11
Fifty-Two Cycles
The requirement with which the last chapter ended sounds noble and means nothing until it turns into a procedure.
Lying to oneself being technically inconvenient: if you translate that from the noble into engineering, how does it work?
It is arranged insultingly simply, and of course I was not the one who thought it up.
Before measuring anything, I write down in a separate file what exactly I expect to see: what quantity, in which direction, what size of effect and at which outcome I am obliged to admit I was wrong. The file goes into the code repository, after which a fingerprint is taken of it, a short string that changes completely if you correct so much as a comma in the file, and this fingerprint is entered into a common register with today’s date.
Only after that is the measurement run.
Cheating, of course, has not gone anywhere: I can still rewrite a prediction after the fact, but then the fingerprint will stop matching, and anyone who opens the register will see it. The garden of forking paths, in which I had such a nice stroll in the last chapter, is closed off here purely mechanically: a single path remains, and it is written down in advance.
Over a few years nine rules have grown up around this procedure, of which I will name three. Read the literature before registering the prediction, not after. Do not count fewer than a thousand independent runs as a result. And as a separate item, hard-won from my own mistakes, there is a check for tautology: whether what I am going to measure might turn out to be a simple consequence of how I defined it.
As of today there are fifty-two predictions in the register.
Tras la compra
Este capítulo forma parte de la edición completa
Los primeros capítulos son gratuitos. El libro completo está disponible tras la compra y queda en tu biblioteca.
· · ¿Ya lo compró? Inicie sesiónUn correo cuando salgan capítulos y experimentos nuevos
Como mucho una vez por semana. Sin registro; baja con un clic.