One run in a monthly export may sit far from the others. Do not remove it just to tidy the table: it may reflect a measurement, matching or clock issue, or be a real observation that needs review.
Three checks
- Device: serial number, logging interval, battery state and sensor position. 2. Event: delay or fault noted on the run sheet. 3. Matching: whether the file belongs to that run and which clock information was used.
Record the treatment
If there is a reason to exclude a value from analysis, record the row or time window, evidence, authorizing role and date. Leave the raw file unchanged; make the action traceable in the analysis copy.
If the value is unexplained
A value review alone does not decide whether product is released or rejected. Assess the discrepancy against product specifications, validated route conditions and the authorized quality process.
Two calculations
Calculate the monthly measure both with and without the value to show its effect, and report the scenarios side by side. This comparison shows the impact of the choice; it does not establish whether the value is valid.
Repeated values. If similar outliers recur, compare device, position and event records using the same criteria. Recurrence alone proves neither a fault nor a cause; state which records were reviewed.
Leave a decision trail
Link the report to the raw filename, device and run identifiers, time window, review note and authorized decision record. A later reader can then reconstruct why the value was included or shown separately.
