Calibrate and verify#

With both MCUs on mainline, the config translated, and plugins in, calibrate the eddy probe and motion, then prove it with plastic.

Eddy and tap#

Run in order, SAVE_CONFIG after each:

LDC_CALIBRATE_DRIVE_CURRENT
PROBE_EDDY_CURRENT_CALIBRATE — builds the freq→height table
PROBE_EDDY_CURRENT_TAP_CALIBRATE — sets tap_threshold

Belt tension#

Aim for ~110 Hz plucking the XY belts over their 150 mm span. (The Z belt is factory ~137 Hz, but that’s a 21.5 cm span — a different length, so don’t compare the two numbers.) Tighter than that only cleans up the shaper graphs on paper and adds ringing. Klipper’s belt similarity test is unreliable here — the head-mounted accelerometer is noisy at standstill from motor hold-current buzz, so its readings swing run to run. Check tension by ear or a phone tuner instead (Android works; iPhone noise suppression corrupts the reading).

Input shaper#

TEST_RESONANCES — mainline’s chunked-FIFO LIS2DW read gives a clean trace, and with a wide [resonance_tester] window (not the vendor’s pinned 35–45 Hz) the real resonance surfaces. On this host the diagonal test needs the camera stopped — see troubleshooting. Measured on this machine, SHAPER_CALIBRATE lands X at mzv ~64.6 Hz and Y at ei ~59.6 Hz with axis noise ~5–8 mm/s² — the real resonance sits in the 55–65 Hz band, nowhere near the vendor’s 40 Hz.

Functional check#

G28 homes via eddy.
BED_MESH_CALIBRATE METHOD=rapid_scan completes.
A test print runs.

Set max_accel to the binding per-axis SHAPER_CALIBRATE limit for clean geometry, not the resonance-test ceiling.

The first print is the real proof#

Klippy ready with both CAN UUIDs connected proves the stack is alive, but only a supervised first print proves the tap reference. The tap reference survives a config restore — but watch the first layer anyway. Only plastic proves it.