Preregistration in a wet lab is harder than it looks

We preregistered our last three studies. Two of those went well. One did not, and that is the interesting one.

Where it worked

For the banding-reversal study, preregistration forced us to commit to a primary endpoint before we saw any echocardiography. When the secondary endpoints turned out more exciting, we could not quietly promote one of them — and the paper is more honest for it.

Where it broke down

Wet-lab work has a failure mode the psychology literature mostly does not: the animal model itself can misbehave. Our second cohort had an unexpected mortality rate, and no amount of pre-specification tells you what to do about that.

What we settled on:

Where I landed

Preregistration is not a guarantee of rigour. It is a record of intent, and that record is most valuable precisely when things go sideways.