Log-Arithmetic-Tropical Decomposition: how each byte's prediction flows through three semiring regimes.
L = log-stochastic counting (support values), A = arithmetic probability (normalization), T = tropical max-min (LPP selection).
Support values w ∈ [0,255]. Each w represents ~2w observations.
Counting via probabilistic increment: P(w→w+1) = 2-w.
Select a position to see L-regime data.
Normalized probability from support values.
P(b|source) = 2w(b) / Σ 2w(b').
Surprise = -log2 P(actual).
Select a position to see A-regime data.
Max-min selection across LPPs. The winning LPP has the highest gap
(s1 - s2). Its prediction becomes the output.
Select a position to see T-regime data.