The gap

A governance mechanism in which every departure from a declared constraint is a fault carries no quantity representing the need or urgency of the action attempted. There is nothing to weigh against the constraint — no magnitude that says the agent is exhausting a resource it needs to keep operating, that a relational commitment it depends on has been withdrawn, that its declared roles have fallen into contradiction. Prior approaches assume departures are violations by construction, and so they hold no input from which a permission condition could ever be computed. The only available responses are compliance and fault.

That architecture treats a sanctioned override and an unexplained breach identically, because it has no vocabulary for the difference. An agent that deviates under genuine pressure and an agent that simply misbehaves land in the same withheld state, incrementing the same refusal counter, leaving a party who later reads the record unable to reconstruct which happened or by what quantities. The conduct is legible only as pass or fail — and the space where a departure might be earned, metered, and accounted for goes unrepresented.

The invention

This chapter discloses the quantities from which deviation is computed and the permission semantics attaching to them. A scoped integrity vector tracks the agent's alignment across three scopes — with itself, with identified counterparties, and with parties beyond direct interaction. A need quantity accumulates a quantified semantic urgency per scope, computed rather than declared, compounding across intervals for as long as its condition obtains. A dynamic ethical threshold resolves per scope the minimum condition for deviation, rising as integrity and self-esteem rise and falling as they fall. An empathy weighting aggregates the anticipated impact of a proposed mutation across affected entities.

The deviation likelihood is the quotient of the difference between need and threshold, divided by the product of the empathy weighting and the self-esteem aggregate. Where it exceeds unity, the crossing is a permission and not a fault: the architecture admits a mutation the signed policy object otherwise forbids, and records it as a permitted deviation. The policy object is not nullified and not amended — it remains authoritative, and the conduct deviates from it in a computably sanctioned manner through a recorded override. The permitted deviation record carries the need-threshold imbalance at admission, a context hash, an integrity displacement vector across the three components, the constraint overridden, and a restoration status.

The inventive step

The departure from prior art is where a departure is allowed to live. In approaches that treat every deviation as a violation, the only place to put a policy-forbidden mutation is a fault log; here the admission is a first-class semantic mutation of the same standing as any other, entered through a per-mutation deviation-preparation state scoped to that mutation alone, and incapable of extending its permission to any further mutation. A crossing that fails the mutation's own policy constraints, or that arises from accumulated state with no mutation pending, writes the authorization gate to its withheld state instead — so the same quotient distinguishes a sanctioned override from a genuine breach.

The permission is metered, not free. A deviation deductible draws the harm of each permitted deviation first against the agent itself, borne as a self-esteem decrement for which no reparation is available, without any determination of whether the deviation was well founded. Harm exceeding the deductible accumulates in a retention register as an arc dischargeable only by a forward restorative mutation; and once the undischarged amount exceeds the declared aggregate retention, the permission condition is foreclosed — a deviation likelihood above unity thereafter withholds rather than admits, until pending arcs are discharged. Deviation is budgeted, accounted, and repayable, never erasable.

Alone, and in composition

On its own, the mechanism is a quotient over four cognitive quantities — need, threshold, empathy, and self-esteem — that converts the pressure to deviate into a single per-scope number and attaches an auditable permission record wherever that number crosses unity. It gives a governed agent a way to depart from its own declared values under recorded conditions without pretending the values changed, and to leave behind a lineage a reader can reconstruct rather than an unexplained fault.

In composition, this quotient is the deviation quantity the wider architecture's deviation engine recomputes after each modification of the integrity vector and self-esteem aggregate. Its permitted deviations open reparation arcs that feed the integrity compliance score governing later mutations; its counterparty-directed arcs hand off to the reparation layer, dischargeable only by that counterparty's own recorded acceptance; and its self-sourced gate writes are restorable without any external principal. The permission semantics and the accounting that bounds them are the substrate on which every downstream reconciliation, suppression record, and restorative mutation operates.