MB4a — Audit Independence

Audit-path legitimacy: the designated measured correction/audit path is uncaptured and its endorsements count. Precise bet: correction integrity implies the measured path is legitimate — a falsifier bridge, not green Verify ⇒ real-loop integrity.

What decision changes?

Before trusting a named audit path, ask whether the judge or monitor can be captured, whether green Verify on the occurrence implies control of the real intervening loop, and whether bypass routes exist outside the measured path.

The field often folds this into corrigibility, but this project types it separately from MB4 — Corrigibility. MB4 asks whether correction preserves the correction process under pressure. MB4a asks a narrower audit question: is the designated measured correction/audit path legitimate and uncaptured?

CIRIS names the same wall as WA independence: green Verify on a certified occurrence does not imply correction-channel integrity  on the real intervening loop when composites, tools, memory, or incentives sit outside the measured path. Trusted-monitor and judge-capture cousins in Redwood-style control research target the same subchannel.

In Lean, MB4a is a falsifier bridge, not a positive certification rule: if correction integrity holds, the measured path must be legitimate and uncaptured. Capture of any measured-path handle refutes integrity (contrapositive). A green path reading alone does not discharge the bridge — legitimacy requires control, reach, persistence, and anti-capture, supplied by the bridge rather than by path construction.

Where agendas agree: CIRIS WA independence; trusted-monitor / judge-capture work. Where they diverge: no field headline for MB4a alone; green Verify ⇏ real-loop integrity; do not merge with MB8 (legacy CEV route) or with MB4 dynamical corrigibility without naming which subchannel was tested.

What would count as evidence?

Evidence would include interventional probes for judge capture, composite/tool bypass outside the named path, and tests that distinguish path telemetry from real correction uptake.