What astrophotography taught me about experiments

I photograph the night sky, and it has taught me more about experimental practice than most of the courses I have taken.

Why you stack

You cannot brute force a faint target with one long exposure. You take many and stack them, because signal adds coherently across frames while noise adds in quadrature. The improvement goes as the square root of the number of frames.

That is the same reason a claim about registration accuracy needs a hundred paired cases rather than one striking example. A single beautiful result is a single sample from a distribution you have not characterised.

Why you shoot calibration frames

Darks capture your sensor thermal signature. Flats capture the vignetting and dust in your optical path. You subtract them because your instrument has a systematic character that has nothing to do with the sky, and if you do not measure it you will publish it as astronomy.

That is the same reason a comparison needs a matched control rather than a number copied from someone else paper. Their number carries their instrument.

Why you learn to distrust a beautiful frame

A Milky Way image can be pushed in post until it looks spectacular and means nothing. The structure you are admiring is stretched noise. It is genuinely difficult to tell by eye, and the only defence is knowing what your processing actually did.

A deformation field can be smoothed until it looks plausible and has stopped describing the patient. Same trap, different domain.

Knowing where that line sits, in both cases, is mostly a matter of having crossed it before and been caught.