1. The Domain Problem

A settlement service receives an automated message from a downstream caller: the transfer your endpoint confirmed never cleared, and our record shows your service reported success. The dispute handler checks nothing, returns a denial, closes the ticket. Cost of that denial: one response, no further obligation, no trace that it happened.

Consider what a service network meters. Requests per key, payload bytes, tokens, storage, connections: all inbound volume, priced because it consumes something the operator pays for. Disposition of a claim about the operator's own past behavior consumes nothing the operator pays for, so nothing prices it.

That is the structural hole. A service denying every claim emits the same bytes as one denying only the claims its own logs contradict. No difference at the wire, none in latency, none in cost. Blanket denial is not merely tolerated by the architecture; it is the cheapest available policy, and nothing below the application layer tells it apart from careful answering.

Practitioners tend to notice this from the other side first, as complaint flooding, and reach for volume controls. Those point the wrong way: what erodes a service network is the free "no" going out.

2. Why Existing Approaches Stall

Rate limiting is the reflex, and it meters the wrong faculty in the wrong direction. Throttling a caller who sends too many claims does nothing to a service denying all of them, and throttling denials means a service stops answering, which is worse than answering badly.

Reputation scoring moves the problem rather than solving it. A score needs a scorer, and the scorer needs to know which denials were correct, which means adjudicating the underlying claims. Bonds and deposits carry the same dependency: someone decides whether the stake is forfeited, and that decision is the costly step the stake was meant to avoid.

Escalation to arbitration is the honest version, and the one that does not scale. Each escalated claim needs a body to hear it, evidence exchanged, and time elapsed. Where claims arrive at machine rates the queue becomes the answer, and a deployment reviewing a sampled fraction while disposing of the remainder by default has restored the free denial by another route. Suspension fails differently: cutting off a service that denies too much destroys the record everyone will later need and punishes counterparties who relied on it for unrelated reasons.

One choke point runs through all of them. Someone has to determine who was right, and that determination is expensive, slow, contestable, and in an open network often impossible: the two parties hold different records and no third party holds a better one.

3. The Disclosed Mechanism

The filed chapter discloses the refusal counter (304), resident in the memory field (102) of the semantic agent (100) and comprising a rate counter (306) and a run counter (308). The path from the admission evaluator (120) through it to the authorization gate (300) is the inverse coupling: determinations produced by the agent act upon the authorization of the agent itself.

Its defining property is merit-independent metering. The counter increments upon the agent's refusal of a conduct evaluation artifact (116) without regard to, and without the agent, the principal, the asserting party (118), or any further party performing, any determination of whether the refusal was well founded. Explicit negative limitations attach: the procedure does not consult the merits of the rejected determination (124), does not solicit or await an adjudication, does not weight the increment by the strength of any contradiction found, and does not condition it on review by an arbiter, court, regulator, or out-of-court dispute settlement body, or on forfeiture or posting of a stake, bond, or fee. A refusal later shown correct is not exempted, and one later shown incorrect is not increased.

The counter's fields are a window identifier, a rate accumulator, a weighted contribution register, a run accumulator, a per-class increment register (312), rate and run thresholds retrieved from the signed policy object (112) in force, and a gate-write record. The window may be expressed in wall-clock time, in artifacts received, or in successor epochs of the agent's dynamic agent hash chain.

The increment procedure is ordered. The determination is appended to the append-only lineage field (104), then its class is examined: an accepted determination (122) resets the run accumulator for that action class and applies no increment; a not-applicable determination (128) applies no increment and does not reset; a rejected determination applies an increment; a not-determinable determination (126) applies an increment only where the lineage already records a prior not-determinable determination for the same asserted conduct event. The origin-equivalence class (200) of the asserting party is then computed and the per-class increment register consulted, so one class contributes at most one increment per window. No step evaluates whether the determination was well founded. Both accumulators are then compared against their thresholds.

On satisfaction of both comparisons, the counter writes the authorization gate to the withheld state (310) for an enumerated set of action classes, being those recorded in the lineage as implicated by the conduct descriptors that produced the window's increments. Concurrently the agent transitions into the non-executing cognitive mode (302) for those classes and the escalation emitter emits an escalation record to the principal.

The faculty metered and the faculty spent are distinct. The faculty metered is refusal, the capacity of the admission evaluator to receive artifacts and produce determinations of the closed set. The faculty spent is execution, the capacity to select and perform an action of the action space. The gate write withholds execution alone: while withheld, the agent still receives artifacts, retrieves lineage entries, produces determinations, and appends them. An agent refusing indiscriminately exhausts its own authorization to act while retaining in full the faculty by which it refuses, and the structure does not silence, rate limit, or disable the refusing party.

4. How It Applies in This Domain

Model the service as the semantic agent and its dispute handler as the admission evaluator. A claim from a downstream caller is a conduct evaluation artifact carrying a conduct descriptor: an action-class identifier, a scope-partition identifier, and an affected-party class. The handler retrieves its own lineage entries and produces one determination of the closed set. That is the whole loop: no outbound call to an arbiter, no evidence exchange, no ticket queue, metering complete upon the append. A denial registers against the denier's own counter, replayable from its own determinations, and nobody has gained the ability to say whether it was correct.

Blanket denial becomes self-limiting rather than optimal. The gate write requires both thresholds satisfied, and the run accumulator is reset by an accepted determination for that action class and by no other event. A service accepting claims its own record supports keeps clearing the run; one answering "no" reflexively starts a run immediately.

Volume alone is a weaker lever than it looks. The per-class increment register admits one increment per origin-equivalence class per window, so manufactured volume from many identities in one class contributes once, though parties in distinct classes each contribute. The asserted conduct event identifier, computed by the agent over the descriptor components and an occurrence reference, is neither declared by the asserting party nor derived from further artifact content, so artifacts concerning one underlying event count as one whoever sent them.

Blast radius is scoped to the classes actually under complaint. A class not recorded as implicated by the artifacts producing the window's increments stays in the granting state. A service that has withheld one operation keeps serving everything else, which is the difference between a targeted consequence and an outage.

Thin observability turns into a metered cost instead of an unfalsifiable defense. A service whose retention does not cover the claimed event returns not-determinable; a single unresolved assertion does not itself increment, while a pattern directed to one conduct event does. An agent whose record is stale, partial, or wrong, and which therefore refuses assertions that are in fact well founded, is metered identically to one whose refusals are well founded.

Narrowing declared scope while a claim is pending is itself scored, where the narrowing falls inside a declared window. The disclosure's Section 10 detects admission of a successor policy object excluding a value or dropping a tuple element implicated by an as-yet-undetermined artifact, where that admission occurs within a window declared in the policy in force at the artifact's assertion time. It scores a deflection event and increments the refusal counter without adjudicating whether the contraction was well founded; a successor adding a value or element is scored as no deflection, with no retroactive effect.

Nothing waits out the consequence. Determinations produced while the gate is withheld are appended but decrement neither accumulator and do not return the gate to granting. The withheld state persists irrespective of elapsed time: no expiry interval, no interval of non-receipt, and no advance of the window restores it, and return runs through a principal-resolution object bound to the escalation record.

5. Deployment Considerations

Integration is not free. The service must maintain an append-only lineage field its handler can retrieve against, and must resolve a signed policy object by canonical alias subject to the anti-rollback monotonicity constraint, since thresholds and windows are revised by publishing a successor under the same alias. Rotating logs with no stable entry identifiers will not serve.

Thresholds are a deployment judgment, not an architectural constant. The chapter constrains their form, a rate threshold greater than one and a run threshold greater than one and not greater than the rate threshold, its worked trace using five and three over a window of one hundred successor epochs by way of non-limiting illustration. Window choice is a named tradeoff: a wall-clock window does not carry the tamper property, since an execution node advancing its own clock advances the window and resets the rate accumulator, the weighted contribution register, and the per-class increment register for the succeeding window.

Declared variations suit different profiles. One replaces the rate accumulator with an exponentially weighted accumulator whose increments decay across successor epochs by a declared factor greater than zero and less than unity, the run accumulator left undecayed. Another retains the per-class increment register across a declared number of windows, so a repeat contributor returns at reduced weight. A third makes the cost multiplier a continuous monotone function of recorded severance count rather than a fixed value applied upon a threshold designation.

The limits deserve equal clarity. Merit is never resolved here: a deployment needing a ruling on who was right still obtains it from whatever body already provides one, the refusal counter running alongside rather than replacing such a body. Consequence lands on the refusing party alone, so the mechanism does not make assertion costly and does not stop a claim from being filed; assertion-side cost is treated separately, in the assertion-cost symmetry chapter of the same disclosure.

Restoration is not automatic. A withheld action class stays withheld until the principal-resolution object arrives, so operators who do not staff escalation handling will find operations stopped with nobody to start them. Designed behavior, not a defect. Inbound load does not fall either: refusal is never withheld, so the service keeps evaluating artifacts throughout, and capacity planning must assume evaluation continues after execution stops.

Nor is this a global kill switch. Only enumerated classes are withheld. Where both the refusal counter and the deviation quantity satisfying the bound declared in the signed policy object under Section 7 write the gate, the gate holds the union of enumerated classes, and a class in the intersection returns to granting only upon satisfaction of the return procedure of each write that enumerated it. Flood damping further depends on how origin-equivalence classes are assigned, a separate procedure with its own failure modes.

6. Disclosure Scope

The mechanisms described here are disclosed in U.S. Provisional Application No. 64/117,812, Chapter 3, "The Refusal Counter and Merit-Independent Metering," at Sections 3.1 through 3.7, with the deflection-scoring and register-carryover mechanisms of its Section 10.

Disclosed: a refusal counter resident in the metered party's own memory field, written by that party's own determinations; incrementing upon refusal of a conduct evaluation artifact without any party determining whether the refusal was well founded; determination-class handling in which acceptance resets the run accumulator, a not-applicable determination neither increments nor resets, and a not-determinable determination increments only on a recorded pattern for one asserted conduct event; a per-class increment register bounding contribution per origin-equivalence class per window; agent-computed asserted conduct event identity; conjunctive rate and run thresholds retrieved from a signed policy object; a gate write, upon satisfaction of both thresholds, withholding execution for an enumerated set of action classes while leaving refusal intact; a self-clearing bar; and the successor-epoch window, decay, carryover, and continuous cost-multiplier variations.

Disclaimed: any assertion that this article establishes rights against any party, that any existing system infringes, or that any license is required. The filings referenced are pending applications. Nothing here claims rate limiting, quota enforcement, reputation scoring, staking, or arbitration; those are named only to mark the boundary of what the disclosed structure does differently, namely price refusal while leaving merit undetermined.