I needed a bracket to hold a small fan against the back of a Pi enclosure at a slightly silly angle. I spent forty minutes searching for one that didn't exist, then realised I was being daft. This is exactly the thing the printer is for. The maths of it is simple: the part is worth about thirty pence in filament and an evening of my time, and I'd already burned most of the evening looking for it.
drawing the thing
I use FreeCAD for this, not because it's lovely (it isn't, the workflow has sharp edges) but because it's free, it's parametric, and the sketch-then-pad approach matches how I think about a part. Draw the profile, give it dimensions as named constraints, pad it to a thickness, cut the holes. When I get the hole spacing wrong, and I always get the hole spacing wrong first time, I change one number and the model rebuilds.
The constraints are the whole point. The first version had the fan screw holes 32mm apart because I'd measured the wrong pair of holes on the fan. Because they were parametric, fixing it was changing 32 to 24 and re-exporting. If I'd modelled it with raw geometry I'd have been redrawing half the sketch.
the iterations
First print: the bracket fit, but the fan fouled a capacitor on the board by about a millimetre. So close. Second print: added a 2mm standoff, fine, but I'd printed it flat and the layer lines ran across the thin neck, so it snapped when I tightened the screw. Layer orientation is the thing nobody tells you and everybody learns by breaking a part. Lines are strong along the layer, weak across it. Third print: rotated it 90 degrees on the bed so the load runs along the layers, dropped the infill from 20% to a more sensible setting for a chunky part, and it held.
Three prints, maybe ninety minutes of machine time spread across an evening, and a part that fits one specific board in one specific orientation. You cannot buy that. That's the actual value of having a printer in the house: not the headline projects, but the thirty-pence bracket that solves a problem nobody else has.
The STL is sitting in a folder named after the project, which I will absolutely fail to find again in a year. That's a problem for future me.