Preregistering research can make it clearer which questions and analysis choices researchers settled before seeing their results. But in experimental biology, trying to specify every step in advance can clash with how lab work unfolds. In an article published in PLOS Biology on November 20, 2026, Olavo B. Amaral and coauthors argue that preregistration should be adapted to the kind of research being done.
Their case begins with the distinction between exploratory research, which develops hypotheses, and confirmatory research, which tests them. When researchers analyze data and then present a resulting hypothesis as if they had predicted it, they engage in “HARKing”—hypothesizing after results are known. The authors say this can let researcher expectations shape which analyses are run or reported, distorting the scientific record.
Registering a protocol before data collection can help readers distinguish decisions made in advance from those made later. The authors describe other proposed benefits, too: planning analysis early, giving peer review a chance to improve study design, and reducing the influence of results on publication decisions in Registered Reports. In that format, a journal reviews a study protocol before data collection and can accept the work in principle regardless of its eventual results.
Yet the authors say preregistration faces particular practical hurdles in lab biology. They draw on their experience in the Brazilian Reproducibility Initiative, a multicenter effort to replicate biomedical experiments. There, researchers were asked to specify protocol decisions and criteria for judging whether experiments were valid. The authors report that researchers struggled to define some choices in advance and often departed from their own protocols. They present this as a challenge in implementation, not evidence that preregistration has no value.
One reason, they suggest, is that lab researchers may find it harder to collect data than to analyze them. Experiments can require adapting a method to make it work, particularly when a method is new or researchers are still learning it. By contrast, analysts can repeat the same data analysis in many ways. Without a prespecified plan, trying different options can become “p-hacking”—searching for a significant or otherwise desired result.
That difference leads to a practical emphasis: be especially clear in advance about analysis plans and rules for including or excluding experiments, animals, or samples. If a decision must depend on what the data show, the authors recommend setting out a decision rule ahead of time. Unplanned analyses can still be done, they say, but should be labeled as unplanned.
How much protocol detail to preregister depends on the purpose. For confirmatory experiments, and especially when seeking peer review through a Registered Report, a protocol should be detailed enough to judge whether the methods are appropriate. Pilot work can help show that the planned approach is feasible. But trying to anticipate every experimental adjustment may be counterproductive: some choices emerge as work proceeds.
For projects that begin with exploration, the authors offer several ways to preserve transparency without pretending that every next step can be known in advance. Researchers can separate an exploratory stage from a later confirmatory experiment and preregister the latter. They can also time-stamp protocols sequentially as a project develops, creating a record of decisions, or preregister general standards for well-established methods. Sharing an experimental log can make changes and discarded experiments easier to assess.
The authors caution that preregistration alone cannot prevent selective reporting. Researchers can still repeat experiments or discard data under criteria chosen after the fact. Prespecified exclusion rules help, but a complete log and sensitivity analyses that include discarded experiments can make those choices more transparent. Their argument is for using preregistration where it can limit bias while keeping a visible record of necessary changes and remaining uncertainties.