Why not use confidence scores?

Model confidence is calibrated against its training distribution, not against the operator's situation. Behavior in scope is the only evidence that matches the situation the trust boundary has to govern.

Does this mean every action has to be reviewed?

No. It means the trust boundary is updated by every recorded outcome. Where behavior has accumulated, the harness can act without asking. Where it has not, it asks.

← Prompted LLC

Canonical · L4 Primitive·Shipped

Trust as behavior

Trust is not a claim or a confidence score. It is the pattern of observed behavior — actions, refusals, escalations, outcomes — that has accumulated against a scope. Ubiquity treats trust as behavior so autonomy can be earned rather than declared.

Plain definition

Trust as behavior is the operational premise that an AI harness has only the trust its history of action supports. Approvals, refusals, rollbacks, escalations, and successful outcomes are all signals. The trust boundary moves with evidence, not with confidence.

Why it matters

Confidence is a property of the model. Trust is a property of the system around the model. Treating trust as behavior keeps the trust boundary tied to what the harness has actually done in scope, in context, against real outcomes — not to how persuasive its last response sounded.

What it is not

A static permission level granted at deployment.
A confidence score reported by the model.
A reputation badge attached to a vendor or framework.
A binary trusted / untrusted flag.

Example

A harness has earned action scope on dependency upgrades after 200 clean outcomes. The same harness still asks on schema changes, because its behavior history in that scope is shallow. Trust is local, scoped, and behavioral.

Where it appears in Ubiquity

Trust as behavior is the substrate-level expression of earned autonomy and the input to trust telemetry.

Ladder context

Demand ladder

L1 Pain→L2 Contrast→L3 Category→L4 Primitive→L5 Branded

← Up the funnel

What is trust telemetry?

Down the funnel →

This is a leaf of the ladder.

Adjacent transitions

Related terms

Frequently asked

Why not use confidence scores?: Model confidence is calibrated against its training distribution, not against the operator's situation. Behavior in scope is the only evidence that matches the situation the trust boundary has to govern.
Does this mean every action has to be reviewed?: No. It means the trust boundary is updated by every recorded outcome. Where behavior has accumulated, the harness can act without asking. Where it has not, it asks.