Report Measurement Uncertainty in a UPS Runtime Result
By Highidea Power · 23 October 2026 · 3 min read
A runtime result needs more than a stopwatch value. Define the start and endpoint, describe how they were observed and report the limitations supported by that method. Display resolution does not establish measurement accuracy. The acceptance authority must also agree how a result near the required duration will be judged; that decision rule is separate from the measurement itself.
Define the elapsed time being reported
“Runtime” can end at a specified application state, a defined electrical threshold, an approved shutdown or another contractual endpoint. Those events need not occur together. Identify the exact quantity before assigning uncertainty to it.
Use one time basis where possible. If different systems record the start and end, their clock relationship becomes relevant. A logger interval, an operator's observation and an event message can introduce different limitations. More decimal places in a spreadsheet do not remove them.
NIST TN 1297 asks for uncertainty components, their evaluation and the basis of any coverage or probability interpretation. That reporting discipline is useful here. Its examples do not impose one coverage factor or one contractual acceptance rule on every UPS test.
Distinguish complete but limited observations from missing evidence
A complete record can still have finite resolution. For example, periodic observations might locate an event within an interval without identifying its exact instant. A lost record presents another question: whether the event was observed at all. Keep the two findings separate.
Contribution | What the report should identify |
|---|---|
Event definition | Physical or application event, required threshold and the observation that represents it. |
Time measurement | Instrument or clock, specification, resolution and applicable calibration evidence. |
Event observation | Sampling interval, detection delay and justified bounds or model. |
Separate clocks | Offset, synchronization evidence and uncertainty where relevant. |
Result model | How contributions enter the elapsed-time result and what assumptions remain. |
Do not automatically add every number as though it were independent, or divide every bound by the same statistical factor. A competent measurement reviewer should select and explain the model, including correlations where they matter.
Work a bounded observation example
Assume, for illustration only, that the start occurred between 0 and 2 seconds on one common reference, and the endpoint between 1,798 and 1,802 seconds. Assume those intervals completely contain the respective events and ignore all other error sources for this exercise.
The shortest elapsed time is 1,798 − 2 = 1,796 seconds. The longest is 1,802 − 0 = 1,802 seconds. The supported interval is therefore 1,796–1,802 seconds, or approximately 29.93–30.03 minutes. Reporting exactly 30.000 minutes would hide the observation limits.
This is interval arithmetic, not a statistical confidence interval or a Highidea test result. It does not justify attaching a confidence percentage. If an event was not actually bounded by valid observations, the calculation cannot manufacture that evidence.
Keep the pass decision explicit
Against a hypothetical requirement of at least 1,800 seconds, the example spans the threshold. It cannot establish a pass under a rule requiring the entire supported interval to meet or exceed the threshold. A different contract may use a different agreed rule; document it before viewing the result rather than choosing a favorable rule afterward.
Preserve measured values, corrections and their bases, the uncertainty expression, the chosen decision rule, the conclusion and the responsible approver. A measurement report may be technically complete while contractual disposition remains open.
Attach the result to the actual configuration
Include host/battery identity, load, starting condition, operating mode, environment, endpoint and authorized procedure reference. Send Highidea these records through our contact page when discussing a configuration result. The project's acceptance authority retains the contractual decision.
References
Related reading
FAQ
Frequently asked questions
Does a one-second display mean one-second accuracy?
No. Resolution is one attribute; timing accuracy, event detection and the complete method still need evaluation.
Is expanded uncertainty always a fixed multiple?
Use a justified model and report the selected factor and interpretation where applicable. Do not import a factor from an unrelated example.
Can uncertainty explain away an unrecorded endpoint?
No. Mark missing event evidence explicitly and use the surviving-record review or an approved retest to resolve it.
Put this into practice
Need help specifying the right unit?
Share the load, voltage, runtime, environment and destination for a configuration review.