Pass the DQA on a normal Tuesday.
A Data Quality Assessment tests five things — validity, reliability, precision, integrity and timeliness — and most organisations meet it by reconstructing months of evidence in the fortnight before the assessor arrives. Karagem runs those tests continuously instead: at the point of capture, in an approval queue, and in an overnight scan across every submission. The assessment becomes a report you export rather than a project you survive.
The five dimensions, and what is checked against each
Validity
Does the number actually measure what the indicator says it measures?
Indicators carry their definition, unit and numerator/denominator with them, and a submission can only feed an indicator through a mapping somebody made deliberately. A question about bags per acre cannot end up counted as households.
Reliability
Would the same measurement, taken again the same way, give the same answer?
Form versions are recorded against every submission, so a change to the question does not silently change the series. Where a definition is revised, the indicator shows the break rather than smoothing over it.
Precision
Is the margin of error small enough for the decision being made on it?
Range and consistency rules run at the point of capture, so out-of-range values are refused in the field. GPS outliers and interviews completed suspiciously fast are flagged overnight for review.
Integrity
Could the number have been manipulated, by accident or on purpose?
Nothing counts until it is approved, and every approval names a reviewer and a time. Duplicate registrations are surfaced as a pair to be judged, never merged silently. Edits leave an audit trail.
Timeliness
Is the data current enough to be worth acting on?
Capture time and sync time are stored separately, so a submission that waited two days in an outbox is not mistaken for one collected this morning. Every indicator shows when it was last recorded.
Why the fortnight before never works
The reason a DQA is stressful is not that the data is bad. It is that the evidence for the data is scattered: the source documents are in a filing cabinet in a district office, the recalculation lives in a workbook one person maintains, and nobody can reconstruct why a number changed between the March report and the June one.
An assessor asks three questions in different words — where did this come from, who checked it, and can you show me the same trail for a different month. Every one of them is a question about provenance, and provenance cannot be added afterwards. It is either recorded as the work happens or it is testimony.
So Karagem records it as the work happens. A number in a report traces to an approved submission, to the enumerator who captured it, to the time and place of capture, and to the form version it was answered on — for any month, without anyone preparing for the question.
What the daily scan looks for
Out-of-range values
A yield of 400 bags per acre or an age of 130. Caught in the field where it can be corrected, not in the report where it can only be explained.
Duplicate registrations
The same household entered twice by two enumerators. Shown as a pair with what differs, for a person to resolve.
GPS outliers
A submission recorded fifty kilometres outside the ward it claims. Sometimes a legitimate referral, sometimes not — either way, worth a question.
Suspiciously fast interviews
A thirty-eight question form completed in four minutes. No accusation, just a flag: the pattern is worth a supervisor looking at.
Missing required evidence
An attendance record with no signatures, a distribution with no photo. The submission is held rather than counted.
Stale indicators
Nothing recorded against a target for six weeks. Silence is a data quality problem too, and it is the one nobody reports.
For the assessment itself
Every indicator exports with its definition, its source submissions, its approval history and its recorded dates — which is the evidence pack a DQA asks for, in the order it asks for it. Tables export to Excel or CSV, and there is a REST API if your assessor wants to pull rather than be sent.
None of that makes a number true. What it does is make the answer to “how do you know” something you can show in a minute instead of a fortnight.
If the gaps this finds are about how data is collected rather than how it is stored:Data Management in Agriculture4 weeks · All levels · USD 30.99