A slicer works with centrelines. It decides where the middle of each extruded bead should go, then assumes that bead comes out a specified width. Every dimension the printer produces is that centreline plus half a bead on each side.
So if the real bead is wider than the assumed one, an external dimension grows by the error twice over — once on each face — and an internal one shrinks by the same amount. That is why the two most common complaints in this category point in opposite directions and have the same cause.
Calibrate first, compensate second
This ordering is the single most useful thing on this page.
Slicers offer compensation settings — hole compensation, X/Y size compensation, elephant's-foot compensation — that shift geometry by a fixed amount to cancel an observed error. Applied on top of an uncalibrated flow rate, they cancel the symptom on one test part and leave the machine wrong everywhere else, because the underlying error scales with feature size and the compensation does not.
Get the extrusion right, then measure, then compensate for whatever residual is left. Reversed, you are tuning two unknowns against one measurement.
Why circles suffer more than flats
An internal circular feature loses accuracy three separate ways at once. The path is approximated by short chords that all sit inside the true curve. The bead is pressed against a concave surface, where displaced material has fewer places to go. And as the loop cools, its contraction closes the hole rather than opening it, because everything shrinks toward its own centre.
The three errors add in the same direction, which is why a hole is nearly always undersized and rarely oversized, and why the fix is a deliberate allowance rather than a search for the setting that went wrong.
How to measure so the number means something
- Let the part reach room temperature. A print measured warm is measured mid-contraction.
- Avoid the bottom couple of layers entirely — they carry the deliberate squash of the first layer and any bulge above it.
- Measure away from the seam, where the start-and-stop of each loop leaves a local excess.
- Close the calipers gently. It is entirely possible to compress a plastic wall by more than the error being chased.
- Measure a printed calibration object with known nominal features, not the interesting part. A part with fillets, chamfers and text gives you nowhere reliable to put the jaws.
When the material, not the machine, owns the error
Above a certain size, contraction during cooling dominates everything the slicer does, and it is a property of the polymer rather than the profile. Compensating for it means scaling the model before slicing, which is what the shrinkage compensation calculator exists to work out from a measured test part.
The distinction is easy to check: a machine error is roughly constant across a dimension, while a shrinkage error grows in proportion to it. Print a long test bar, measure it in two places, and the two look nothing alike.
The estimate that is also an accuracy problem
One page in this category is not about a physical dimension at all — it is about a print that used far more material or far more time than the slicer predicted. It sits here because it is the same class of failure: a model of reality that assumed a number, and reality using a different one. The assumption in that case is the density value in the filament profile and a motion model that does not know the machine's real acceleration limits.