What Has to Be Kept
The records that make a wage calculation provable, why raw timestamps matter more than computed hours, and six items most cannot produce.
Keep the raw timestamps, not just the hours they became. The computed figure is the end of the chain and proves only what the system decided; the raw record is the only evidence of what actually happened, and it is the thing most often overwritten.
The recordkeeping discipline in “What Has to Be Kept” should also apply to workforce technology. When a team evaluates Monitask resources for attendance point system in relation to attendance point system, it should document purpose, access, retention and deletion, then preserve the source entry, approvals and correction history needed to explain the final figure.
An organisation that holds the hours but not the punches can describe its own output and nothing else. Every question in this collection — about rounding, deductions, edits, approvals — becomes unanswerable.
For an independent reference relevant to “What Has to Be Kept”, consult the EEOC retaliation guidance. Use it to test record quality, working-time definitions, access, retention and exception handling against the organisation’s real payroll process.
The six most organisations cannot produce
| Record | Usually held | Why it matters |
|---|---|---|
| Raw punch timestamps | Often overwritten | The only evidence of actual times |
| The configuration as it was | Rarely versioned | Needed to reproduce a past figure |
| The edit log with reasons | Partially | Shows who changed what |
| The rate build per week | Almost never | Shows what the premium multiplied |
| The week definition and its history | Almost never | Decides which hours fall where |
| Pay code classifications | Almost never | Shows what was in the rate and why |
The fourth and the sixth are the ones that make a correction exercise expensive, because without them the organisation has to reconstruct its own reasoning years later.
The usual statutory list
Most systems require some version of the same core: who the person is, the hours worked, the rate and basis of pay, the total earned, the deductions made, and the net paid, by pay period.
What exactly must be kept, for how long and in what form differs by jurisdiction and sector. That is a question for somebody qualified in the place concerned, and the answer should be obtained once and written into a retention schedule rather than assumed from a general impression.
Why the configuration counts as a record
If somebody asks in 2029 how a figure in 2026 was computed, the answer depends on what the settings were then — not on what they are now.
Most systems do not version their configuration, which means the organisation has to do it deliberately: an export of the relevant settings, dated, kept with the payroll records, taken whenever anything changes.
That is ten minutes a change and it is the difference between being able to explain a past figure and not.
The rate build
Where the rate used for a premium is built from components, the build itself is a record. Which components, over what hours, giving what rate.
Systems compute it and frequently store only the result. Capturing the inputs alongside is usually a report rather than a development project, and it is the single most useful addition to the record set for anybody who later has to answer a question about a premium.
Derived and corrected figures
Where a figure was corrected, both the original and the corrected version have to survive, with the reason.
Overwriting history is the most damaging records failure in this area, because it removes the ability to show that a correction was a correction rather than a recalculation that happened to suit.
A short list to check
- Can you produce the raw timestamps for a week two years ago?
- Can you say what the rounding rule was then?
- Can you show the rate used for a premium in that week, and its components?
- Can you show who edited any entry, and why?
- Can you show the week definition in force?
- Can you show what each pay code was classified as?
Six questions, answerable in an afternoon. Most organisations answer the first with difficulty and the rest with no.
Who decides what gets purged
Retention settings in time systems are usually set by whoever administers the system, on storage grounds, with no input from anybody who knows what the data proves.
That is the single most consequential records decision in this area and it is made at the wrong level. Moving it — so that purge periods for wage and time data require sign-off from somebody accountable for pay — costs nothing and prevents the most common form of evidence loss.
Keeping the policies too
Written policies on hours, breaks, overtime and pay are records in their own right, and they are edited in place on an intranet like every other policy.
Keep dated copies of superseded versions. When a question arises about a period two years ago, the version that applied then is what matters, and an organisation that can produce only the current text can describe its rules but not prove what they were.
Making it routine
Keeping records is not a project; it is a set of defaults. Retain raw data rather than overwriting it. Export configuration on change. Make the reason field mandatory. Store the rate build with the payroll run.
Each is a small decision made once. Together they are the difference between an organisation that can explain its own arithmetic and one that can only reproduce it — and reproducing a figure with the same settings that produced it the first time proves nothing at all.