# Data Quality Dimensions for Business Reports: Separate Complete, Valid, and Accurate

<a id="main-content"></a>

Operations &amp; Systems

October 28, 2026  ·  8 min read  ·  Victor Valentine Romo

On this page

1. [Name the Use Before Choosing the Checks](<#name-the-use-before-choosing-the-checks>)
2. [Keep the Six Dimensions Separate](<#keep-the-six-dimensions-separate>)
3. [Work Through Complete and Valid Records That Are Wrong](<#work-through-complete-and-valid-records-that-are-wrong>)
4. [Specify the Population and Reference for Every Result](<#specify-the-population-and-reference-for-every-result>)
5. [Record Timing as a Rule With a Clock](<#record-timing-as-a-rule-with-a-clock>)
6. [Copy the Purpose-Specific Assessment Worksheet](<#copy-the-purpose-specific-assessment-worksheet>)
7. [Show What the Error Changes](<#show-what-the-error-changes>)
8. [Expose Tradeoffs Before Combining Scores](<#expose-tradeoffs-before-combining-scores>)
9. [Sources](<#sources>)

Every required cell contains a value. Every [timestamp](</articles/event-time-processing-time-business-reports/>) passes the format check. One closing time is wrong.

A green validation result can coexist with a false report. Name the property each check establishes before treating the data as ready for a decision.

The [UK Government Data Quality Framework](<https://www.gov.uk/government/publications/the-government-data-quality-framework/the-government-data-quality-framework>) presents six core dimensions drawn from DAMA UK. Its primary focus is input data. It does not replace quality assurance of analytical outputs. Its adoption request addresses central government, while its concepts can inform other contexts.

The worksheet below adapts those dimensions to a fictional request report. The rules, evidence, and tolerances are proposed examples. They describe no customer records or measured business outcome.

<a id="name-the-use-before-choosing-the-checks"></a>

## Name the Use Before Choosing the Checks

The fictional operations owner wants to compare elapsed duration across four completed requests. The report will support a retrospective discussion about those four records.

It will not assign individual performance ratings, issue invoices, or make a live dispatch decision. Those uses would require their own evidence and acceptance conditions.

Define the population: requests with a confirmed closing event on October first, using UTC dates. Define the measurement: closing timestamp minus opening timestamp, expressed in elapsed minutes.

The owner expects one row per request. The report requires a request identifier, opening timestamp, and closing timestamp. Its cutoff occurs after the selected day's events.

Put those definitions in a [business metric dictionary](</articles/business-metric-dictionary-template/>). The quality assessment then asks whether a specific dataset supports that defined use.

[Choose tolerances with the person responsible](</articles/control-limits-targets-service-process-reports/>) for the decision. A developer can implement a check without owning the consequences of accepting its failures.

For this example, the owner requires resolution of any confirmed timestamp error before comparing the four durations. That is a fictional acceptance condition. It is not a universal threshold.

<a id="keep-the-six-dimensions-separate"></a>

## Keep the Six Dimensions Separate

The framework's [dimensions section](<https://www.gov.uk/government/publications/the-government-data-quality-framework/the-government-data-quality-framework>) identifies these six properties:

| Dimension | Question for this assessment |
| --- | --- |
| Completeness | Are expected records and essential values present? |
| Uniqueness | Does each represented request occur once? |
| Consistency | Do related representations avoid contradiction? |
| Timeliness | Is availability and currency suitable for the use? |
| Validity | Do values meet the specified format and range? |
| Accuracy | Do recorded values correspond to reality? |

The framework treats this list as adaptable rather than prescriptive. It also recognizes tradeoffs between dimensions according to user needs. These are separate assessment questions. A pass in one column cannot answer another column's question.

For the proposed worksheet, write each check as a rule with an explicit population and evidence source. Keep results at the level the check supports.

A populated closing field establishes presence. A parser accepting that closing value establishes [compliance with its parsing rule](</articles/ai-structured-outputs-schema-business-validation/>). Neither result verifies when the request closed.

Use [null, blank, and zero distinctions](</articles/null-blank-zero-business-data/>) when deciding what counts as present. Do not fill a missing closing time merely to improve a completeness percentage.

<a id="work-through-complete-and-valid-records-that-are-wrong"></a>

## Work Through Complete and Valid Records That Are Wrong

All dates below are October first, 2026, in UTC. These records and their reference evidence are synthetic.

The fictional reference register lists exactly R1, R2, R3, and R4 as the expected completed requests. The export contains exactly those four identifiers, once each.

| Request | Export opening time | Export closing time | Export duration, minutes | Reference finding |
| --- | --- | --- | --- | --- |
| R1 | 09:00Z | 09:30Z | 30 | Opening and closing events agree |
| R2 | 10:00Z | 11:00Z | 60 | Confirmed closing event is 10:30Z |
| R3 | 12:00Z | 13:30Z | 90 | Closing evidence unavailable |
| R4 | 14:00Z | 16:00Z | 120 | Opening and closing events agree |

The export passes the example's presence check. Four of four expected records appear. All twelve required cells contain values: three required fields across four records.

It also passes the proposed validity rule. Each timestamp is parseable as UTC, occurs on the selected day, and places closing after opening. The identifiers follow the example's agreed pattern.

R2 still contains a confirmed error. The reference packet supplies an accepted closing event at 10:30Z. The export says 11:00Z.

R3 has a different status. Its closing evidence is unavailable. The record may be right or wrong. Record accuracy as unknown for that closing field.

Do not put R3 in the confirmed-error count. Do not count it as verified correct either. The missing evidence is a result that requires its own action.

<a id="specify-the-population-and-reference-for-every-result"></a>

## Specify the Population and Reference for Every Result

The completeness [denominator](</articles/two-way-tables-business-reporting/>) comes from the fictional reference register. Comparing the export against itself would only establish that its own rows exist.

The required-field check uses twelve cells. The record check uses four identifiers. Report both denominators rather than presenting a single unexplained completeness score.

Uniqueness in this exercise means one export row per expected request identifier. Four rows contain four distinct identifiers. That result does not prove two different identifiers could never describe the same underlying request.

The example supplies the identity boundary through its reference register. A real assessment needs evidence for that boundary. [Keep identity resolution outside](</articles/unicode-normalization-business-record-matching/>) a format-validation claim.

For consistency, compare each export's closing timestamp with a second synthetic reporting copy. Suppose both copies contain 11:00Z for R2. Agreement across four of four corresponding closing fields would pass that narrow check.

The two copies can agree because [one copied the other](</articles/data-lineage-small-business-reports/>). Their agreement does not overturn R2's conflicting event evidence.

For accuracy, define the assessed field: closing timestamp. Three closing fields have accepted reference evidence. Two agree and one disagrees. One additional field value remains unverified.

Write “two verified agreements, one confirmed discrepancy, one unknown across four closing values.” That statement preserves the evidence boundary.

If a summary requires a verified-subset rate, two divided by three is about sixty-six point seven percent. It describes only the three checked values. Keep the fourth value's unknown status beside it.

<a id="record-timing-as-a-rule-with-a-clock"></a>

## Record Timing as a Rule With a Clock

Timeliness needs a stated reference time. A file's recent modification date does not establish how current its underlying events are.

In this example, the reporting cutoff is midnight after October first. The export becomes available at 08:00Z on October second. The fictional retrospective review starts at 09:00Z.

The owner accepts availability before that review, provided the export represents events through the cutoff. The example states that its four-record reference register is complete through that time.

Under those assumptions, the export meets the availability condition. The eight-hour interval between cutoff and availability remains visible.

That same timing would fail a hypothetical dispatch requirement for updates within fifteen minutes. The retrospective result cannot certify the export for live dispatch.

Keep event time, extraction time, and availability time distinct. If the source cannot establish its coverage cutoff, mark currency unknown even when the file arrived on schedule.

[Record the clock and timezone](</articles/automation-time-zones-daylight-saving/>) used by the rule. Do not compare a local review deadline with a UTC extraction time without an explicit conversion.

<a id="copy-the-purpose-specific-assessment-worksheet"></a>

## Copy the Purpose-Specific Assessment Worksheet

Use one row per check. Some dimensions need several rows because their populations or evidence differ.

This is a proposed worksheet, not a government certification form. “Report owner” is a fictional role that must become a named responsibility in an actual assessment.

| Use | Dimension | Rule | Population | Evidence | Result | Unknowns | Tolerance owner | Action |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Four-request retrospective | Completeness | Include expected requests | Four register identifiers | Reference register and export | 4 of 4 present | None within register boundary | Report owner | Retain reconciliation |
| Same use | Completeness | Populate three required fields | Twelve cells | Export field inspection | 12 of 12 populated | Presence does not verify values | Report owner | Preserve separate accuracy findings |
| Same use | Uniqueness | One row per expected identifier | Four export rows | Identifier counts | Four distinct identifiers | Identity outside supplied register unassessed | Report owner | Keep identity boundary stated |
| Same use | Consistency | Closing values agree between copies | Four matched closing fields | Export and second reporting copy | 4 of 4 agree | Copies may share an error | Report owner | Retain R2 discrepancy |
| Same use | Timeliness | Available before 09:00Z review with stated cutoff | One export | Availability and cutoff record | Available 08:00Z; eight-hour lag | Upstream capture behavior outside example | Report owner | Limit use to retrospective |
| Same use | Validity | Timestamps parse and closing follows opening | Four timestamp pairs | Explicit format and order rule | 4 of 4 pass | Actual closing time not established | Report owner | Keep accuracy review separate |
| Same use | Accuracy | Closing value agrees with accepted event evidence | Four closing values | Three reference packets | Two agree; one differs; one unknown | R3 closing evidence absent | Report owner | Resolve R2 and obtain R3 evidence |

Attach the rule version, assessment date, reviewer, and evidence location above the worksheet. [Preserve the assessed export version](</articles/spreadsheet-version-control-business-decisions/>).

When an unknown becomes resolved, record the new evidence and a new assessment outcome. [Keep the earlier finding available](</articles/late-arriving-data-report-revisions/>) for anyone reviewing the original report.

The worksheet directs attention toward the next action. It should let a reader distinguish a missing record, an invalid timestamp, and a timestamp contradicted by evidence.

<a id="show-what-the-error-changes"></a>

## Show What the Error Changes

The exported durations sum to three hundred minutes. Dividing by four produces [a mean of seventy-five minutes](</articles/mean-median-business-reports/>).

Correcting R2's closing time changes its duration from sixty to thirty minutes. Holding the other export values unchanged produces two hundred seventy total minutes and a mean of sixty-seven point five.

That recalculation isolates the known discrepancy. It does not verify R3's ninety minutes. Label the revised mean provisional while that evidence remains unavailable.

Do not silently remove R3 and retain the same population description. A three-request subset would represent a different analysis, with its own denominator and selection limitation. Verified closing times alone would not establish the accuracy of every field in those records.

If the owner elects to review those three records separately, record that decision. State which request remains outside the verified discussion and why.

The numerical change shows why validation alone cannot establish fitness for this use. Every export timestamp passed the specified format and ordering checks before the error was corrected.

<a id="expose-tradeoffs-before-combining-scores"></a>

## Expose Tradeoffs Before Combining Scores

The owner could publish a provisional retrospective before R3's evidence arrives. That choice prioritizes availability over a fully verified closing-time assessment.

Record the unresolved value beside the output. Identify the decision owner and the permitted use. A conditional release should not become blanket acceptance for unrelated decisions.

[Avoid averaging dimension percentages](</articles/weighted-averages-business-reports/>) by default. Completeness counts cells or records. Timeliness may assess one delivery. Accuracy here includes an unknown. Those results have different populations and meanings.

If the business requires a combined score, document the formula, weights, treatment of unknowns, and blocking conditions. Display the underlying findings beside the score.

An equal weighting is still a weighting decision. Explain why that choice suits the stated use. A high average should not conceal an unresolved condition the owner considers blocking.

Keep remediation distinct from assessment. [CRM data quality workflows](</articles/crm-data-quality-automation/>) address validation and cleanup. This worksheet records which checks support a particular reporting decision.

Name the use. Preserve the evidence. Leave unknowns visible.

<a id="sources"></a>

## Sources

- [UK Government: The Government Data Quality Framework](<https://www.gov.uk/government/publications/the-government-data-quality-framework/the-government-data-quality-framework>). Published 2020.12.03. Scope, six dimensions, adaptability, and purpose-dependent tradeoffs. Checked 2026.10.28. The worksheet, records, and acceptance rules are synthetic editorial adaptations.
