A data quality issue register connects a failed rule to the affected record IDs, a count and a named next action without changing the source. Use the small exercise below to create a reviewable exception list, explain its priorities and save a record that another person can check.
Content reviewed: 21 October 2026. The four records are an explicitly hypothetical author-created worksheet, not customer data or a reported consulting engagement. The exercise checks stipulated rules. It does not establish that the underlying transactions are true, fix a workbook or approve a financial report.
What should a data quality issue register establish?
Imagine receiving a small dispatch export with a missing quantity, a negative quantity and an absent amount. Editing the empty cells until the sheet looks complete would conceal the questions you actually need answered. Your first useful output is a list of the exceptions, their implications and the person who should investigate each one.
A register is simply a structured issue list. It separates an observation from a correction. “Quantity is missing in R2” is an observation against a rule. “Enter two units” is a proposed source change requiring evidence. The first does not authorise the second, even if the suggested replacement looks plausible.
The UK Government Data Quality Framework distinguishes dimensions such as completeness, validity and accuracy. Those distinctions help us avoid calling a present, correctly formatted value necessarily true. We use the framework as conceptual support, not as a legal requirement for every business or a certification for this worksheet.
Keep this review separate from choosing products or interpreting sales movements. The AI data analysis guide addresses comparisons and business reporting. Here, the decision is narrower: which data questions need investigation before someone relies on the records for the stated purpose?
Which rules must be explicit before counting anything?
CSV is a text format for rows and fields. Our starting point is an already correctly parsed file with stable record identifiers. We are not demonstrating delimiter repair, spreadsheet import settings or text conversion. If field boundaries are uncertain, settle that first using the separate CSV quoted-comma audit.
Write the purpose at the top of a rule card. In this invented scenario, the team is preparing dispatch exceptions before a later financial reconciliation. The local policy puts quantity questions first. Another organisation could require an amount check before dispatch; the example is not a universal operating policy.
- record_id: the unchanged identifier used to trace each observation back to its source record.
- quantity: required; a supplied value must be an integer of at least one.
- amount: required in this executed worksheet; numeric and non-negative validation is a separate check outside this run.
- Missing-value treatment: an empty field fails the required-value rule and is excluded from the corresponding value-range evaluation.
That final rule prevents an absent quantity being counted again as a quantity below one. It also avoids quietly turning absence into zero. State the treatment before inspecting the file rather than adjusting it afterwards to get a convenient total.
These rules depend on meaning. A negative quantity could legitimately represent a return in another export. If your real source includes returns, ask the responsible person which records belong in this dispatch review. Do not label every negative value an entry error merely because the practice rule excludes it.
What do the four records actually show?
Copy the following hypothetical source exactly. The header names the fields and is not an additional business record. Preserve the empty quantity in R2, the negative sign in R3 and the empty amount in R4.
record_id,quantity,amount R1,2,20 R2,,15 R3,-1,10 R4,1,
| Rule checked | Failed record IDs | Failure count | Review observation |
|---|---|---|---|
| Quantity required | R2 | 1 | No quantity was supplied |
| Quantity integer and at least one | R3 | 1 | The supplied -1 fails this rule |
| Amount required | R4 | 1 | No amount was supplied |
The recorded amount check covers missing values only; this run does not establish a general numeric or non-negative amount validator. R1 produces no issue under the rules supplied. That is a limited finding, not a declaration that R1 is a genuine transaction or that every possible check has passed.
The expected result is three rule failures across three distinct affected records. Keep both numbers. In a different file, one record might fail two independent rules, so a failure total would exceed the number of affected records. Record IDs let a reviewer distinguish those situations instead of relying on a headline count.
Identifiers must survive the review intact. A product code that loses a leading zero may no longer identify the same item. The leading-zero preservation check covers that different prerequisite. This exercise assumes the four R identifiers have already retained their meaning and spelling.
How do you justify priority without inventing a scoring system?
Frequency and urgency answer different questions. A single missing quantity may block a picking instruction, while several secondary omissions might not prevent the next operational step. Explain the intended use and the consequence before assigning a label such as high or medium.
The official Data quality issues framework relates prioritisation to purpose and impact. Our worksheet borrows that broad principle. It does not reproduce the framework's numerical model or claim that our labels are government-assigned ratings.
For this exercise, quantity-required and quantity-range issues receive high priority because the local scenario requires a reliable picking instruction first. The missing amount receives medium priority for the subsequent reconciliation. Both labels need their reasons beside them. Without the scenario, the ranking would be an unsupported opinion.
Give the quantity issues to a dispatch lead role and the amount issue to a finance reviewer role. These are fictional responsibilities for the exercise, not named people who have accepted assignments. In actual work, confirm the owner. Until then, write “owner unassigned” rather than supplying a name merely to make the register appear finished.
What fields make the register actionable?
Use one entry for each column-and-rule combination. Include the IDs, count, consequence, priority reason and next action. Add the source reference and rule-card version at the document level so a reviewer knows what was assessed. A different rule version can produce a different result without either run being arithmetically wrong.
Register: data-quality-issues-v1 Source reference and rule-card version: [record these] Column: quantity Rule: value required Failed record IDs: R2 Failure count: 1 Priority: high within this dispatch exercise Reason: missing quantity blocks a reliable picking instruction Next action: ask the source owner for the recorded quantity Owner role: dispatch lead State: investigation needed; no source correction authorised
Create the second entry for quantity, the integer-at-least-one rule and R3. Its action is to ask whether the negative value represents a return or an entry error. Do not write “remove the minus sign.” The third entry concerns the required amount and R4; ask the finance reviewer to locate the source document rather than estimate a price.
An assistant is optional. If you use one, provide only information approved for that service and ask for rule failures, IDs and counts without source edits. Require it to exclude missing values from the range check and keep observations separate from suggested actions. The Excel prompting templates explain how to describe inputs and checking expectations; this exercise does not require an Excel integration.
Compare any assistant response with the four original records yourself. A confident explanation is not an independent source document. If a response includes a replacement quantity or an invented owner, remove that proposed commitment from the confirmed record and leave the underlying question open.
What can you verify after saving and reopening?
On Windows, open Notepad from Start and create a document. On Mac, open Finder → Applications → TextEdit, choose File → New and, if needed, Format → Make Plain Text. Copy the CSV sample. Use File → Save As on Windows or File → Save on Mac to save source-fixture.txt in Documents, then close it. This separate copy preserves CSV text; it is not an import test. Add a date if the filename exists. Choose File → New and enter the rule card and three register entries.
Choose File → Save As on Windows or File → Save on Mac. Save data-quality-issues-v1.txt in Documents, adding a date rather than overwriting an existing file. Close it, then reopen that exact file from Documents in File Explorer or Finder. Check R2 under required quantity, R3 under the value rule and R4 under required amount: three failures, three IDs. Open the source file under the actual name you saved, including the date, and compare the empty fields and negative sign. Never replace the source with the register.
For the technical preparation behind this article, the fixture was actually checked in a Linux terminal using Python 3.11.2 and its standard csv, json and hashlib modules. The observed IDs matched R2, R3 and R4. JSON and text versions of the register were written, closed, reopened and compared with their expected content. The source records were also read again.
The SHA-256 value of the source was identical before and after that run. A hash is a digital fingerprint calculated from file bytes. Matching hashes support the finding that this source was unchanged during the check; they do not establish business accuracy. This was a Python command-line check, not an observed Notepad, Excel or AI-provider test.
When is the first review complete?
Allow roughly fifteen minutes to read the rules, identify the exceptions, assign provisional next actions and inspect the saved note. This is a planning allowance, not a measured completion time. The useful result is a traceable investigation list, even though none of the source questions has yet been resolved.
Hand the document to the people responsible for those questions. A later correction should retain its evidence and approval separately from the original observation. If the team changes the rule instead, preserve why it changed; a team decision log can hold that history. Do not silently overwrite the old review and call the issue solved.
Frequently asked questions
Does a missing value count as zero?
No. In this exercise it fails the required-value rule. Replacing it with zero would alter the source and change the meaning of the check.
Should R2 also fail the quantity range rule?
No. The stipulated rule card excludes missing quantities from range evaluation. R2 is recorded only under required quantity.
Does high priority authorise a correction?
No. Priority determines investigation order. A source change still needs supporting evidence and appropriate authority.
What should the saved exercise contain?
Keep the unchanged source, the rule card and a register containing failed IDs, counts, priority reasons, actions and owner roles.
