Look-ahead extraction from a nearly uniform right state #
A refresh call supplies a right state close to uniform given the current transcript. This is sufficient for the next actual two-round look-ahead: no pointwise entropy bound for that approximately uniform state is needed. The original left source has an explicit average joint envelope, and the initial-seed map sends a uniform state to a uniform seed.
The proof repairs the distinguished state coordinate while preserving the original sources, applies the checked two-round theorem, and transports the result to the actual output law. Since the retained honest first output can change under repair, the stated bound charges the state error twice. The final law retains the entire original right state and both first outputs.
This supplies the repair between look-ahead passes in the approach of Chattopadhyay--Goyal--Li, Non-Malleable Extractors and Codes, with their Many Tampered Extensions, Lemma 6.8: https://arxiv.org/pdf/1505.00107. The finite coupling and explicit error bound are the deductions proved here.
A uniformity-preserving seed map contracts the original state's joint discrepancy.
A balanced image of an exactly uniform retained state is independently uniform.
A balanced seed map has zero joint discrepancy on an exactly uniform retained state.
A nearly uniform state supports the next two-round call, retaining the actual transcript.