About this record504,153 readings across 3 worksheets. Source timestamps are displayed as recorded; no timezone was supplied. The final reading is on 1 July 2026.
Units are not specified in the workbook. The average is a sample mean, not a time-weighted mean. Lines break where readings are more than 1 hour apart; gaps do not imply zero flow.
For the cleaned series, every reading before 2 January 2026 is set to zero because this is a confirmed no-flow period. After that date, cleaning targets suspected isolated spikes only. Original values remain available through the series switch.
Cleaning method & source details
P4_flow_20251201-20260630.xlsx
realflex6/NTE1/MUKWUX_FLOW/A/AIPS0LMP4F1
Confirmed rule: all readings before 2 January 2026 00:00:00 are set to zero in the cleaned series. The cutoff uses the source timestamps as recorded.
After the cutoff, each spike candidate is compared with five original readings on either side. It must differ from their median by at least 100 flow units and 8 times their scaled median absolute deviation. All ten neighbors must lie within 30 units of their median, with a range of at most 40 on each side. Every interval must be positive and no longer than 10 minutes, and the next reading must return within 10 seconds.
Flagged spike values are replaced by linear interpolation in time between their immediate original neighbors. One pass is applied; sustained changes are not smoothed. These are suspected spikes, not confirmed sensor faults.
All timestamps retained, including 114 repeated timestamps. 64 gaps exceed 1 hour. Dense views retain local minima and maxima for display; zoom in to inspect individual readings. Statistics use all values in the selected series and period.