Publications
Peer-reviewed work, preprints and the registrations, data and code behind them.
Peer-reviewed
Journal article
Published Vol. 84, 2027
Moral Violations Are Made, Not Found: A Theory of Motivated Violation Construction
Peer-reviewed workshop poster (non-archival)
Preprints
Theory
Preprint Starting point of the programme
Trait Judgmentalism: Contextual Moderation Failure and the Architecture of Evaluative Rigidity
Abstract
Psychological paradigms routinely treat comprehension of mitigating context as sufficient evidence that it entered the evaluation. Whether it received any weight is not separately tested. This paper names that default the Revision Assumption and names its proposed exception Trait Judgmentalism: a stable individual difference in which an evaluator comprehends relevant, credible mitigation and does not revise the evaluation of the person. The defining mechanism, Contextual Moderation Failure, is a failure of post-comprehension weighting at an evaluative gate, specified as a computational-level model that must earn its interpretation. The paper states the conditions any valid test must satisfy, separates the construct from its neighbours, and distinguishes overt from unexpressed judgmentalism. A preregistered reanalysis of public blame-updating data serves as Stage 0. Its registered hypothesis was supported. That support turns out not to help. The registered criterion sits inside the dispersion of the individual change it classifies, fires on 81 per cent of a no-information retest control, and the data contain no matched baseline to correct the count. Stage 0 therefore supplies no evidence for or against existence. What it supplies is the design requirement carried into a staged validation ladder with explicit falsification conditions: a matched within-person baseline.
Preprint
Moralised Essentialism: How Moral Violations Become Identity
Abstract
That moral character is inferred as a kind of hidden essence is now a working assumption across several literatures. What remains unspecified is the representational architecture this produces. We propose Moralised Essentialism (ME): a framework in which a moral violation, once encoded as identity-disclosing through an operation we call Moralised Essentialist Encoding (MEE), is sealed off from subsequent contextual evidence. The framework’s contribution is a two-level architecture. After essentialist encoding, the character verdict and the contextual explanation occupy structurally independent levels, and context is routed to a representation with no revising path back to the essence inference. The claim is that inference at the essence level is closed, and a specified task, the Context Integration Task, tests it under a pre-registered equivalence rule. The architecture predicts an enrichment paradox: additional context elaborates the situational narrative while leaving the character inference unchanged. Its signature is a correlation close to zero between contextual understanding and character judgment, held flat as exculpatory evidence accumulates. Trajectory, not magnitude, is the discriminating variable, because competing accounts can produce a small correlation at one measurement point but not a flat function across escalating exoneration. We state seven propositions, give explicit failure conditions for the two that carry the architectural claim, and identify four recent findings that point against the central prediction. The framework is offered as a contested architectural proposal with a pre-committed decision procedure, not as a settled account.
Concept notes
- Evaluation Without Revision (April 2026)
- Moral Repair Is Not Reversal (May 2026)
Language models
Revised preprint
Same Prompt, Not the Same Instrument: A Preregistered Test of Persona Framing and Moral Concern in Four Large Language Models
Abstract
Cross-model comparisons of moral judgement assume a shared prompt is a shared instrument. We report a preregistered test of Motivated Violation Construction (MVC), on which a loaded actor identity can move ambiguous behaviour into violation, in Claude Opus 4.7, GPT-5.5, Gemini 3.1 Pro, and DeepSeek-V4-Flash, which classified 20 ambiguous-by-design scenarios under a neutral frame and six loaded personas across 11,200 trials. Two models classified almost every scenario as concerning under the neutral frame, leaving little headroom for the predicted increase. One failed the registered test-retest gate, which by post hoc analysis falsely excludes a stable model 83.9% of the time at that baseline, while a non-registered diagnostic finds a retained-model shift the gate cannot see. Three of the four registered MVC predictions were not supported in the retained three-model panel, the fourth indeterminate under an amendment-fixed rule. The P1 gateway failed, so the remaining battery is exploratory by rule, and its three-test adjudication returned a mixed pattern with no adjudicated winner: two tests favoured valence priming under registered failure semantics, one met its MVC-consistent threshold, concentrated in one model. Persona effects reversed sign across models under one prompt, an inverted question format collapsed expressed concern, and one model expressed concern on mundane controls at rates others never approached. Baseline, headroom, retest behaviour, and concern on those controls differed by model at certified standard, and position sensitivity differed in an exploratory recomputation. We call this configuration, post hoc, a model’s response regime. A forced binary moral-concern prompt did not function equivalently across these four models.
Preprint Under review
The Cell, Not the Call: A Preregisterable Audit Protocol for Repeated‑Call Behavioral Experiments With Large Language Models
Abstract
When a hosted large language model answers the same prompt many times, the resulting completions can estimate that model’s response distribution for that one prompt. They cannot, on their own, stand in for a sample of scenarios, participants, or models, and treating them as if they could undermines the claims that psychological researchers most want to make with these systems. This article presents an auditable protocol for closing that gap: decisions, records, and prespecified consequences that a researcher adopts before, during, and after data collection, organized around five modules covering the inferential unit, calibration, response handling, endpoint stability, and provenance. We illustrate the protocol’s use, rather than test its overall effectiveness, with a four-model moral-judgment experiment (11,200 main trials) in which it surfaced substantial calibration ceiling effects, provider-specific parsing problems, and a registered stability gate that excluded a model with a maximum observed change of 0.10 while retaining models that changed by 0.55 and 0.45. A supporting simulation, reported in the supplementary material, shows why aggregating repeated calls into cell-level counts does not, by itself, correct for unmodeled dependence: aggregated and call-level analyses are mathematically equivalent, so both inherit the same inflated false-positive rate once cell heterogeneity is ignored. The protocol’s contribution is practical rather than statistical. It gives psychological researchers a way to convert dependence-and-validity concerns into decisions, thresholds, and inspectable records that a reviewer or replicator can check, with explicit limits on what one worked case and one focused simulation establish.
Registrations, data and code
- Same Prompt preregistration (OSF 6m3sk)
- The Cell companion (OSF fyjsx)
- The Screen data and code (Zenodo 22945740)
- Blame-updating reanalysis, reported in Trait Judgmentalism (OSF z9jkd)