Drift metrics

To measure how much a sensor has drifted, Open Gauge compares two of its calibration curves — a reference calibration (typically the baseline/first one) and a current one — by evaluating both fitted polynomials across their shared operating range and looking at the difference curve between them, $d(x) = f_{\text{current}}(x) - f_{\text{reference}}(x)$.

Max drift

The largest absolute difference between the reference and current curves across the shared operating range:

$$\text{max drift} = \max_x \left| d(x) \right|$$

The single biggest weight in the health score (30%) — it answers "at its worst point, how far has this sensor's calibration moved?"

Mean drift

The average absolute difference between the two curves across the shared range:

$$\text{mean drift} = \frac{1}{n}\sum_{i=1}^n \left| d(x_i) \right|$$

RMS drift

The root-mean-square difference between the two curves — penalizes larger deviations more heavily than mean drift does, the same way RMSE penalizes large residuals more than a simple average would:

$$\text{RMS drift} = \sqrt{\frac{1}{n}\sum_{i=1}^n d(x_i)^2}$$

Offset

The difference between the curves at the start of the shared range — a constant (zero-point) shift. A sensor that has drifted by a roughly constant amount across its whole range, rather than growing worse toward one end, shows up mostly as offset.

Gain

The slope of the difference curve $d(x)$ — a change in sensitivity (span) between the two calibrations, as opposed to a constant offset. A sensor whose span has stretched or compressed (e.g. from a reference-junction or amplifier gain shift) shows up mostly as gain, not offset.

Residual drift

The difference between the curves at the end of the range minus the difference at the start — i.e. drift that isn't explained by a constant offset alone. Combined with gain, this separates "the whole curve shifted uniformly" from "the curve tilted."