Threshold normalization for SuccinctMCSP #
This module caps sampled-circuit thresholds at a semantics-preserving linear interpolation bound. A direct DNF uses one exact-input term per positive sample; an existing witness guarantees that no negative sample has the same input as a positive one. The resulting normalized raw witness relation is polynomially balanced in the canonical instance-code length, including empty sample lists at very large binary-encoded arities.
Every feasible instance remains feasible at the explicit sampled DNF upper bound. Contradictory instances are not claimed to become feasible.
The effective threshold never exceeds the threshold supplied in the code.
The effective threshold never exceeds the sampled interpolation bound.
Capping the threshold at the sampled interpolation bound preserves feasibility in both directions.
The sampled interpolation bound is no larger than the canonical instance code itself.
The explicit normalized raw-witness envelope is polynomial.
Every sampled yes-instance has a normalized raw witness whose code is polynomially bounded by the canonical instance-code length.
The normalized raw witness relation is polynomially balanced.
SuccinctMCSP membership is exactly existence of a normalized, polynomially bounded raw witness.