An unexpected result can be the most interesting part of an experiment. The important reading question is whether the result generated a new hypothesis or tested one specified before those results were known. Both activities contribute to research, but they support different statements about the evidence.
Identify what the experiment was meant to do
The Center for Open Science distinguishes generating hypotheses from testing predictions made in advance. Its preregistration guidance allows unplanned analyses, provided their exploratory status is clear; discovery and confirmation can therefore appear in the same research project.Center for Open Science — Preregistration (opens in a new tab)
Exploration asks what patterns might deserve attention. Confirmation asks how a specified prediction fares under a defined test. Neither label is a synonym for a positive or negative finding, and an exploratory result is not automatically unimportant.
| Activity | Useful output |
|---|---|
| Exploration | A clearly described pattern and a new question |
| Confirmation | Evidence about a prior prediction under specified conditions |
| Further exploration | Possible explanations for an unexpected result |
| Follow-up test | New evidence that can challenge the proposed explanation |
The distinction concerns how the evidence was obtained. A polished explanation written after an experiment cannot, by itself, tell the reader which decisions were made before the outcome was visible.
Follow a discovered signal into a separate test
Consider an original hypothetical peptide experiment measuring twelve cellular readouts. The investigators inspect all twelve and notice a particularly large change in one. That observation can motivate a focused hypothesis about that readout.
If the paper then presents this readout as though it had been the sole prediction from the beginning, the reader loses information about the search that produced it. The other measured outcomes help explain the opportunity to encounter an unusual pattern.
A subsequent experiment can specify the selected outcome, comparison and analysis before collecting or inspecting its relevant data. It then supplies a distinct test of the prediction arising from the first experiment.
Look beyond whether a hypothesis was mentioned
A broad prediction that a peptide changes cell behaviour leaves many choices open. Which outcome, time point, subgroup and comparison will count as the test? A confirmatory claim becomes easier to evaluate when those choices were specified with sufficient detail.
The relevant issue is not whether every laboratory action was frozen forever. Equipment failures, missing measurements and unexpected observations may justify changes. The report should explain those changes and distinguish decisions informed by the results from decisions made without seeing them.
An analysis chosen after examining several alternatives can still be useful. Describe it as the analysis that produced the reported pattern, alongside the alternatives considered, rather than implying that its selection was unrelated to the data.
A single paper may contain a planned primary comparison, exploratory subgroup work and a new validation experiment. Assign the evidence role at the level of the analysis or experiment instead of forcing one label onto the entire paper.
Use wording that preserves the evidence stage
For discovery, a precise summary might say that an analysis identified an association warranting a targeted follow-up. For a planned test, describe the prediction, relevant experimental setting and result without suggesting that all possible explanations have been eliminated.
A result consistent with a prediction supports that prediction under the tested conditions. It does not establish a universal mechanism merely because the test was planned, nor does it make contradictory evidence irrelevant.
When timing is unclear, state that the report does not establish which choices preceded data inspection. This is more accurate than assuming deliberate misrepresentation or awarding confirmatory status from the wording of the abstract.
Sources and further detail
- Center for Open Science — Preregistration (opens in a new tab)
Guidance on discovery, confirmation and transparent unplanned analyses read. The twelve-readout example and reading questions are original; no empirical result is claimed.
Sources checked 19 September 2026. Worked examples are illustrative unless a supplied report is explicitly identified. This article has not undergone independent scientific peer review.