Bed mesh visualizer: range, tilt and flatness from a Klipper or Marlin mesh

What is a good bed mesh range?

Aim for a range below your first layer height, usually 0.2 mm. Under that, the mesh only evens out small differences. No firmware publishes a target. Makers’ own figures run from about 0.05 mm, where Voron’s docs say a mesh may not be needed, to 0.5 mm “normal” on the Snapmaker U1 and 3 mm on the Creality K1.

The calculator compares your range with your first layer height. A range larger than the first layer means no single Z offset gives the whole bed a usable first layer, so mesh compensation must stay enabled. What printer makers write about their own machines:

Source	Figure	What it says
Voron docs	about 0.05 mm	“For total variance under ~0.05mm, a bed mesh may not be necessary.”
Raise3D, Pro3 series	0.15	Flatness at or under 0.15 (no unit given): “the printer can print normally”
Snapmaker U1	0.5 mm	“Range under 0.5mm considered normal.”
Sovol SV08	0.6 mm	Within 0.6 mm “will not have any impact on the printing”; between 0.8 and 1.2 mm, measure the bed’s curve with a ruler
Creality CR-10 SE	2 mm	“less than 2mm from the maximum to the minimum is considered normal”
Creality K1	3 mm	Under 3 mm “is within the normal range”; above 5 mm, retighten the Z axis

The large figures are fault checks (is something broken?), not print-quality targets. Bambu Lab, Elegoo, QIDI, Anycubic and Flashforge publish no number. Neither do Klipper, Marlin or Prusa.

Range, variance and deviation

• Range is the highest probed point minus the lowest. Mainsail’s “Range” and the range in Fluidd’s profile list are exactly this, from the probed points. In Mainsail’s code the getter is called “variance”, and Voron’s docs say “total variance” for the same number.
• Fluidd’s chart shows the range of the interpolated mesh by default, the smoothed grid between probe points. It can read a little different from the probed range; switch the chart to “Probed matrix” to compare like with like.
• Standard deviation (σ) is the typical distance of a point from the mean, worked out the same way as Klipper’s graph_mesh.py analyze. A bed with one high corner has a big range but a small σ; a bed with ripples everywhere has both large.
• Deviation usually means one point’s distance from the mean or from the reference point. The highest and lowest points the calculator lists are those deviations.

Check your probe before reading small differences. Klipper’s docs call a spread up to 0.005 mm between repeat readings normal; above 0.025 mm, the probe “does not have sufficient accuracy for typical bed leveling procedures.” Klipper’s PROBE_ACCURACY command measures it. Two meshes that differ by less than your probe’s spread differ only by noise.

What should my bed mesh look like?

Mostly flat, with a range below your first layer height and smooth steps between neighboring points. A steady slope from one edge to the other is tilt, which the leveling screws remove. A bowl, dome or twist is the bed’s own shape, which the mesh corrects. A single point far from its neighbors is worth probing again.

What the map shows	What it is	What to do
One edge or corner high, a steady ramp to the other side	Tilt	Level the screws: the calculator links the tilt to the bed leveling screw calculator
Center high or low, edges the other way (dome or bowl)	The bed’s shape	Leave it to the mesh, measured with the bed at printing temperature
Two opposite corners high, the other two low (a twist)	The bed’s shape	Leave it to the mesh
One point far from its neighbors	A probe misread, debris on the sheet or a real dent	Clean the sheet and probe again. If it stays, it’s real

Heat soak the bed before you judge its shape. For an SV08 range over 1 mm, Sovol advises holding the bed at 65 °C for 30 minutes before meshing.

How the numbers are calculated

The calculator reads every probed point into a grid, front row first, then works out:

range = highest − lowest
mean  = sum of all points / number of points
σ     = √(sum of (point − mean)² / number of points)

Tilt comes from a least-squares plane through all the points: the flat, tilted surface that sits closest to the mesh. Its slope from the left edge to the right edge is the left-to-right tilt; front edge to back edge is the front-to-back tilt. Subtract the plane from every point and what’s left is the bed’s own shape; its range is “Shape left without the tilt”. That is roughly what you’d see after perfect leveling.

The example mesh (a saved Klipper mesh from an Elegoo Neptune) shows the split well: 0.803 mm of range, with the front 0.723 mm higher than the back. Take the tilt out and about 0.10 mm of shape is left. The two parts don’t sum to the total range: the highest and lowest points of each part fall in different places.

The “Turn this tilt into screw turns” link sends the plane’s height at the four corners of the mesh to the bed leveling screw calculator. Screws usually sit outside the probed area, so the real adjustment is a little larger than the corners suggest; run the mesh again after one round of turns.

Where to copy your mesh from

Firmware	Command or file	Rows printed
Klipper	BED_MESH_OUTPUT, the “Mesh Leveling Probed Z positions” lines	Front row first
Klipper	printer.cfg, the #*# [bed_mesh default] block’s points	Front row first
Klipper	BED_MESH_OUTPUT, the “Measured points” lines	Back row first, and interpolated
Marlin bilinear or mesh bed leveling	M420 V, or the grid G29 prints	Row 0 is the front
Marlin UBL (Unified Bed Leveling)	G29 T or M420 V	Top row is the back
Prusa MK4, Core One, XL, MINI	The UBL report: their firmware uses Marlin’s code (Prusa documents no text output)	Top row is the back
Prusa MK3, MK2.5	G81	Back row first

The calculator recognizes each of these from the lines around the numbers and puts the rows in the right order. Two traps:

• Klipper’s “Measured points” is not what was measured. It’s the interpolated mesh, printed back row first, and the “Mesh Range” line above it comes from the same interpolated grid. Paste the “Probed Z positions” lines, or the saved profile, for the real probe readings.
• Prusa’s G81 always prints 7 × 7. With the factory setting of 3 × 3 probing, 40 of the 49 numbers are interpolated, so the standard deviation and the tilt rest on nine real points.

Klipper subtracts the Z at zero_reference_position from every point when one is set, so the mesh reads 0 at that position. That shifts every value equally: range, σ, tilt and shape stay the same. Only the mean and the highest and lowest values move.

After the mesh: fade and per-side correction

• Fade. Klipper’s fade is off by default (fade_end: 0), so the mesh correction applies at every layer. Its docs suggest fade_end: 10 if you want the correction to taper out over the first 10 mm. Marlin’s DEFAULT_LEVELING_FADE_HEIGHT is 10 mm in the stock configuration, and M420 Z0 disables fade.
• Prusa’s Bed Level Correction shifts the left, right, front or back by up to ±100 µm. Prusa’s knowledge base says even 20 µm can make a big difference. Next to the ranges above, a usable correction is tiny.