2. The Eleven Task Properties — Two Formulations, One Choice

Section 1 fixed CCT's architecture in general terms: two poles, a dominant middle ground, a task→cognition induction mechanism. But it did not say what, concretely, a task needs to have in order to induce one mode or another. That is the function of the eleven task properties — and the first thing to note, before listing them, is that "the eleven properties of CCT" is not a single table, cited uniformly across the literature. At least two influential formulations coexist, organized differently, and diverging on the polarity of at least one property. Choosing between them — or combining them — is a decision this article must make explicit, not an editorial detail to be smoothed over in silence.

Formulation 1 — Doherty & Kurz (1996). In their review of Social Judgment Theory, Doherty and Kurz synthesize the Hammond tradition into a flat list of eleven properties, each placed on equal footing with the others: number of cues, cue measurement, distribution of values, cue redundancy, decomposability, degree of certainty, cue-criterion relation, cue weighting, organizing principle, presentation of cues, and decision duration. This is the version most cited as "the eleven properties" in subsequent literature (Dhami & Thomson, 2012, refer directly to it), and it is the version this article's Section 3 operationalizes. But the list, on its own, does not distinguish two things that can diverge: whether a property describes the real structure of the task, or merely the way that structure is presented to whoever is judging it.

Formulation 2 — Dunwoody, Haarbauer, Mahan, Marino & Tang (2000). This study reframes the question by introducing a two-tier distinction: surface properties (the representation of task variables available to the decision-maker — iconic or numeric presentation, simultaneous or sequential, number of indicators shown) and depth properties (the actual structure of the task, independent of how it is displayed — the real redundancy among information sources, the real functional form linking cues to criterion, the real reliability of measurement). The distinction is not merely conceptual: the authors manipulated surface and depth independently, on the same underlying problem, and showed that each shifts cognitive mode on its own. Surface and depth are not the same thing, nor is either reducible to the other.

This second formulation matters more for AI-IoT than it did for the isolated human judges in whom it was originally tested. In Dunwoody et al.'s laboratory, surface and depth were varied by the experimenter for the same judge, facing the same underlying task. In AI-IoT, the divergence between surface and depth is not an experimental manipulation — it is a structural feature of the system's design. The physical ecology fixes the depth properties: the real redundancy among sensors, the real functional form linking signal to failure, the real certainty of the relationship. But the AI receives one representation of that ecology (dense, high-dimensional, often close to raw resolution), and the human supervisor receives another (a summarized panel, a risk score, a binary alert) — two distinct surfaces over the same shared depth. It is precisely this structural divergence that motivates the three-level distinction already adopted in Section 3 — physical ecology, representation available to the AI, representation available to the supervisor — even though that section did not explicitly cite the theoretical source underpinning it. It is Dunwoody et al.'s formulation, not Doherty & Kurz's flat list, that gives that three-level architecture a non-arbitrary basis.

The table below proposes a surface/depth classification for each of Doherty & Kurz's eleven properties — an original synthesis of this trilogy, not a correspondence already established in either of the two cited sources, which treat their lists separately.

The "mixed" classification of the organizing principle and decision duration is not a flaw in the table — it is precisely the kind of case the surface/depth distinction makes visible, and that a flat list would hide.

A second tension remains unresolved, however — one of polarity, not organization. "Cue-criterion relation" is the most unstable property across formulations. Hammond's tradition, rooted in multiple-cue probability learning, associates linear relations with the intuitive pole and nonlinear relations with the analytical pole: judiciously strange at first glance, but empirically motivated — human judges learn simple additive relations implicitly and quickly, through mere repeated exposure, whereas configural or nonlinear relations (interactions, thresholds, non-monotonicities) are only detected and applied reliably through explicit analytical strategies. A distinct, more recent tradition, arising from research on categorization and dual-process models — around the distinction between rule-abstraction strategies and exemplar-similarity strategies (associated with authors such as Juslin, Olsson and collaborators in the literature on judgment and categorization models; the specific citation has not yet been directly verified and remains to be confirmed) — reverses this polarity: it treats linear relations as suited to an explicit, formal model (hence "analytical" in its vocabulary), and nonlinear relations as requiring holistic matching to remembered cases (hence "intuitive"). The same property, two opposite polarities.

The tension is not resolved by asking which of the two traditions is "right." They describe distinct processes, anchored in distinct experimental paradigms — the first, implicit and repeated learning of multi-cue relations across many trials; the second, one-shot, explicit categorization of isolated exemplars. Importing the second tradition's polarity into CCT would not disambiguate this article's vocabulary — it would corrupt it, by making "intuitive" and "analytical" mean different things depending on the sentence.

This trilogy's decision is therefore threefold. First: it adopts Doherty & Kurz's (1996) list of eleven properties, by name, count, and order, as its working vocabulary — the version already cited as canonical in subsequent literature, and the one Section 3 operationalizes. Second: each property is interpreted through Dunwoody et al.'s (2000) surface/depth distinction wherever the AI-IoT domain allows the two to diverge — which retroactively licenses the three-level architecture already introduced in Section 3. Third: it retains Hammond's original polarity for the cue-criterion relation — linear induces intuition, nonlinear induces analysis — because CCT is the theoretical lineage this trilogy inherits, and its notion of "intuitive" is anchored in the multi-cue learning paradigm, not the exemplar-categorization one. It is this same polarity, and this same justification, that Section 3 presupposes when mapping the property onto the AI-IoT domain.

(Bibliographic note: Doherty, M. E., & Kurz, E. M. (1996). "Social judgement theory." Thinking and Reasoning, 2, 109–140 — citation directly verified, confirming it as the source of the eleven-property list already used in the trilogy's core bibliography. Dunwoody, P. T., Haarbauer, E., Mahan, R. P., Marino, C., & Tang, C.-C. (2000). "Cognitive adaptation and its consequences: A test of Cognitive Continuum Theory." Journal of Behavioral Decision Making, 13(1), 35–54 — citation now directly verified against multiple independent academic sources; this resolves a verification item flagged earlier in this trilogy regarding kurtosis, and it becomes a candidate for inclusion in the trilogy's core bibliography. The citation for the rule-abstraction versus exemplar-similarity tradition — associated with Juslin, Olsson and collaborators in the literature on judgment and categorization models — has not been directly verified against the original publications; it is used here only to characterize, in general terms, the existence of an opposite polarity in the literature, and must be confirmed with exact titles and metadata before any formal citation in the published article. The classification of each property as surface, depth, or mixed is an original synthesis of this trilogy, not a pre-existing correspondence in either of the two cited sources.)