Measurement History¶

The measurement history shows, per measuring point, how the deviation of a device developed across several calibrations — the question behind every interval discussion: is this device drifting, and how fast.

Access¶

On the device detail page, the Calibrations tab carries a switch between List and Measurement history. It requires the "view measurement history and trend" permission (measurement_history_view). Until 09/2026 calibration_view was enough; it is a permission of its own since, so that the history can be switched off without hiding the calibrations themselves. Without it the switch is not offered at all.

The measuring points are selected on the left, the history curve of the selected point is shown on the right, with status and trend statement in plain words below it.

What the chart shows¶

Element	Meaning
Points	The deviation per calibration, plotted over time
Tolerance band	The limits that applied to this measuring point
Trend line	Only where a slope is statistically demonstrable

The series is interrupted at two kinds of event instead of fitting a line across the break:

Adjustment. After an intervention on the device, the earlier slope no longer describes the same device. A routine adjustment that was already within tolerance beforehand does not count as a break.
Tolerance change. If the band changes, the deviation before and after is no longer the same quantity.

When a statement is made — and when not¶

The trend line is only drawn if the slope passes a significance test. If it fails, the page says so instead of showing a line that looks like a statement. The status per measuring point knows four values:

Status	Meaning
Evaluable	Enough points, unambiguous matching, trend statement possible
Too few points for a trend	Fewer than four calibrations at this point
Ambiguous point matching	It is not clear which measurement row corresponds to which point
Resolution-limited	The values jump in steps of the display resolution; a trend would describe the display, not the device

No statement is a statement

The three refusals are the point of this view, not its shortcoming. A drift line through three points, or through quantised values, would be a number with the appearance of precision — and such a number turns into a calibration interval at the next step.

Where no slope is demonstrable, an upper bound is stated instead of a rate: "with 95 % confidence smaller than X per year". That is the statement the data actually supports.

Rows not taken into account¶

Rows that contribute nothing to a history are excluded and counted — visibly, not silently:

Reason	What is missing
No date	The calibration carries no date, so the point has no position in time
No measured value	The row was left empty
No set point	Without a reference there is no deviation
Unit mismatch	The row measures something other than the rest of the series

Limits¶

The evaluation describes the past. A forecast ("when is the device expected to leave tolerance") is deliberately not part of it yet.

How many devices in your own stock carry enough data for an evaluation at all is answered by the drift:coverage command:

docker exec calserver-api-v2 php artisan drift:coverage