Reading the metrics

After each sample the optimizer reports two groups of metrics: accuracy — how closely the fit matches the measurements — and diagnostics — whether the data is rich enough to trust that fit. A calibration is good when accuracy is low and the diagnostics confirm the Workspace was well covered.

Accuracy

Metric	What it means
RMSE	The headline accuracy number — the root-mean-square of the residual position errors across all samples, in mm. Roughly: how far the calibrated model still misses the measured target. ≤ 1.0 mm reads green, 1.0–2.5 mm amber, above 2.5 mm red. Lower is better.
Max error	The single worst residual error, in mm. If it sits far above the mean, one bad sample or pose is likely dragging the fit — worth inspecting in the samples table.
Std deviation	The spread of the residual errors, in mm. Low spread means the remaining error is consistent across the Workspace rather than concentrated in a few poses.

Diagnostics

Metric	What it means
Observable params	How many model parameters the data can actually identify (e.g. "7 / 10"). When the two numbers match (shown green) every parameter is constrained; a shortfall (amber) means some parameters stay unobservable until you add more varied poses.
Generalization	How well the result is expected to hold on poses outside the sample set rather than overfitting the ones you captured. Read it together with observability: strong observability plus a low generalization factor is a robust calibration.
Joint coverage	One ring per joint, split into 36 sectors of 10°, showing how much of that joint's rotation the samples actually touched. The percentage in the middle is the share of sectors holding at least one sample — 60% and above reads green, 30–60% amber, below that red. An amber sector holds a single sample, a green one two or more. Whichever ring is lowest tells you which joint to move more; while collecting, a needle marks that joint's current angle, so you can drive it into a sector that has not been sampled yet.