The first layer is not consistently wrong; it is wrong differently in different places. One region is transparent and ridged, another shows round separated lines, and there is usually a diagonal across the plate between them.
Two different faults are hiding in this description
The plate is not parallel to the plane the toolhead moves in. A tilt. Distance from nozzle to plate varies smoothly across the bed, usually worst along one diagonal.
The plate is not flat. A dish, a dome, or a wave. Distance varies non-linearly, so the middle can be wrong while all four corners are right.
Mesh levelling addresses the second and, on most machines, corrects the first as a side effect. That is exactly why so many people are stuck: a large mechanical tilt corrected in software eats up the compensation range, drags the toolhead through a large Z movement on every pass, and leaves the machine printing acceptably but never well.
Tram the bed mechanically first, then mesh. Not instead of, and not the other way round.
Map your plate in one print
Guessing from a failed part is slow. Print a diagnostic instead: a single-layer square patch in each corner and one in the middle, in a cross or a grid across the whole plate. Any slicer will do this from five copies of a thin square; some ship one ready-made.
Then read it:
- All five identical: the plate is fine and the problem is the overall offset, not its distribution.
- Opposite corners differ, middle in between: a tilt. Adjust the bed screws or the gantry, not the mesh.
- Corners right, middle wrong: the plate is dished or domed. This is what mesh compensation is genuinely for.
- One corner wrong and the rest fine: a local defect — a screw over-tightened, debris under the sheet, or a damaged magnet layer.
- Pattern changes between prints: not a shape problem at all. See bed mesh no longer accurate.
Probe the bed hot
A heated plate is a different shape from a cold one. Aluminium expands, the sheet's constraint at its mounting points resists that expansion, and the result is a plate that bows measurably as it warms.
A mesh generated cold and used at 100 °C describes a plate that no longer exists. Probe at the temperature you print at, with the plate soaked for long enough to be evenly hot — several minutes, not thirty seconds. On machines that heat the chamber as well, soak that too.
Check the gantry before blaming the bed
On a machine with two Z motors or two lead screws, the gantry can sit lower on one side than the other. Every symptom then looks like a tilted bed and no amount of bed adjustment will fix it, because the moving plane is the tilted one.
Measure from the top of the frame down to the gantry at both ends with a ruler or a printed spacer. If the two differ, correct that first: drive the gantry gently up against its top stops so both sides bottom out together, or adjust the belt or coupling depending on the machine. The full treatment is on gantry not square.
Things that quietly cause this
- A single grain of plastic under a removable sheet. It lifts a whole region by more than the tolerance you are chasing.
- Bed springs at different compressions. Springs are a compromise that relaxes over time; solid mounts or silicone spacers hold their setting.
- Too few mesh points on a large plate. A plate over about 300 mm across needs a denser grid to describe its shape — a sparse mesh interpolates straight lines across a curve.
- A probe reading the sheet's texture rather than its surface. Inductive probes read the steel under the coating; that is fine, but a textured sheet with a thick powder coat changes the relationship between what the probe sees and where the nozzle actually is.
When the whole plate is wrong in the same way
If the whole plate is uniformly too close or too far, this is not your page — see first layer too high or first layer too squished, both of which are single-number offset problems and take minutes rather than an afternoon.
Where it bites hardest
Large beds, thin plates and any material that demands a perfect first layer. Polyamides and polypropylene are unforgiving because their bond is marginal to begin with. Long, flat parts spanning most of the plate suffer most of all, since they cross the entire error rather than sitting in one well-behaved region of it — which is a good argument for printing test parts at the plate's edge rather than always in the comfortable middle.