Nisan--Wigderson fixed-advice reconstruction #
This module turns the overlap tables into the exact query made by the Yao
next-bit predictor. It also removes irrelevant seed and tail coordinates from
the fixed advice and proves the resulting Boolean payload bound
overlapCostAt + (d - ell) + 1, before codec and parameter overhead. The
averaging submodule fixes one advice choice without losing success probability,
repeated sampling finds such advice with an explicit geometric guarantee, and
global sampling chooses the hybrid coordinate and advice in one fixed-width
trial matching the randomized reconstruction algorithm. Certificate search
then checks both test orientations by exact finite agreement and returns the
first successful sampled trial. Finally, explicit reconstruction programs
materialize the predecessor tables so that evaluating a selected predictor no
longer accesses the hard function, serialize the complete program and decoder
index exactly, compose its agreement guarantee with finite list decoding, and
turn the resulting short string into a machine-relative bounded-complexity
certificate whenever the fixed decoder has a TM realization, then transfer it
to any efficiently universal machine through explicit compiler and clock
overheads.
There are exactly d - ell seed coordinates outside one design block.
There are exactly m - (i + 1) output coordinates after coordinate i.
Exact number of outside-seed assignments.
Exact number of later-tail assignments.
A reconstructed seed reads the supplied challenge on the current block.
Restricting a seed to the outside and current-block coordinates and then reconstructing it recovers the original seed exactly.
The target bit on a reconstruction background is the hard function applied to the varying challenge.
The table-based reconstruction query is exactly the hybrid query built from the corresponding candidate background.
Evaluating the test on the reconstructed query agrees exactly with the canonical next-bit experiment.
Pointwise agreement of the fixed-advice reconstruction predictor with the hard function is exactly the success event in the canonical experiment.
Exact reconstruction payload before encoding the test, design parameters, coordinate, and polarity.
A weak-design budget bounds the fixed predictor's non-codec payload by the budget, the outside seed, and one candidate bit.