Concept study / Tests & research methodsConceptual illustration. Shapes, colors, and motion are not measured data.
A research program exploring many geometries, scales and memory measures can generate an enormous number of possible comparisons. Some will look striking by chance. Preregistration makes the chosen question and decision rule visible before the relevant data are used. It does not guarantee a good experiment, and exploratory work remains valuable. The purpose is to identify which conclusions were genuinely predicted and which were developed after inspection. For Hill’s framework, a concise, versioned protocol can make a small study more informative than a much larger collection of attractive plots whose selection history cannot be reconstructed.
Register the decisions that can change the answer
Include the target observable, primary hypothesis, baseline, meaningful effect size, sample unit, sampling plan, stopping rule, exclusions and analysis. Specify preprocessing, fitting windows and the number of primary comparisons. For a cross-scale study, name the included systems and eligibility criteria. For a memory test, define training, readout, delay and decoder evaluation. If the dataset already exists, disclose which parts have been inspected and what remains held out. A document written before analysis is useful only if its timing and relationship to the data are stated accurately.
Pilot without spending the confirmation set
Use pilot data or simulations to determine whether the protocol is feasible and adequately sensitive. Freeze the confirmatory analysis after those checks. If a necessary change arises during collection, timestamp the amendment and explain its effect on the original question. Do not conceal exploratory analyses; place them in a separate section with their rationale and a proposal for independent testing. The project’s protocol generator should export a complete reviewable record, but an exported local file is not automatically a public time-stamped registration with an independent service.
Publish outcomes against the plan
The final report should show each primary prediction alongside its result, including failed and inconclusive tests. List exclusions and deviations, and distinguish predeclared from newly selected analyses. A beautiful visual that emerged during exploration can inspire a new hypothesis; it does not become an advance prediction retroactively. For Alpha 1, a preregistration is premature until a measurable consequence is specified. The immediate contribution may therefore be a public model specification that others can critique before any observational analysis begins, with unresolved assumptions openly marked.
The mathematical lens
This is a registration record, not a statistic. It documents what was decided and which observations were available at that time.
Turn it into a question.
- Question
- Can a reviewer distinguish predictions from post-observation choices?
- Method
- Archive the protocol, data-access history and amendments, then report all primary outcomes.
- What would count against it?
- Missing timing, hidden deviations or selective outcomes prevent a clean confirmatory interpretation.
Exploration finds questions; preregistration makes their later tests interpretable.
Follow the original work.
These sources support the mechanisms and limits discussed here. They do not endorse Hill’s framework. Notes distinguish research proposals, reviews, and observations.
Continue the investigation
Edition 02 · 18 October 2026 · AI-assisted synthesis. See editorial method.