A work record begins at 22:00 on Tuesday and ends at 08:00 on Wednesday. Assume both timestamps use the same reporting zone and no clock change occurs during the interval.
Ten hours have elapsed. Two belong to Tuesday. Eight belong to Wednesday.
Assign the entire duration to the start date and Tuesday receives ten hours while Wednesday receives none. The total still adds up. The daily picture is wrong.
This is an uncomfortable class of defect because a report can pass its simplest check. Nothing is missing from the grand total. The error appears when somebody asks when the recorded time occurred.
Split the coverage, preserve the record
The original interval should remain identifiable. Daily allocation is a calculation over that interval, not a rewrite of what the source supplied.
For our example, the allocation is:
| Reporting day | Interval | Elapsed time |
|---|---|---|
| Tuesday | Tuesday 22:00 to Wednesday 00:00 | 2 hours |
| Wednesday | Wednesday 00:00 to Wednesday 08:00 | 8 hours |
Use half-open intervals: include the start, exclude the end. Written compactly, the first span is [Tuesday 22:00, Wednesday 00:00) and the second is [Wednesday 00:00, Wednesday 08:00).
They meet at the same boundary. No time is lost and the boundary contributes no duration twice.
Ending Tuesday at 23:59:59 does not improve precision. It removes a second. With finer timestamp precision, that shortcut becomes even less defensible. Midnight is the boundary; “almost midnight” is not.
If a record ends exactly at midnight, it needs no extra zero-duration segment on the following day. If it spans several dates, apply the same rule at each intervening local-day boundary.
First decide whose midnight you mean
Midnight is local. An instant can belong to Tuesday in one zone and Wednesday in another.
A daily report therefore needs a declared reporting zone. The analyst’s current location is not a sufficient reason to choose it. Establish how the source timestamps were recorded, then establish which local days the report is intended to describe.
Ghost Observatory’s documented work-time calculation splits admitted work intervals at local-day boundaries in run.timezone. Its time-zone reference explains how zone-free timestamps receive a declared interpretation and how that assumption is reported.
An IANA zone such as Europe/Stockholm describes more than a fixed offset. Its clock rules include transitions. During a daylight-saving transition, a local calendar day can contain 23 or 25 elapsed hours. Counting the change in displayed clock labels as though every day contained exactly 24 hours can produce another wrong answer.
Keep two questions separate: which local date owns a segment, and how much elapsed time the segment contains. The zone determines the first. Correctly interpreted instants determine the second.
Splitting cannot repair missing evidence
A tidy allocation does not make a doubtful input trustworthy.
If the end precedes the start, check the dates, source clocks, zone interpretation and field mapping. Do not turn the duration positive just to keep the row. Observatory documents such reversed intervals as rejected records with a reason.
If an end time is absent, midnight is not a replacement end time. If a record states a duration without establishing where it belongs on the clock, do not place it on a day by convenience. The missing position matters.
And a valid ten-hour interval is still only evidence of what its source records. An open process step does not prove ten hours of continuous human work. A desktop observation is not interchangeable with a reported work entry.
The details are the product
Daily comparisons depend on boundaries that most people never see. An error at midnight can move coverage between dates and distort the pattern someone is trying to understand.
Getting the allocation right does not turn a process-analysis tool into a payroll system, nor does it settle which hours qualify for payment. Those are additional questions with their own rules and evidence.
It does give the next decision a sounder starting point: the right recorded interval, interpreted in an explicit zone, allocated without gaps or invented time.