Your Chart Looks Fine. Did the Data Load Twice?
A chart can look plausible with duplicated data. Define the daily-channel key and reconcile row counts and totals before trusting a new Power BI report.

The chart has the right shape. Monday is busier than Tuesday and the channels appear in the expected order. Nothing looks obviously wrong.
Then an independent check finds that the source data were included twice. A familiar pattern survived while every affected count grew. Visual plausibility was never a sufficient control.
Define one row before counting rows
Imagine a fictional report built from daily service totals. One row represents one date and one channel. That combination is the intended key for this exercise; it is not a list of individual customer cases.
Write down the expected grain before importing or combining files. A repeated date may be normal because several channels report on that day. A repeated date-channel combination needs investigation under this particular rule.
A duplicate check must reflect the source’s actual meaning. Do not remove rows simply because several values happen to match. Two legitimately different observations can contain identical numbers.
Keep an independent control
Record an expected row count and a few totals from the approved fictional source. Compare them after loading and after any transformation that could change the structure.
Use a control that is independent of the same transformation being tested. Two visuals built from the same duplicated table can agree perfectly while repeating the same mistake.
Check both the number of records and their values. A correct row count does not prove every record is the right one, and a matching total can occasionally conceal offsetting errors.
Review combinations and relationships
Combining files or attaching descriptive information can change a dataset unexpectedly when keys are not unique. Inspect unmatched and repeated keys before relying on the joined result.
Microsoft’s star-schema guidance emphasises a consistent grain for fact data. That principle is useful even in a modest learning project: know what observation each row represents, then make the model preserve it.
For a daily-channel exercise, use deliberately distinct test values. They make a duplicated contribution easier to trace than a table full of similar figures.
Keep the exception visible until it is explained
When a repeated key appears, record where it came from and whether it is a source duplicate, an import problem or a legitimate change to the data structure. Correct the cause rather than hiding the symptom in the chart.
After the correction, rerun the row-count and value checks and record the observed results. Keep any unresolved question in the validation note.
The goal is an explainable report, not a chart that happens to look familiar. A small control table can protect that explanation better than an hour spent adjusting colours.
Continue with a related question
Explore Build a Power BI Report You Can Check and Hand Over, or read The Report Opens. Can Someone Else Refresh It Next Month?.
For a written practice option, see Power BI with Matze: contents and requirements.
Review the provider’s course details. Affiliate link: Superstore may earn a commission if you buy.
Technical reference: Microsoft Learn: Power BI model design.




Discussion 0
What did this article make you think about?