The model says 20 mm and the calipers say more. Not by much — a couple of tenths — but consistently, on every face and every part, and enough that press fits are tight and mating parts do not go together.
Eliminate in this order
Each step below rules something out cheaply, and doing them out of order is how people end up with a machine that is correct on one part and wrong on the next.
1. Check the model. Open the file and measure it in the slicer. Downloaded models are sometimes not the size their description claims, and scaling applied and forgotten is common. Thirty seconds, and it removes an embarrassing possibility.
2. Measure somewhere legitimate. Not across the seam, where every layer leaves a small excess. Not within the bottom couple of layers, where elephant's foot lives. Not across a corner, which bulges for its own reason. And not with the calipers squeezed hard — plastic compresses, and a firm grip can hide the entire error you are chasing.
3. Let it cool. A part measured warm is measured mid-contraction and will shrink further on the bench.
4. Do the two-measurement test. This is the useful one, and it is described below.
5. Calibrate flow. If the bead is wider than the slicer assumed, every external dimension gains that excess twice — once on each side.
6. Only then apply size compensation.
The two-measurement test
Print one object containing both a short feature and a long one — a bar with a 20 mm step at one end and a 100 mm overall length is ideal. Measure both.
- Both dimensions are out by the same absolute amount — say 0.2 mm each. The error does not scale with size, so it comes from the bead: extrusion width, flow, or a worn nozzle laying a fatter line than the profile expects.
- The long dimension is out by roughly five times as much as the short one. The error scales with size, so it is contraction. That is a material property, and it is corrected by scaling the model rather than by adjusting the printer — the shrinkage compensation calculator takes an intended and a measured dimension and gives you the factor.
One print settles which of two completely different fixes you need, and skipping it is how people apply shrinkage compensation to a flow problem and make small parts wrong in order to make large ones right.
The worn-orifice case
A brass nozzle that has printed abrasive filament wears its orifice larger, and a larger orifice lays a wider bead than any setting accounts for. Suspect it if the machine used to be accurate, if it has run carbon fibre, glass fibre, wood-filled or metal-filled material, and if fine details have also become mushy. Nozzles are consumables — fitting a fresh one is faster than measuring a worn one.
Using size compensation properly
X/Y size compensation shifts every external perimeter inward by a fixed amount. It is the right tool for the residual that remains after flow is correct, and the wrong tool for anything else.
Two cautions. It shifts internal features outward at the same time on most slicers, which is usually what you want — but it means a value tuned on an outside dimension will change the fit of every hole in the part. And because it is a fixed offset, it cannot correct an error that scales with size; using it for shrinkage will get one dimension right and everything else wrong.
Enter a negative value equal to half your measured excess if the slicer applies it per side, or the full excess if it applies it per dimension. Which convention your slicer uses is worth confirming with one test print rather than assuming.
Where this shows up worst
Press fits, bearing seats, parts that stack, and anything designed against a metal component. A 0.2 mm error is invisible on a decorative print and fatal on a shaft that was meant to slide.
It compounds, too. Two mating printed parts each oversized by 0.2 mm are out by 0.4 mm at the joint, which is why assemblies fail long before individual parts look wrong.
Materials that behave differently
Fibre-filled grades are the most dimensionally stable material on this site and hold external dimensions best. The polyamides are the least stable and will happily move more than the whole error described here just from absorbing moisture over a fortnight — a nylon part measured on the day it was printed and again a month later is genuinely a different size. ABS and ASA contract enough that the two-measurement test above is worth doing for each new spool, not just each new machine.