There are two types of reproducibility crisis in biomedical sciences. Type 1 is "shooting first, drawing bullseye later". Type 2 is the misunderstanding of scientific modeling.

It's very easy to understand Type 1 crisis, so I will focus on Type 2. Let me start with two extreme cases of lacking reproducibility. The first one is the discovery of a supernova. After that, it can be analyzed and modeled theoretically. Can you predict when and where the next supernova is? Very unlikely. Therefore, this study lacks reproducibility.

The second is the famous Wow! signal. In 1977, Ohio State University's "Big Ear" radio telescope that support Search for Extraterrestrial Intelligence (SETI) project received a strong signal unlikely from any known natural source. Astronomer Jerry R. Ehman was so impressed by the result that he wrote the comment "Wow!" on the signal chart, leading to the event's widely used name. However, the signal was never received again, though it remains the strongest candidate for an extraterrestrial radio transmission ever detected.

Intuitively, most of the scientists would not consider the first case a "crisis". The occurrence of supernova follows the rule of star evolution, and its distribution in the universe is the events of probability. Both can be modeled from the knowledge of astrophysics. No one will complain why you could not find the next supernova. In contrast, without the second detection, Wow! signal could not support the existence of Extraterrestrial Intelligence. It could be a random, singular event in this big universe, not produced by any intelligence.

In the recent discussion of "reproducility crisis", cases of type 1 and 2 are often confused. To distinguish them from each other, we have to identified the drifting factors in our studies. Cell lines in culture and mouse colonies can change over time, simply because of unintended selection and genetic drifting from geographic segregation. Patient cohorts can be intrinsically varied even by the same admitting criteria, because that is the nature of heterogeneous population. Therefore, the data of the studies need to include the current status or features of the drifting factors, and the modeling can be done accordingly. By comparing the drifting factors, we will get the opportunity to discover the real driving factors.

When we see two study outcomes, e.g. therapeutic response of tumors in mice, are different, we should not rush to conclude there is a problem in reproducibility. We should ask, what are the drifting factors in these two studies? How should we find the difference between their status? We have to be very aware that overemphasis on the absolute reproducibility will drive researchers to commit Type 1 crisis.