Even understanding the maths it is still slightly amazing that it works!
You want a pinhole with as much of the lamp light as you can get hitting it nearly parallel to begin with. An unresolved point source would be ideal but you will have to settle for a finite sized pinhole.
It was always a bit of a bugbear for calibrating aperture synthesis - to get fringes on the longest baselines you needed unresolved point sources but to calibrate the amplitude you wanted them to have a constant brightness too. The number of such stable point sources is rather small.
Try with a 0.1mm hole and 1m drift range as a starting guess.
Might be more fun to play around with it on the yellow sodium D-lines where you can see immediately what the results are. It was a favourite for fourier optics practicals back in the days when lasers were rare exotic beasts that required a lot of nuturing to work.
Symmetrical around the brightest objects. Snag is many different images give the same autocorrelation and picking the right one can be tricky.