Use when stating the theory, formalizing the model, and deriving predictions for a Cognitive Psychology (Elsevier) manuscript. The journal rewards a formal/computational account whose parameters mean something and whose predictions discriminate it from rivals. Structures the theory and the model that the experiments test; it does not fit the model or run analyses.
SkillScore79Maintained
Automatic
◆Research pack
Last verified 9/1/2026
What it changes
Passed 18 of 23 test cases with the skill, and 4 of 23 without it.
Without skill17%
With skill79%
Checks it passed that the model alone missed
Uses the exact header block 【Rival(s)】.
Uses the exact header 【Status】 to mark prediction status.
Assigns cogpsych-literature-positioning to the 【Next】 block.
and 24 more in the full report
18 fixed1 passed either way2 still failing3 broke
Pass-rate lift
+63 pp
Tokens
+11%
Agent turns
0%
Response time
-19%
Model gemini-3.6-flashMeasured 9/1/2026v119 of 24 cases passed with skill · 23 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 18 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 18
Prompt
I'm designing an experiment on working memory forgetting to adjudicate between temporal decay and resource interference. I need to establish a rigorous computational hypothesis section before gathering data. Please outline the theoretical setup using bracketed headers for 【Theory】, 【Model】, 【Rival(s)】, 【Discriminating prediction】, 【Status】, 【Disconfirming evidence】, and 【Next】. Make sure to request psychological meanings for all free parameters and clearly define the qualitative signature that distinguishes the accounts.
What was checked, without and with the skill
Checked
Without skill
With skill
Explicitly designates pre-committed predictions as confirmatory under 【Status】.
Missed
Passed
Specifies cogpsych-literature-positioning under the 【Next】 section.
Missed
Passed
case-01 · gemini-3.6-flash · 9/1/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
Cognitive Psychology rewards a formal account of a cognitive process — a computational or
mathematical model whose parameters have interpretable meaning and whose predictions can be fit to
data and compared against rival models. The cardinal move here is to turn a verbal theory into a
model that makes the experiments discriminating, and to separate predicted (confirmatory) from
discovered (exploratory) results.
When to trigger
Specifying the theory and the formal/computational model that the experiments will test
Deriving the predictions that separate your account from rival models
Co-designing the model with the experiments (iterate with cogpsych-study-design)
A reviewer said the work is "atheoretical," "the model is just a curve fit," or "your data don't
distinguish the accounts"
Build the theory-and-model
State the cognitive theory. What mechanism or representation explains the phenomenon, and why —
in words, before equations. Name the rival accounts you intend to adjudicate.
Formalize it. Write the model: its representations, processes, free parameters, and what each
parameter means psychologically. A model whose parameters lack interpretation is a red flag here.
Name the rival model(s). Specify the competing account(s) in the same formal language so the
comparison is fair (nested or matched-flexibility where possible).
Derive discriminating predictions. Identify the data pattern that the models predict
differently — that qualitative or quantitative signature is what your experiments must produce.
Mark prediction status. Separate confirmatory (pre-committed/preregistered) predictions from
exploratory model exploration done after seeing data; do not present a post hoc fit as predicted.
State what would disconfirm the model. Which data pattern, or which parameter estimate, would
count against your account — this is what makes the model a theory, not a fitting exercise.
Avoiding the "just a curve fit" objection
A model that fits anything explains nothing. Show the model is falsifiable (some data it cannot
produce) and identifiable (its parameters can be recovered — handoff to cogpsych-data-analysis).
Prefer qualitative signatures that one model predicts and the other forbids over a small numerical
edge in fit; reviewers trust a crossed prediction more than a smaller AIC.
Worked micro-example — theory to discriminating prediction (illustrative)