Answering a supervisory data-quality query
A data-quality query is not an accusation, and treating it as one produces the worst possible response. Supervisors query figures constantly — because a value moved, because it looks inconsistent with another return, or because an automated check flagged it. What distinguishes a firm that handles it well is not that its data was perfect but that it can reconstruct what it filed, quickly, and say plainly whether the figure was right.
1. The five artefacts that answer it
| Artefact | Answers |
|---|---|
| The submitted file and its acknowledgement | What was actually filed, and that it was accepted |
| The pinned extract for that reference date | What the source said at the time |
| The mapping version then in force | How the figure was derived |
| The validation output | What the rules said, and which exceptions were accepted |
| The sign-off record | Who approved it, and on what basis |
2. Scope the question before answering it
The first substantive step is to establish what is actually being asked, because three quite different questions arrive in similar language:
- “Explain this figure.” A derivation question. Answered from the mapping and the extract.
- “Why did this move?” A trend question. Answered by identifying the business change, and it usually has a perfectly good answer.
- “This is inconsistent with X.” A reconciliation question — and the most important one, because the difference may be entirely definitional. Answered with a bridge, not a number.
Answering the third with a corrected figure when the difference was definitional is a common and costly error: it concedes an error that did not exist, and it invites a restatement that was never required.
3. When the figure was actually wrong
Sometimes it was. The response then has three parts, and firms consistently under-deliver the third:
- What the correct figure is, and the corrected periods — noting that cross-period validation compares consecutive reference dates, so a correction to one period usually requires the following ones too.
- Why it was wrong — source, mapping or transformation — stated plainly.
- What has changed so it does not recur. This is the part supervisors weigh most, because it distinguishes a one-off error from a control weakness. A response that corrects the number and says nothing about the control invites the follow-up question.
A correction is built against the framework version applicable to the reference date being corrected, not the current one — which is the practical reason to record the version per reference date rather than as a global setting.
4. What good looks like
Three habits, none of which require better data:
- Answer the question asked. A query about one data point is not an invitation to a general narrative about the reporting programme.
- Distinguish fact from assessment. “The figure is X, derived as follows” is fact. “We believe this is immaterial” is assessment, and should be labelled as such.
- Volunteer the known limitation. If the figure carries a data gap you already knew about, say so. A disclosed limitation is a controlled outcome; the same limitation discovered by the supervisor is a finding.
5. A worked case
Facts: a supervisor notes that a counterparty-sector breakdown in a statistical return is inconsistent with the equivalent breakdown in a financial return for the same period.
What the analysis does: reconciles the two populations before touching the amounts. Part of the difference is definitional — the two frameworks define the population differently, and that part is stable and expected. Part is not.
What the response says: the definitional component, quantified, with the framework provisions that cause it; the residual, identified as a genuine mapping gap on a product added two quarters earlier; the corrected figures for both affected periods; and the change made to the product-launch process so that a new product cannot reach production without a reporting scope assessment.
Why that response works: it separates the expected difference from the defect, which demonstrates that the firm understands its own frameworks — and it closes with a control change rather than a promise.
FAQ
Should two returns covering the same population agree?
Usually not exactly. Frameworks define populations and valuation differently by design, so the right answer to an inconsistency query is an attributed bridge rather than a corrected number.
Which framework version applies to a correction?
The one applicable to the reference date being corrected, not the current one.
What do supervisors weigh most in a response?
What changed so it does not recur. Correcting the figure answers the query; the control change answers the underlying question.
Related: Reconciling returns · Resubmissions and corrections · Reporting governance
