The reproducibility problem isn't what you think it is
Most people talk about the reproducibility crisis in science as if it's a recent collapse. It's not. It's been ongoing since at least the 1980s, and the actual mechanics of why results don't replicate are far more mundane than the dramatic headlines suggest. I've spent over a decade running experiments, publishing them, and watching other people try to replicate what I did. The thing nobody wants to admit out loud is that replication failure is usually caused by tiny methodological divergences that neither side considers worth reporting. You run an experiment using a particular strain of lab mice, bought from a specific vendor on a specific day, under fluorescent lights with a particular cycle length. Someone else three years later buys mice from a different vendor, uses LED lighting, and gets a completely different outcome. Both papers get published. Neither is wrong. Neither is right in the way either author claims. The effect size from the original study drops to near zero in independent hands. This happens constantly. It's not fraud. It's just sloppy documentation culture.Are We Managing To Destroy Science
The question comes up regularly on forums like this, and the answer is complicated because it depends on what version of "destroy" you mean. If you mean collapsing the scientific enterprise entirely, no. Science is too entrenched. If you mean slowly degrading the quality of peer-reviewed output through structural incentives that reward speed over rigor, then yes, that's happening. The core mechanism is simple and well-documented. Academic publishers operate on a model where journal impact factor directly influences researcher promotion, grant funding, and institutional reputation. This creates pressure to publish novel findings with large effect sizes. Null results don't get published at the same rate. Negative findings disappear into file drawers. The literature becomes skewed toward optimistic results simply because the system filters them in. I encountered this firsthand during a project investigating dose-response relationships in a commonly used cell line. My initial experiments showed clean, statistically significant results. I submitted to a mid-tier journal. The review process took eleven months. When it came back, the reviewers asked for three additional experiments that I couldn't reasonably complete within my funding timeline. I re-submitted elsewhere and got published in eight weeks with substantially lighter review. The paper was weaker for it, but the journal's metrics went up. This is not an outlier story. This is the standard workflow for most working researchers. The file drawer problem compounds over time. Meta-analyses that attempt to correct for publication bias often find that the true effect size in published literature is roughly twice what it actually is. This isn't speculation. It's been demonstrated across multiple fields using statistical techniques like p-curve analysis and trim-and-fill methods. Here's the part that most people miss: the problem isn't primarily bad actors. It's systemic incentive design. A researcher who rigorously documents every protocol detail, who publishes negative results, who openly shares failing experiments, will fall behind a colleague who chases publications using looser standards. The colleague gets promoted. The rigorous one gets passed over. After twenty years of this, the rigorous ones leave academia or stop trying.Some fields handle this better than others. Clinical trials have regulatory oversight that enforces registration and result reporting. Experimental physics publishes preprints on arXiv before journal submission, creating transparent timestamps. Psychology has made genuine institutional changes since 2015, including mandatory pre-registration and registered reports. But these are exceptions, not the rule. Most fields still operate on the old model where speed and novelty trump transparency.
I had another experience that illustrates this more concretely. A collaborator and I built a custom apparatus for measuring a specific physical property. We documented everything—materials sourced, dimensions, environmental conditions, calibration procedures. Two groups tried to use our setup to extend our work. One succeeded. The other failed repeatedly and concluded our results were unreproducible. The issue turned out to be that the second group used a slightly different thermal isolation material that introduced a systematic offset they never measured. They published a paper saying our work couldn't be replicated. Our paper stayed valid. The failure was theirs, but the damage to the perceived reliability of our work was real. This is how the degradation happens. Not through corruption, but through accumulated small failures of documentation, communication, and incentive alignment.
