upon its own records, asking whether it and the followed agent are assignable to a common origin-equivalence class. Where they are, the record is appended as an attempted deference record and given no effect: it satisfies no dispatch-authority predicate, is transmitted to no party, decrements no authorization budget, enrolls no class, and appends nothing adverse. Declining is itself a positive abstention, and a fleet sharing an origin cannot credential itself.
Otherwise the agent retrieves a followed determination record: the determination as recorded, identifiers of the conduct evaluation artifact and its asserting party, the conduct descriptor asserted, the signed policy object in force at the recorded assertion time, and a tested-entry enumeration naming each lineage entry tested against the artifact and not contradicting it. Upon that record it performs a retrievability-form verification of three ordered conjuncts: continuity against its own counterparty identity record, the enumeration against a lineage commitment carrying a lineage-extent value and an accumulator root, and the determination for membership in the closed set with a matching action class. Form and retrievability are all it reaches: the agent retrieves none of the enumerated entries, weighs no sufficiency, re-derives nothing, and produces no determination concerning the followed agent's conduct. A failing conjunct yields a further attempted deference record naming what failed.
With both tests satisfied it appends a deference record carrying a counter-signature window field and a deference abstention entry, the latter a positive abstention naming the determination withheld from its own admission evaluator, and transmits that record as the first governed observation of a matched pair. Should the followed agent elect to counter-sign, it appends a deference counter-signature and by that append decrements its own authorization budget under a schedule declared in its signed policy object, neither agent having determined whether the deference was well founded.
4. Qualification rebuilt as exposure rather than presentation
Consider a practitioner agent authorized on its own record within an action class, yet holding almost no exposure there: few counterparties from whom a correction could ever reach it. Presented credentialing would have it fetch a certificate; here it acquires nothing it can show anyone, and defers.
A counter-signature, where the followed agent elects to give one, costs that agent capacity denominated in the same units that gate its own non-advisory dispatch. The declared schedule is a monotonically non-decreasing function of the counter-signatures it has appended within a declared window, bounded above by a declared maximum, so the marginal cost of being followed rises with the rate of adoption. In one disclosed linear form, the fifth counter-signature within a window costs twice the first. That budget stays replenishable by the agent's ordinary procedure and by no other, so being followed spends a quantity rather than ending an agent.
Responsive to that counter-signature, and not otherwise, the deferring agent retrieves from records the followed agent presents a corrector-class enumeration for the action class, admitted by a two-part filter. Correction: the records must show an accepted determination produced by the followed agent upon a conduct evaluation artifact implicating the action class and originating from an asserting party assignable to that class, so the class is recorded as having corrected that agent rather than as having interacted with it, rated it, or been introduced to it. Conformity: within an observation interval declared in the deferring agent's own policy object, the records must show a later conduct record of that agent evaluated against that determination and recorded as compliant. A class passing the first part and failing the second is excluded, the exclusion recorded. A further disclosed embodiment raises the second part to a declared minimum count of complied records.
What passes between the two agents is therefore no rating, weight, score, endorsement, vouch, or standing quantity, but an enumeration of parties recorded as having corrected the followed agent, each named by a class-defining reference both agents can evaluate against their own records rather than by a local identifier meaningless outside that agent. The deferring agent verifies those records under the same retrievability-form verification and appends an exposure edge record naming the enumeration and each excluded class with the filter part that failed, then dispatches to each enumerated class an exposure dispatch record: its own execution records, in the categories its counterparty scope object admits, declaring that it admits conduct evaluation artifacts from that class. Where the reference names no identity primitive for which it holds a dispatch address, no such record issues and the omission is appended, enrollment unaffected. Nothing transmitted describes the followed agent's standing.
Each enumerated class is written into the deferring agent's own replenishment register, for the action class and scope partition of the deference, as an admissible-but-unexercised entry. The entry has two effects and no third. An artifact from an assignable party becomes admissible to that agent's admission evaluator for the action class notwithstanding that it holds no prior encounter history with the party, and is not held in the pre-settlement inert state on that ground. For renewal the entry is not counted while unexercised, so the agent's quantities stand as they would have stood had it never been written. Enrollment buys reachability and no credit.
The rest waits on an artifact arriving. Assignability is settled by an ordered test upon the deferring agent's own records: direct membership, then definition membership against the relation type carried, then, upon plurality, by transitioning the entry earliest written, so one artifact exercises at most one enrolled class. On satisfaction the agent appends an exercised transition record binding the artifact to the class-defining reference, the exposure edge record, the deference record, and the determination adopted, and by the same append a deference correction linkage record keyed on that determination. Where the class is absent from the replenishment register for the declared window, an entry is written there and the authorization budget replenished, without regard to whether the artifact was well founded. Where no test is satisfied, a non-assignable-arrival record is appended, the artifact is not admissible on the ground of enrollment, and nothing is replenished.
Further embodiments in the same filing convert the event into standing that persists. Where an agent's count of determinations invariant across a policy succession sequence, and that succession's depth, both satisfy declared minimums, a standing-determination designation is written into the counterparty identity record, its whole effect being one substituted conjunct of the dispatch-authority predicate, confined to the designated action classes, counterparty, and partitions.
5. Running the mechanism, and the guarantees it withholds
Deference transfers no competence, and the filed disclosure says so. A deference records no corrective encounter, advances no compliance count of the deferring agent, and does not lengthen the decay constant applicable to that agent's authorization quantity for the action class. Where it has recorded no complied corrective encounter in the class, the class remains an untested class and its decay constant remains the short one declared in its policy object, however many deference records it has appended or counter-signed. Exposure can expire too: under a further disclosed embodiment an entry still unexercised upon expiry of a declared enrollment window is written to a lapsed value and ceases to confer admissibility, nothing adverse recorded.
A declination does not read as failure either. The followed agent may decline expressly, or by non-response within the counter-signature window carried in the record as received; where a record carries no such field, no window applies and non-response is not a declination. The declination is recorded as a not-determinable determination identifying the counter-signature as an unavailable input, and increments no counter of either agent. The deferring agent may still execute, appending an ungoverned reliance record and a reliance-bound narrowing record that narrows its own counterparty scope object for the followed agent by the categories its policy object declares narrowed upon ungoverned reliance. That narrowing is not a sanction, and it is released upon a counter-signature within the reinstatement window.
Adoption unwinds forward rather than backward. On a superseding or retracting determination the followed agent emits an adopted-determination supersession record; once the deferring agent appends it, the deference record satisfies no dispatch-authority predicate for further dispatch, while a dispatch already performed stands and becomes the fault of neither agent. Where two followed agents present determinations of differing outcome class upon one conduct descriptor, the deferring agent appends a deference conflict record, adopts neither, and either determines the matter itself or withholds.
Two limits belong in plain sight. Verification reaches form and retrievability, so a formally impeccable determination can still be substantively wrong; the architecture supplies not correctness but a record binding each correction thereafter received through the enrolled exposure back to the deference that earned it. No dashboard number sits at the end of it: the conversion bar forbids turning a deference abstention entry, a correction linkage record, or a count of either into a scalar value, a rating, or a threshold operand.
6. Disclosure Scope
The mechanisms described are disclosed in U.S. Provisional Application No. 64/117,812, principally Chapter 9, "Deference and Earned Exposure," at sections 9.1 through 9.10: the inversion of cost, the origin-disjointness test, the followed determination record and retrievability-form verification, the deference record and counter-signature window field, the merit-independent decrement, declined counter-signature and ungoverned reliance, non-renewal upon deference, enrollment of conformance-filtered corrector classes, the admissible-but-unexercised entry and exercised transition, and chained deference. Further embodiments appear at section 10.7, including determination invariance across policy successions, the standing-determination designation, bounded dispatch counts, repeated-conformity filtering, the exposure-dispatch omission branch, and enrollment-window lapse.
This article is published to establish public, timestamped prior art for applying this architecture to the credentialing of autonomous practitioners. It describes a pending application. Nothing here asserts that any product, service, standard, or organization infringes anything, and nothing here states that a license is required.
