Full nonlinear overlap map
If attenuation is diffuse, a complete monomial-by-monomial overlap map should show broadly distributed small differences rather than one dominant reproducible cancellation term.
Diagnostic work, exact algebra and local capabilities are not treated as end-to-end mining advantage.
What was tested?
If attenuation is diffuse, a complete monomial-by-monomial overlap map should show broadly distributed small differences rather than one dominant reproducible cancellation term.
Why the test is meaningful
A complete map avoids selecting a visually favorable monomial after observing the data. For the four-source-bit ANF, all eleven nonlinear monomials of degrees two through four are enumerated and their presence rates compared between fixed groups.
M={S⊆{0,1,2,3}: |S|≥2}, |M|=11Δ_S = P(S present | survivor) − P(S present | non-survivor)replication(S) ⇔ sign(Δ_S^original)=sign(Δ_S^fresh)How it was tested
Enumerate the full eleven-monomial support for every exact output mask, compute survivor/non-survivor rate differences without feature selection and require sign replication in the independent 300-record cohort.
What happened
Semantic errors were zero. Nine of eleven monomials retained the same difference sign across original and fresh cohorts; none formed a dominant complete selector. Fresh differences ranged from about +0.030 to −0.086, supporting a diffuse overlap-loss profile.
Exactness and statistical controls
The family was defined before comparing groups, so all nonlinear monomials are reported. Replication used the untouched cohort and explicitly records the two sign disagreements. Exact mask reconstruction remained the semantic gate.
What the result means
The map supports distributed attenuation rather than a single exploitable algebraic term. Nine sign agreements are an audit pattern, not proof of predictive or economic value.
Limitations
- Sign agreement alone does not quantify uncertainty or multiple-testing evidence.
- The groups are observed outcomes, not an online classifier.
- No full-round work reduction was measured.
Evidence trail
Enumerate all eleven nonlinear monomials for each exact mask, publish every rate difference, preserve cohort membership and repeat the predefined sign comparison.
Canonical variants
SUBENGINE-V32ASUBENGINE-V32BSUBENGINE-V32CSource: internally audited canonical reports. Local filesystem structure, private headers and operational identifiers are excluded from publication.