Almost every failed print I've ever had traces back to the same thing: the first layer. Not the slicer, not the filament, not some exotic retraction setting I read about on a forum at midnight. The nozzle was the wrong distance from the bed, and everything downstream was a consequence of that one sin. Bed levelling is the eternal war, and this is the story of finally winning a battle in it.
the paper-drag years
For ages I did it the way everyone starts: a sheet of paper under the nozzle, four corners and the middle, turning the thumbwheels until I felt a slight drag. It works, in the sense that a stopped clock works. The trouble is that "slight drag" is a feel you have to relearn every time, the bed moves with temperature, and the paper itself is about 0.1mm thick, which is a meaningful fraction of your first-layer height. You're calibrating against a fuzzy reference with your fingertips. No wonder it drifts.
The symptom catalogue became familiar. Filament not sticking and curling up into spaghetti meant the nozzle was too high. A first layer that looked translucent and smeared, with the nozzle ploughing through it, meant too low. Sticking on one side of the bed but not the other meant the bed wasn't flat, which paper-drag at five points papers over but never actually fixes.
heat it up first, every time
The single biggest improvement before I bought any hardware was simply levelling hot. Bring the bed and hozzle up to printing temperature, let them soak for five or ten minutes, then level. Aluminium beds bow as they heat; the gap you set cold is not the gap you print with. I'd been levelling a different machine to the one that actually ran the job.
While I was at it I checked the bed for physical flatness with a steel rule and a torch behind it, looking for light under the edge. Mine had a gentle dip in the centre, the classic result of a single central mounting screw pulling the middle down. Some careful adjustment got it closer, but a warped bed is a warped bed, and that's where the electronics finally earned their keep.
the probe changes the game
I fitted an inductive Z-probe, the sort that triggers off the metal bed without touching it. Suddenly the question stopped being "can I feel the drag" and became "what does the firmware measure", which is a much better question because it's repeatable.
The important step that people skip is the Z-offset calibration. The probe triggers at some height above the bed, and that's not where you want the nozzle. You measure the difference once, carefully, and tell the firmware. In Marlin that lives in the configuration and you can tune it live:
G28 ; home all axes
G29 ; probe the bed and build a mesh
M851 Z-1.85 ; set probe Z-offset (mine, yours will differ)
M500 ; save to EEPROM
The number that worked for me was around -1.85mm, found by the time-honoured method of printing a single-layer square, watching the first line go down, and nudging the offset 0.025mm at a time with M851 until the extrusion looked like a flat ribbon rather than a round string or a smear. Save it to EEPROM with M500 so it survives a reboot, because asking it the cold way every print is exactly the manual misery I was trying to escape.
mesh levelling, for the bed you actually have
The real revelation was mesh bed levelling, the G29 above. Instead of pretending the bed is a tilted-but-flat plane, the printer probes a grid of points and builds a height map, then compensates in software as it prints the first layer. My warped centre, the thing no amount of thumbwheel turning could fix, just stopped mattering. The nozzle rises and falls a few hundredths of a millimetre to follow the surface, and the first layer comes out even from corner to corner.
There's a subtlety worth knowing here. The mesh only compensates over the first few layers and then fades out, because you don't want the printer riding the surface contours all the way up a 200mm-tall vase; that would put waviness into the walls. So the mesh fixes adhesion, not geometry higher up, which is exactly the right division of labour. Get the height map honest and the rest of the print stops caring about the bed entirely.
A few things I'd tell my earlier self:
- Probe more points than feels necessary on a warped bed. A 3x3 grid smooths over a dip that a 5x5 actually captures.
- Run the probe sequence on a clean bed. A stray blob of dried glue stick under one probe point throws the whole mesh off and you'll chase a ghost for an hour.
- Re-probe after any mechanical change. New nozzle, new bed surface, you moved the printer: re-mesh. It takes ninety seconds and saves a failed three-hour print.
was the fuss worth it
Completely. First-layer failures went from "most prints, eventually" to genuinely rare, and the ones that do fail now are honest filament or adhesion problems rather than geometry I set wrong with a piece of paper. The whole stack, level hot, set the Z-offset once with care, let mesh compensation handle the warp, turned the most temperamental part of the machine into something I barely think about.
I won't pretend the war is over. A new build surface or a knock to the gantry and I'm back out probing. But I've stopped fighting the bed by feel and started measuring it, and measuring beats feeling every single time. The thumbwheels can rust shut for all I care, and the sheet of paper has gone back in the drawer where it belongs.