Additivity of the modal operator — the property that necessity distributes over disjunction — has long been understood to imply the deduction-detachment theorem. If your logic has additive necessity, it has deduction. The implication runs one way; the converse was an open structural question.
Note (2026-11-13; read against the paper's full text): the paper's main results are stated for the local deduction-detachment theorem, not for deduction in general. It proves continuum many 'strongly non-additive' minimal discriminator varieties of Boolean frames (a notion it defines: a variety not equivalent to any additive variety), equivalently continuum many strongly non-additive maximal congruential modal logics with the theorem, and that every normal modal logic can be transformed injectively into a strongly non-additive one preserving the local theorem. I did not find in the text a statement that the converse was an open question; the paper says its theme is that deduction properties 'persist far beyond the additive setting'. The paper's non-elementarity result (its Theorem 5.5, shown through the congruence extension property) says that neither the class nor its complement is elementary, meaning not axiomatisable by any set of first-order sentences, finite or infinite; the essay's 'not captured by any finite set of algebraic axioms' is weaker than that, and it leaves out that the complement is not elementary either. 'Deduction is not downstream of algebra' and the building analogy are the essay's: the paper's own framework is algebraic semantics, with its examples inside that setting.
The converse utterly fails.
There exist continuum-many strongly non-additive maximal congruential modal logics that still possess the deduction-detachment property. The algebraic structure that was thought to generate deduction can be completely destroyed while deduction survives. Furthermore, any normal modal logic can be injectively transformed into a strongly non-additive version while preserving deduction — the transformation is systematic, not a collection of isolated counterexamples.
The class of congruential modal logics with local deduction is not even first-order definable. The property cannot be captured by any finite set of algebraic axioms. Deduction is more structurally wild than the algebraic framework can express.
This matters because deduction is the most fundamental inference rule — it says that if assuming A lets you prove B, then you can conclude “A implies B” as a theorem. The assumption was that this basic inferential capacity depended on the modal operator having nice algebraic properties. The paper shows the dependency doesn't exist. Deduction is not downstream of algebra. It is a more primitive organizational principle that persists through transformations that destroy the algebraic structure traditionally invoked to explain it.
The analogy: finding that a building stands after removing what you believed was the load-bearing wall. The building's stability had a different source — one that the architectural theory didn't describe. Inference is more robust than the foundations we attributed to it. The tool survives the destruction of its supposed support.