Replication needs more than a budget

In late September, NIH laid out R3PEATS, a five-year program to make biomedical research more rigorous and reproducible. Science reported that it includes a literature search tool and a network modeled on the UK Reproducibility Network — but not the large replication effort some had expected.

Critics were quick to note what’s missing. Earlier plans for dedicated replication centers drew objections from NIH’s advisory council, and the current plan scraps them. By one estimate, the program works out to roughly 0.07% of NIH’s annual budget.

Where replications actually stall

Whatever the budget, every replication starts the same way: someone reads a methods section and tries to do what it says. That is where the trouble begins. When the Reproducibility Project: Cancer Biology planned to repeat 193 experiments, none were described in enough detail, and in the end only 50 experiments from 23 papers were repeated. It’s a familiar story — in Nature’s survey of 1,576 researchers, more than 70% had failed to reproduce another scientist’s experiment.

Money can pay for the second attempt. It can’t recover the detail that was never written down the first time: the mixing step, the incubation that ran long, the reagent lot that changed mid-study.

You can’t replicate a step nobody recorded.

Fix it at the source

The cheapest replication is the one designed in from the start. If each run produces a time-stamped, step-by-step record alongside its data, the next lab doesn’t have to reconstruct the method from memory and email threads. It can read it.

R3PEATS is right to treat reproducibility as a culture problem. Records captured automatically at the bench make that culture easier to keep — which is why we built Augmentiv for academic groups as well as industry. Why every result needs a run record →