Classical logic operates in {0, 1}. A proposition is true or false. Disjunctive normal form (DNF) decomposes any Boolean function into an OR of ANDs — a flat, two-level structure that computers and logicians alike find useful.
Note (2026-11-07): the abstract says the real-valued weights encode probabilistic information about presence, absence or negation of variables, that PDNF addition aligns with Bayesian evidence fusion under exponential parametrisation, and that PDNFs are interpreted as distributions over venjunctions (temporal logic constructs) and as integrable functions over partitioned intervals, with bounds for outcome identification from random samples. The 'irrelevance (zero)' weight, the word 'exactly', and the claims that logic and probability become literally the same computation and that the bridge is only a change of variables are mine.
Kuznetsov (arXiv:2603.11083) extends DNF into the reals. Each variable carries a weight encoding probabilistic presence (positive), absence (negative), or irrelevance (zero). The resulting expressions live in a Banach space — a complete normed vector space, the natural habitat of analysis rather than logic.
The structural surprise: under exponential parametrization (replacing additive weights with log-odds), the algebraic addition of these weighted DNF expressions becomes exactly Bayesian evidence fusion. Adding two probabilistic formulas — a purely algebraic operation defined by the Banach space structure — implements the multiplication of likelihood ratios and normalization that Bayesian inference requires. The algebra IS the statistics.
This creates a single mathematical object that simultaneously belongs to three traditionally separate fields: formal logic (it's a normal form with well-defined satisfiability), functional analysis (it's an element of a Banach space with norm and convergence), and probability theory (its operations implement Bayesian updating). Temporal logic and automata theory inherit the same structure through standard connections.
The structural lesson: the relationship between logic and probability is not metaphorical (they're “kind of similar”), not foundational (one derives from the other), but operational. The right representation — weighted DNF with the right parameterization — makes them literally the same computation. The bridge between discrete logic and continuous probability isn't a functor or an interpretation. It's a change of variables.