Same reference, different reading
Imagine a scale checked with the same 100 gram reference under comparable conditions. Its readings are 100.02 g in January, 100.05 g in March, and 100.08 g in May. Relative to the reference, the signed deviations are +0.02, +0.05, and +0.08 g. If the limit is ±0.10 g, all three results are within tolerance on those numbers alone. Yet the deviations are moving toward the high side.
Drift describes a change in the instrument’s behavior, not a change in what you are measuring. Wear, component aging, mechanical strain, and repeated use can contribute. A sudden step after a drop may be damage rather than gradual drift, but it is still a reason to inspect the instrument.
Separate drift from noise or a changed setup
Not every sequence of larger readings proves drift. A warmer room, different fixture, new operator, or suspect reference can shift the results. Repeatability also means readings scatter. One unusual result deserves examination, not an automatic trend line. Compare measurements at the same point in the range, with the same units and comparable procedures. Keep notes on conditions and any interventions.
Avoid joining a series straight through a repair or adjustment as if nothing happened. A reset can change the baseline. Keep both records so another person can reconstruct the history, but review the observations before and after the intervention separately when the behavior clearly changes.
Use readings taken before adjustment
As-found results tell you how the instrument performed when it arrived for calibration. As-left results tell you how it performed after service or adjustment. If you only keep the final values near zero, you can erase the signal that helps explain what happened between calibrations. Ask your provider for both when relevant and keep the date, measurement point, and applied limit with each result.
Write the deviation as a signed number so an increase and a decrease do not look alike. Do not average away a problem at one important measurement point by mixing it with a point that happens to be stable. Investigate separate locations or ranges separately.
What a trend prediction can and cannot say
A simple method fits a straight line to deviations by date and asks when that line reaches a tolerance limit. With the scale above, equal 60-day spacing and a rise of 0.03 g per interval would leave roughly 40 days from the last 0.08 g reading to the upper 0.10 g limit. The date is conditional on the slope continuing; it is not a test result or a guarantee.
In the free interval calculator, enter the three dated deviations and a positive limit of 0.10 for each row. It expresses readings as a fraction of the limit so it can estimate the direction. It assumes symmetric limits and does not account for measurement uncertainty, nonlinear behavior, or changing shop conditions.
Use the estimate to trigger a review, not to move a date
If the predicted crossing is close to the normal interval, you might review sooner, add an intermediate check, or inspect the instrument for wear. Assess which products or tests depend on it and what happens if it reads incorrectly. Consider the cost of an extra check against the cost of accepting bad measurements. A forecast can help you decide where to look first.
A stable history may justify evaluating a modest extension, but not if a customer, manufacturer, contract, or procedure requires a shorter maximum. Use enough representative observations, document your rationale and authorization, and verify the revised interval in service. The calculator’s suggestion is not an approval.
Respond to an actual limit breach now
If an as-found reading exceeds the applicable limit, you no longer need a prediction of when it might happen. Follow your procedure for taking the instrument out of service or restricting use and evaluating prior results. Establish the last evidence that it was suitable and identify work that could be affected. Keep the actual readings and the disposition together.
Then review why the interval or intermediate checks did not detect the problem sooner. The answer might be a shorter interval, a change in handling, different equipment, or a better reference check. CaliberOS offers a public drift calculator as a simplified demo; predictive drift is not available in the current product.
References: NIST: Recommended calibration interval · ILAC G24 / OIML D 10: Determining recalibration intervals.