Cognitive Bias Paradigms as Dimensional Probes: A Structural Typology of What the Heuristics-and-Biases Literature Was Actually Measuring

The heuristics-and-biases tradition has been read as a catalog of named biases. This paper reads it as a catalog of experimental probes. Each canonical paradigm was constructed by an experimenter to elicit a specific judgment, and the data each paradigm produces probes one or more dimensions of a four-dimensional judgment process the participant runs to produce the response. We specify the four dimensions as recognition (what crosses the participant's threshold of attention), contextualization (what frame the participant holds), resolution (what determination the participant produces), and action (what the participant reports or carries forward). They are ordered by dependency, each dimension is a necessary condition for the next. The four-dimensional judgment architecture admits three structurally distinct closed paths from recognition back to recognition (the full four-dimensional judgment cycle, the direct recognition-to-resolution path, and the resolution-and-action-return path; architectural derivation in Stewart, 2026e, 2026f, 2026g). The architecture admits five non-empty violation classes corresponding to dependency failures within these three paths (2-1 at the recognition-to-contextualization step within the full cycle, 3-2 at the contextualization-to-resolution step within the full cycle, 4-3 at the resolution-to-action step shared by all three paths, 3-1 at the direct recognition-to-resolution step, and 4-1 as the catastrophic failure of the full cycle), plus one structurally empty cell at the contextualization-to-action reversal (Class 4-2) where the architecture admits no direct dependency for the failure to occur on. The participant's judgment cycle runs continuously, with each cycle's commitment carried forward to the next cycle's recognition through cross-cycle content propagation. Class signatures index within-cycle failures on the propagated content. We read what the literature labels "bias" as the gap between the participant's accumulated default at the probed dimension and the experimenter's normative target. The typology classifies each probe, names what dimension it targets, and reads the participant's response as the operational output of the accumulated default content at that dimension. The canonical paradigms distribute across three categories. Clean single-class paradigms occupy four dependency-reversal classes. The 3-1 class catches probes where the resolution operates on a recognition admission that does not include the targeted referent (Linda, availability, representativeness). The 3-2 class catches probes where the resolution operates on a contextualization frame whose accumulated content is not the frame the situation called for (cab problem, anchoring, Asian disease, IKEA, endowment, hindsight, fundamental attribution error, base-rate neglect, confirmation bias, planning fallacy, in-group bias, status quo, false memory). The 4-3 class catches probes where action operates without resolution having delivered against the currently relevant content (belief perseverance, sunk-cost). The 4-1 class catches probes where the participant's action commits forward content the participant's recognition did not admit (confabulation). Devine's three-study sequence is a separability demonstration of the contextualization-resolution distinction rather than a single class. The bystander and bandwagon effects are multi-signature paradigms whose data isolates 3-1, 3-2, and 4-3 instances at the participant level under the experimenter's flattened label. We identify two unfilled cells. The 4-2 cell is structurally empty because no direct contextualization-to-action dependency exists in the architecture; any apparent 4-2 effect decomposes into a 3-2 frame failure with a 4-3 manifestation at the action stage. The 2-1 cell is joint-occluded by the cross-cycle frame delivery: the joint loads the new cycle's contextualization with the prior cycle's pragmatic output before the within-cycle recognition-to-contextualization step can fire in isolation. The two unfilled cells are unfilled for different architectural reasons, one because the dependency does not exist, one because the within-cycle isolation the class requires is occluded by the joint's normal operation. We propose three forward openings. The paper predicts that probes targeting the same dependency produce correlated responses across participants on the same content domain, and probes targeting different dependencies do not, regardless of the named-bias label. For heterogeneous multi-signature paradigms the prediction operates at the subtype level, each subtype correlating with paradigms targeting the same dependency rather than the paradigm as a whole correlating with other paradigms. A second opening is the structural consequence at the 4-2 dependency. A third opening is the construction problem at the 2-1 dependency, which requires a design that isolates the experimental cycle's recognition admission against the carried-forward frame before the joint integrates the new admission. The correlation prediction's operationalization is bound at five structural facts the architecture forces. Content matching at the accumulated frame the experimental cycle activates at contextualization. Dependency-content separation at the dimensions' invariance across participants. The 3-2 block's homogeneity at the convergence at resolution. The 4-1 class's testability on healthy samples at the graded admission-threshold default at recognition. Multi-class subtype resolution at the cycle's two-direction readability at resolution. The five binding forces remove the analyst's content-matching and subtype-resolution latitude and specify the prediction in terms that follow from the architecture rather than from the analyst's discretion. **Keywords:** cognitive bias, heuristics and biases, Linda problem, anchoring, framing effect, base-rate neglect, confirmation bias, planning fallacy, fundamental attribution error, bystander effect, representativeness heuristic, availability heuristic