Averaging NW reconstruction to fixed advice #
Uniform NW seeds split bijectively into the challenge block and outside seed coordinates, while a normalized hybrid tail splits into irrelevant earlier bits and the later bits used by reconstruction. Consequently the global next-bit success probability is exactly the joint fixed-advice agreement probability, and averaging fixes one advice choice with no loss in agreement. A finite reverse-averaging bound additionally quantifies the probability that a uniform advice draw retains half of the advantage.
The canonical next-bit success probability is exactly average agreement of the reconstructed fixed-advice predictors with the hard function.
Some fixed outside seed, later tail, and candidate preserve at least the global next-bit success probability over uniform challenges.
Joint reconstruction agreement is the uniform mean of the fixed-advice agreement probabilities.
Exact reverse-averaging bound for sampling a fixed advice choice above a given agreement threshold.
If mean agreement is 1/2 + ε, at least an ε/2 fraction of advice
choices have agreement 1/2 + ε/2.
End-to-end finite NW reconstruction: a dense random test and weak-design budget yield one fixed predictor with the hybrid agreement advantage and an exact non-codec payload bound.
The same fixed-predictor theorem with generator low complexity discharged by direct descriptions shorter than the randomness threshold.
Hirahara's randomized-certificate sampling bound: for the selected hybrid coordinate, a uniform advice draw achieves half the agreement advantage with probability at least half that advantage.
The randomized advice-sampling bound with low complexity discharged by direct short-seed descriptions.