The gap

A party that refuses can refuse without end, and the usual lever against indiscriminate refusal is to ask whether each refusal was justified. That question routes through an adjudicator — a reviewer, an arbiter, a body that rules on the merits — and the consequence arrives only after the ruling, from outside the refusing party. Prior approaches assume that shape: the meter lives with the adjudicator rather than the metered party, and nothing moves until merit has been decided.

That coupling carries a cost. Merit adjudication is slow, contestable, and itself a fresh surface for dispute. A refusing party whose record is stale or wrong looks, from the outside, exactly like one whose refusals are sound — until someone does the work of telling them apart. Where agents evaluate one another's conduct at machine speed, the adjudicator becomes the bottleneck, and the refusing party keeps every incentive to refuse.

The invention

Merit-independent metering places the counter inside the refusing party. A refusal counter resident in the agent's own memory field increments when the agent refuses a conduct evaluation artifact — and it increments without the agent, its principal, the asserting party, or any further party deciding whether that refusal was well founded. The counter carries two accumulators: a rate accumulator over a policy-declared window, and a run accumulator counting consecutive refusals uninterrupted by an accepted determination. A per-class register admits at most one increment per origin-equivalence class within a window, so no single asserter can manufacture the count alone.

When both the rate threshold and the run threshold are satisfied, the counter writes the agent's own authorization gate to a withheld state, transitions the agent into a non-executing cognitive mode, and emits an escalation record to the principal. Two properties follow from that construction. The metering is complete on append — it takes effect with no external ruling. And an agent whose record is stale, partial, or wrong is metered identically to one whose refusals are sound; the structure never asks which it is, because none of its fields hold a merits outcome.

The inventive step

The departure is the cross-faculty split. The faculty metered and the faculty spent are distinct: the write withholds execution — the capacity to select and perform actions — and withholds nothing else. The faculty of refusal is left untouched, so in the withheld state the agent goes on receiving conduct evaluation artifacts, retrieving its lineage, producing determinations, and appending them. An agent that refuses indiscriminately therefore spends down its own authority to act while keeping, in full, the faculty by which it refuses; the structure does not silence, rate-limit, or disable the refusing party.

Reinforcing that split is a self-clearing bar. Determinations produced while withheld do not decrement the counters and do not reopen the gate; the withheld state persists irrespective of elapsed time; and the decay that later relieves the rate accumulator is a function of successor epochs of the agent's own hash chain, not of any determination the agent can produce. Because the consequence is denominated in a faculty other than the one exercised, and because clearing is driven by an interval the agent does not advance, the agent cannot refuse its way back to authority.

Alone, and in composition

On its own, merit-independent metering is a self-contained governor a semantic agent runs against itself. It leans on no tribunal, no counterparty record to check against, and no shared clock — the window can advance on the agent's own hash chain, beyond the reach of a wall clock an execution node might manipulate. The counter's fields are recoverable by replay of the append-only lineage, so the meter is auditable without ever being external.

In composition, it is the metering primitive the surrounding layers lean on. The same authorization gate it writes is also written by deviation-driven governance elsewhere in the architecture, and the two writes are reconciled as a union with independent return procedures per action class. The origin-equivalence classes it consults, the signed policy object it reads thresholds from, and the escalation record it emits to the principal are shared with those layers — so the refusal counter stands as one writer among several onto a common gate, not a bolt-on.