How Far Back It Goes

Establish when the error started before asking how far back it has to be corrected. The first is a question about the organisation's own systems and it can be answered definitively; the second depends on rules that differ by jurisdiction and is a question for somebody qualified in the place concerned.

The workflow in “How Far Back It Goes” becomes more reliable when captured time, approvals and later changes can be followed separately. For teams exploring how to handle multiple clients, monitask.com can provide useful operating context, while payroll rules, employee explanations and final decisions remain with accountable people.

Organisations reverse the order, ask the legal question first, receive an answer in the form of a period, and correct only that period — without ever establishing whether the error is older than it.

For an independent reference relevant to “How Far Back It Goes”, consult the Atlassian project-management guide. Use it to test record quality, working-time definitions, access, retention and exception handling against the organisation’s real payroll process.

The three dates

When the error started: the date the setting was configured, the scheme was introduced or the rule changed.

When it stopped: the date it was corrected, which should be known precisely.

What period must be corrected: a question of rule, of contract, and sometimes of what the organisation decides to do beyond what is required.

The first two are facts in the organisation's possession. The third is advice.

Finding the start

• Configuration change logs in the payroll or time system.
• Change tickets, implementation documents and project records.
• The date a pay code was created.
• The date an agreement or scheme took effect.
• Payroll data itself: the first period in which the pattern appears.

The last is the most reliable. Run the detection query backwards through history until it stops returning rows, and the boundary is the start.

When records do not go back far enough

Systems get replaced, and data from a predecessor system is frequently unavailable or in a form nobody can query.

That is a finding in itself and it should be recorded rather than treated as an answer. "The error is present from the earliest data we hold, which begins in March 2021" is a true statement; "the error began in March 2021" is not, and the difference matters if anybody asks later.

Deciding to go further back than required

Organisations frequently correct beyond any required period, and there are good reasons: fairness, consistency between people whose circumstances differ only by timing, and the practical difficulty of explaining to one employee why the person next to them was paid more.

That is a decision rather than an obligation, and it should be made deliberately, recorded as a decision, and applied consistently. What reads badly is correcting further back for the people who complained and not for the ones who did not.

Former employees

An error affecting people who have since left raises the same question with an extra step: finding them.

1. Identify affected former employees from the same query.
2. Compute what each is owed on the same basis as current staff.
3. Find current contact details, using the last known address and any later correspondence.
4. Write, explaining what happened and what is owed.
5. Keep a record of attempts where somebody cannot be found.
6. Decide what happens to unclaimed amounts, and record it.

Step five is the one that matters if the question ever arises. An organisation that tried and recorded the attempts is in a different position from one that quietly dropped the leavers from the list.

The cost of a long tail

Each additional year roughly adds its own proportion of the annual amount, plus the administrative cost of locating and correcting it, which rises sharply as the period lengthens.

That is an argument for finding errors early rather than for limiting the correction. An error found in its first quarter costs a quarter; the same error found after four years costs four years and is harder to explain.

Errors that started and stopped

Not every error runs continuously. A setting can be wrong, corrected by accident during an upgrade, and reintroduced later — which produces two separate periods with a clean stretch between them.

Running the detection query across the whole history rather than searching for a single start date catches that. An exercise that finds the most recent start and stops there will correct one period and leave an older one, which is the version that is discovered later by somebody else.

Changes in the rule itself

Where the underlying requirement changed — a threshold moved, a definition widened — the correct figure differs between periods, and the model has to reflect that rather than applying today's rule backwards.

Applying a current rule to a historical period overstates or understates what was owed, and in either direction it produces a number the organisation cannot defend. Establishing what applied when is part of the advice, and it should be asked for explicitly rather than assumed.

The five lines to keep

The start date and the evidence for it, the stop date, the period corrected, the basis for that period, and anything the organisation chose to do beyond it.

Five lines, kept with the correction records. That document is what makes the exercise explicable years later, and it is also what prevents the organisation from quietly correcting a shorter period the second time something similar happens.