Skip to main content

Data Quality

Hold incomplete order records before they distort the monthly report.

For reporting, data and operational teams that need to decide whether a scheduled report can be issued, limited or held.

Quanta Meridian shows which order or delivery row failed, which service or supplier result would be wrong, who must correct the customer or product record and what the independent retest found.

When data needs a controlled response

The problem is not only that a value failed a check.

A delivered order with no customer code cannot be assigned to the customer group used in the monthly service report. An unknown product code can remove a complete supplier category from the monthly score.

A count of failed references is not enough. The reporting analyst needs the affected order keys, the result that would be wrong and the data owner who can correct the customer or product table.

The row remains held until the same rule runs again. The original failure, corrected customer or product record, independent retest and release decision stay connected for the reporting owner.

Validation field

Keep the failed record connected to the reporting decision.

Rules test the batch in order. Material failures retain their record keys, affected report and owner, then return through the same rule for retest.

INITIAL VALIDATION RUNOrder and delivery records
Rows
550,000
Rules
15
  1. G1PASS

    Schema

    Columns · types · keys

  2. G2WARNING

    Data

    Required · unique · timely

  3. G3FAIL

    Business

    Sequence · references · totals

REPORTING DECISIONHold monthly service report

Release only after all 15 checks pass

FAILED RECORDS244 rows failed; 238 orders held

Record key · rule · value · expected condition

AFFECTED REPORTMonthly service report

Customer, product and on-time delivery results

OWNER RESPONSECorrect · accept · escalate

Evidence and target retest remain in history

RETESTSame rule, retained result
01 · Input and run metadata550,000 order, line and delivery records
02 · PASSSchema rules

Columns · types · keys

03 · WARNINGData rules

Required · unique · timely

04 · FAILBusiness rules

Sequence · references · totals

05 · Failed-record branchRetain records and assess report impact

Assign an owner only when the threshold or impact warrants it.

06 · Resolve and retestRun the same rule and retain both results
07 · Reporting decisionHold until all 15 checks pass
PassThe rule met its threshold.
WarningThe result needs review but does not automatically block.
FailThe rule breached its agreed threshold.
ErrorThe check could not complete, so no quality conclusion is made.

Result states

State what happened before deciding what to do.

Status is written as text and paired with its consequence. Colour is used only to help scanning.

PASS

Pass

The observed result met the agreed threshold.

Retain the run result and continue.
WARNING

Warning

The result needs review but does not automatically block the report.

Assess impact and record any limitation.
FAIL

Fail

The observed result breached the threshold for this reporting purpose.

Hold or limit the affected report and assign the failed rows as agreed.
ERROR

Error

The check did not complete, so it cannot support a quality conclusion.

Resolve the execution problem and run it again.

Data-quality review history

Carry the rule, failed records and decision through one history.

The same configured check is used for the first run and retest so the correction can be compared with the observed failure.

Each check names the order or delivery fields tested, the threshold and the report consequence. The run keeps its input period, identifier, time, observed result and failed record keys.

A missing customer code is assigned to the customer-data owner; a missing product mapping goes to the product-data owner. The reporting owner records whether the affected page is held or issued with a stated limitation.

After the customer, product or delivery record is corrected, the same rule is run again by a second analyst. Only a passing retest and reconciled row total allow the held row or report page to return.

Check result and incident

A failed check is evidence. An incident is a managed response.

CHECK RESULT

What one rule observed in one run

It records the rule, threshold and severity alongside the input, run time and result state. Failed record keys retain the observed values and show which measure or report is affected.

DATA INCIDENT

How a material failure is resolved and closed

It adds impact, urgency and recurrence, then keeps the data owner’s response, correction or accepted limitation. Retest evidence supports the final closure decision and history.

The check and retest package

Controls that can be run, investigated and handed over.

01

Define the controls

Agree what should be tested and what each result means for the reporting purpose.

The retained package contains rule, severity and threshold register and dataset, field and affected-report mapping.

02

Execute and investigate

Retain the run and the records needed to explain where the data failed.

The retained package contains configuration-driven SQL or Python checks and run metadata, failed records and quarantine table.

03

Resolve and monitor

Connect material failures to an owner, a reporting decision and evidence from the retest.

The retained package contains issue or accepted-limitation record and named owner, retest history and readiness decision.

Executable project evidence

Four builds connect failed records to a report decision.

The retained builds cover validation, issue history, SQL controls and metric reconciliation. Their project pages explain the input records, execution, tie-out and limitation behind each view.

01

Executed rule results

Order and Delivery Data Quality Checks

Executed Python check results listing nineteen order and delivery rules, their dimensions, initial states, failed-row counts and responsible roles

A repeatable set of checks that stops incomplete or contradictory order records from entering the monthly service report and gives each failed row a specific correction reason.

The initial run separates 15 failures, three warnings and one execution error. Each row also retains the affected report section and correction reason.Inspect this project
02

One reporting incident from detection to release

Reporting Data Issue and Retest Register

Excel incident trace for order line OL-0001001 from missing product category through correction, independent retest and report release

A reporting data incident register that starts with a retained failed row and ends with verified closure, an approved limitation or a continued hold.

PRD-UNKNOWN is replaced by PRD-01001. The retained SQL returns Product category 06 and PASS before the reporting owner releases the corrected report.Inspect this project
03

Executed mart and release results

SQL-backed Reporting Foundation

Executed DuckDB evidence showing sales, purchasing and stock fact counts, the May 2016 commercial result and passing release controls

A wholesale reporting mart built from Microsoft’s Wide World Importers sample database so sales, purchasing, stock and margin reports use the same customers, products, dates and totals.

The retained result combines the actual query result with the release decision. It shows what loaded, what the latest monthly mart contains and whether the checks allow publication.Inspect this project
04

Exact percentage bridge

On-time Delivery Metric Reconciliation

Numerical bridge from five published on-time delivery percentages to the approved result

A row-level reconciliation explaining why the SQL reporting mart, Power BI model, Excel management pack, operational extract and manually maintained report show different on-time delivery percentages.

Calculation, event-date, record-grain and extract-population adjustments bring every reported version to the same numerator and denominator with no residual.Inspect this project

Technical patterns

Put the rule at the layer that can explain and maintain it.

Schema rules

Use when

Columns, types, keys or file structures must remain stable enough for the next process to run.

Boundary

A valid structure does not show that the values are complete, timely or meaningful for the report.

Data rules

Use when

Records need tests for completeness, uniqueness, validity, freshness, sequence or reference integrity.

Boundary

Thresholds must reflect the reporting purpose; a universal quality score cannot make that decision.

Business rules

Use when

Amounts, populations or states must reconcile to an agreed definition before a report can be issued.

Boundary

The rule needs a business owner and must not silently replace a disputed definition.

Issue response

Use when

A recurring or material failure needs ownership, impact assessment, a recorded data correction and independent retest.

Boundary

Minor warnings can remain check results. Creating an incident for every failed row adds noise without improving the data.

Related capabilities

Execute the check in the simplest maintainable environment.

Evidence boundary

A check can show what failed. It cannot prove unrecorded activity is complete.

Checks test observed data against agreed rules. They do not prove that unobserved real-world activity is complete, and a failed check becomes an incident only when its threshold or reporting impact warrants managed resolution.

Bring one report, the records that cause concern and the decision due next. Confidential or protected data is not needed for the first conversation.

Discuss a data-quality reviewReturn to all solutions