Preregistering means recording an experiment’s question, criteria and planned analysis in advance. It limits stories reconstructed after observation.
01STORY BRIEFThe idea in three screens.3 SCREENS
Write possible verdicts before knowing the score.
Preregistering means recording an experiment’s question, criteria and planned analysis in advance.
A dated document is enough to begin: hypothesis, groups, duration, primary metric, exclusions and expected decision. You can modify the plan when reality demands it, provided the previous version and explanation remain. Exploration is still possible; label it as exploration.
Add an acceptable conclusion for each outcome: favorable signal, uncertain difference or measurement failure. This often exposes protocols that have no response to negative results. A useful experiment should remain informative when it contradicts the initial plan. Otherwise it mainly produces retrospective justification. Recording these possibilities also helps reviewers distinguish the intended decision from additional questions discovered while the experiment was already underway.
Write the decision while you still dislike all outcomes equally
A preregistration does not need to be a grand academic document to improve an operating experiment. Before publishing, write the question, units, treatment, primary outcome, observation window and stopping rule in a dated record. Add the exclusions you already know you will need. This prevents the team from promoting whichever metric looks best after the fact. It also exposes vague questions early. If nobody can agree what result would change the publishing process, the experiment may not yet be ready to consume accounts, creative work and budget.
A small contract with your future self
- 01Before
Freeze the question, outcome and stopping rule.
- 02During
Log deviations without rewriting the original.
- 03After
Separate planned findings from new hypotheses.
Editorial model: a way to reason about the process, not measured platform results.
Keep deviations visible instead of pretending the plan survived contact with reality. A platform outage, missing asset or account access problem may require a change; record when and why it happened. Distinguish planned analysis from exploratory findings in the retrospective. Exploration is valuable, but it should generate the next question rather than masquerade as a prediction made in advance. A lightweight protocol works when another operator could read it and understand what was intended, what actually happened and how the team reached its decision.
“Write possible verdicts before knowing the score.”
02FIELD KITFrom concept to practice.2 SCREENS
Freeze a plan version before collection.
Adapt this to your audience, budget and test scope.
Put it to work.
- 01Freeze a plan version before collection.
- 02Log departures and exploratory analyses.
Notes & sources.
The limit. Preregistering a poor protocol does not validate it; it makes its choices visible.
Further reading is linked throughout this article. Worked examples explain a process; they are not campaign results.
THE ORGANIC CLUB is an editorial project initiated by TokPortal. That relationship is separate from the evidence behind claims. Editorial signatures and image credits are explained in our policy. Our editorial policy →
