6.2Quick troubleshooting
The essentials in 30 seconds
- Diagnose: one symptom = likely causes = a sequence of actions — we don't fix at random.
- Roots: relationships, types, date table, filter context, column vs measure, performance — 90% of blockers.
- Verify: read the cell context, mark the date table, time in the Analyzer, prove on an edge case.
Zineb's thread — the table stuck beside the screen
Zineb keeps this table beside her screen in Algiers. Verdict after a quarter: three rows served her the most — totals identical on every row of a visual, prior-year revenue that stays blank, and a report that drags. For each, she ended up applying the sequence (symptom, causes, actions) instead of rewriting measures at random.
Symptom observed, causes verified, actions in order: the reflex holds in three steps.
One reflex per symptom
Most Power BI blockers are classics. Linking each symptom to its likely causes and to a sequence of actions saves precious time.
The context before the formula
A « weird » DAX result rarely comes from the formula alone: it's almost always a misunderstood filter context or an incomplete relationship. We observe, verify, then fix.
Optimize on evidence
For any slowness, we time it in the Performance Analyzer before touching the model: we optimize in the order model → measures → visuals.
| Symptom | Likely causes | Actions |
|---|---|---|
| Identical totals on every row of a visual | missing, broken or inactive relationship between the crossed tables | Model view: create or repair the relationship (1:* cardinality, single direction); check the key column |
| « Weird » DAX result or wrong at the total | misunderstood filter context; average of ratios on the Total row | read the cell context; recalculate by iteration (SUMX); check CALCULATE and ALL |
| Prior-year revenue (or YTD) blank or wrong | date table not marked; relationship through the wrong column; history too short | mark as a date table; check the date relationship; check the history depth |
| Months sort out of order (April before January) | alphabetical sort of the Month column | sort the Month column by a MonthNumber column (Sort by column) |
| My calculated column doesn't react to slicers | a column is frozen at refresh; a measure was needed | create a measure (recalculated on the fly according to the context) rather than a column |
| Slow report (several seconds per visual) | unoptimized model, useless columns, too many visuals, auto date/time | Performance Analyzer; remove useless columns; disable the auto date/time hierarchy |
Ready-to-say script — diagnosing without random rewrites (30s)
« My symptom: [identical totals / DAX wrong at the total / blank prior year / slow]. I check the likely cause in the table (relationship, context, date table, performance), I apply the targeted action and I prove it on an edge case. »
The reflex holds in three steps: symptom, cause, action.
Ready-to-say script — checking a measure's context (30s)
« My measure shows [an unexpected result]. I read the cell context out loud: which filters are active? I check the relationship and the date table, then I fix the cause, not the formula. »
Always verify the context before rewriting a measure.
Copy-paste template — DAX band-aids
Line Revenue = SUMX(Sales, Sales[Quantity] * Sales[UnitPrice])
Growth % = DIVIDE([Total Revenue] - [Revenue PY], [Revenue PY])
Revenue PY = CALCULATE([Total Revenue], SAMEPERIODLASTYEAR(Dates[Date]))
-- PY blank? Mark the date table and check the relationship.Copy-paste template — safe fallback
Rank = RANKX(ALL(Products[Product]), [Total Revenue])
Line Average = SUMX(Sales, DIVIDE(Sales[Amount], Sales[Quantity]))
-- Check: 1:* relationship, marked date table, column vs measure.SME / small budget
No formal table: one symptom, the likely cause, the action and who proves it, noted down.
SME Algiers
Integrate real exports into the diagnosis: local separator, FR dates, known reference person.
Remote / hybrid
Add a screenshot to the diagnosis: at a distance, a written symptom and proof avoid duplicates.
Intercultural
Make the findings explicit in writing: what seems obvious (which department, which format) is not for everyone.
48h challenge — apply tomorrow
Spot 1 current symptom (identical totals, blank prior year, disordered months, slowness), apply the table row (named symptom + verified cause + targeted action) and prove it.
- I can link a symptom to its likely causes.
- I check relationships, context and date table before fixing.
- I prove every fix (totals, edge case, Analyzer).
Zineb sees the same totals on every row of a visual. Which reflex does the table apply?
Reflex: diagnose the symptom, identify the cause, apply the targeted action — and time any slowness in the Analyzer before touching the model.