Explore a finding.Test it on new data.
Record exploratory work, lock one candidate before confirmation data are opened, and test it on independent samples. Keep exploratory findings and confirmatory evidence distinct throughout the research.
A boundary between finding and testing.
- 01
Explore
Use the supported summaries, plots and comparisons. Keep the work labelled exploratory.
- 02
Lock one candidate
Fix the outcome, hypothesis, analysis, sample size and stopping rule before confirmation data are opened.
- 03
Confirm independently
Test the locked candidate using independent, unused samples in the supported confirmation route.
- 04
Report the outcome
Retain the confirmation result and its scope whether significant or not.
Recording exploration does not adjust for selection. The first confirmation route is one untransformed outcome and one two-sided Student’s t-test.
New data means new evidence.
| Situation | What follows |
|---|---|
| Independent confirmation samples are ready | Lock the candidate before accessing them through the confirmation route. |
| Confirmation samples have not been collected | Save the candidate and prepare a confirmation-study plan. |
| The candidate changes after seeing confirmation | New confirmation data are required. |
Separate animals or donors are needed; re-splitting already seen data does not make them unused. Whether data were seen elsewhere remains a research assumption.
A candidate with a route to confirmation
- A recorded exploration with its candidate and the reason it was chosen
- A locked confirmation plan
- A confirmatory result and supported claim, significant or not
- A confirmation-study plan when new data are still needed
Exploration and confirmation conditions.
Explore, on the record.
Use descriptive statistics, plots, and condition comparisons across a set of continuous outcomes. Data versions, operations, and results are recorded as exploration.
Lock in one candidate.
Before any confirmation data are opened, lock the outcome, comparison direction, preprocessing, analysis, hypothesis, significance level, planned sample size, and stopping rule.
Confirmation data kept out of sight.
Confirmation data are received through a separate route and are not shown to you or your AI until the candidate is locked.
A real confirmatory test.
The locked candidate is tested on independent samples with a standard two-sided t-test. Significant or not, the result returns with the claim it supports.
No confirmation data yet?
Keep what exploration established, save the candidate, and get a plan for the confirmation study you still need.