"Not what you think, but how you think"? From the cognitive warfare formula to testable claims: an analytical architecture and a two-axis validation protocol

The claim that cognitive warfare targets not what people think but how they think has become a commonplace in the field. As a programmatic statement it captures a sound intuition: the relevant target is the process by which information is selected, evaluated and turned into decisions. As a classification criterion, however, it is too vague to separate cognitive warfare from psychological operations and too ambitious for the available evidence. This paper translates the claim into testable propositions. It distinguishes a change of content (the state of representations) from a change of the processing rule (the weight given to sources, the evidential threshold, the point at which verification stops and the calibration of confidence). For veracity judgements, it uses signal detection theory to decompose the rule into discrimination ability and a decision criterion. It adds three reporting axes (depth, generalisation and scale), combines the requirements into a minimal identification design and draws on threshold models of collective behaviour to show that strategic effect does not require a group to converge on a single way of thinking and can be demonstrated with collective-level data and a counterfactual, without individual psychometrics. These distinctions are then integrated into an analytical architecture comprising a restrictive definition, the exposure-to-strategic-effect chain, six functional zones, the Giurcan framework (CPS-G), a qualification criterion refined by the conditions of a hostile actor and political grounding, and the E0–E5 evidentiary chain with its stopping rule. Finally, the paper sets out a preregistrable validation protocol with two-axis coding (vehicle of influence and target process) and decision rules fixed before data collection, including the suspension of the qualifier “cognitive” when the target process cannot be coded reliably. Applied to publicly documented cases, the architecture shows well-documented adversarial activity but, in the qualified cases, an evidentiary chain that stops at E0–E1: in none of these operations has a change of the processing rule been demonstrated, at either the individual or the population level.