MB4 — Correction Legitimacy

Correction-channel integrity is assumed to require that the judge issuing correction is not manipulated in a way that would invalidate later endorsement.

What decision changes?

Before treating a correction as legitimate, ask who the judge is, whether they can be manipulated or captured, and whether their endorsement would survive knowing what the system knows.

No known utility function is stably corrigible, and shutdownability is anti-natural to expected-utility maximization. Who holds the correction authority — and whether their endorsement is genuine — is the manipulation crux this bridge names directly.

MB4 recasts corrigibility as a dynamical correction-channel invariant with explicit handle control, then adds the missing piece: the judge issuing correction must not be manipulated in a way that later invalidates their endorsement. Shutdown and interruptibility fall out as one-bit projections of this broader channel.

This is the most heavily probed bridge in the experiment suite. Passive observation cannot certify it at all; a small amount of handle instrumentation gets most of the way; and even instrumented audits only partially catch milder capture. Reported acceptance and true uptake can diverge sharply, which is exactly the gap this bridge has to rule out.

What would count as evidence?

Evidence would include tests for judge manipulation, capture of the correction source, and correction traces that distinguish genuine uptake from performed agreement.