Software Troubleshooting practice questions

Re-run all queries

Why it fails: Re-running queries reproduces the same conflicting outputs unless the underlying cause, such as a join, filter or refresh timing issue, has changed. It is tempting because rerunning is a valid step after identifying and fixing a failed job or stale extract, where execution itself was the fault.

Notify stakeholders of potential error

Why it fails: Alerting stakeholders before diagnosing the discrepancy risks broadcasting an unverified error and eroding trust; notification belongs after root cause is established. It is tempting because communication plans rightly require prompt disclosure, and in a confirmed, material data incident that has already been validated, notifying affected parties would indeed be the correct action.

Replace the data source with the one believed to be correct

Why it fails: Swapping in the source assumed correct discards evidence and may propagate the actual fault, since neither source has been validated. It is tempting because replacing a known-bad feed is legitimate remediation, and once investigation proves one source defective, substituting the verified source would be the right move.

The database server is offline.

Why it fails: An offline database server yields connection-refused or host-unreachable errors, not SQL syntax errors. Syntax errors are generated by the parser processing the query text, which requires a reachable server. Server-availability checks suit a refresh failing with connection errors instead.

The user does not have permissions to access the table.

Why it fails: Permission failures surface as authorisation or access-denied errors, not SQL syntax errors. The parser rejects malformed statements before permissions are evaluated, so the logged syntax errors point to faulty query text. Permission troubleshooting would be correct for a table-access-denied message.

The network connection timed out.

Why it fails: A network timeout produces connectivity or timeout errors, not repeated SQL syntax errors. Syntax errors are raised by the query parser evaluating malformed SQL, independent of network state. Timeout diagnosis fits a refresh that hangs or drops connection mid-execution.

ETL (Extract, Transform, Load)

Why it fails: ETL extracts, transforms and loads data between systems; it is the mechanism that would have moved returns into the sales dataset. Its absence explains the gap, but the question asks which concept was missing, and ETL describes the pipeline rather than the cross-system ownership and definition of returns data.

Data profiling

Why it fails: Data profiling examines datasets for anomalies, nulls and value distributions; it would flag odd sales figures but cannot merge returns from another system into the report. Profiling is correct when assessing source data quality before migration, not for reconciling records across systems.

Data governance

Why it fails: Data governance defines ownership, definitions and policies for data assets; it would assign accountability for returns data but performs no physical integration. Governance is correct when establishing stewardship, standards and access rules, not for combining records from separate operational systems.

Start with the current scores, then show historical trend

Why it fails: Showing current scores then historical trend establishes a timeline but never isolates the software update as the cause, leaving correlation unaddressed. It is tempting because trend narratives suit showing change over time, and would fit a question asking how satisfaction evolved rather than why it dropped.

Present all data points without a story

Why it fails: Listing every data point without narrative gives the audience no causal link between the update and the satisfaction drop, so they must infer it themselves. It is tempting because raw data presentation suits exploratory analysis or data-quality review, not explanatory storytelling where a specific cause must be communicated.

Use a pie chart of satisfaction categories

Why it fails: A pie chart shows parts of a whole at one moment, so it cannot express the drop's timing relative to the software update. It is tempting because pie charts suit proportional breakdowns, such as satisfaction categories' share of total responses, but cause-and-effect needs a temporal sequence.

Add a filter to exclude null values.

Why it fails: Filtering out nulls hides the affected regions entirely, so the map still cannot display them; the stem asks for values to appear, not disappear. Filtering suits suppressing known-irrelevant rows, but here the nulls signal a data-join or region-name mismatch that must be corrected at source.

Change the mark type from Map to Pie.

Why it fails: Pie marks cannot be plotted on a geographic map, so switching mark type discards the spatial dimension that produces the regional nulls. It is tempting because pie charts display proportions clearly, and would be the correct choice for comparing category shares in a non-spatial view.

Remove the color encoding from profit.

Why it fails: Removing the colour encoding leaves the null regions still null; the missing geographic or profit data, not the colour shelf, causes the blanks. It is tempting because colour legends often look broken, and dropping colour would be right when the encoding itself misleads, for example an inappropriate diverging palette.