Review questions
Integrated summary: Core data concepts
Classify a dataset by inspecting each record's fields and their stability. A JSON file may contain optional keys; sharing a container does not impose one fixed schema. Images and free text can be stored and retrieved, but file count cannot reveal how often a term appears inside content before extraction and accuracy review. Choose storage from the unit of reading and change: an object suits an image opened by ID, while an operational database manages mutable order state under constraints. Link them by a checked identifier without making one service perform both jobs. Preventing a double sale is a concurrent operational decision; aggregating sales over months is a historical analytical question. When a report count differs, trace source, ingestion, transformation, model and visual counts. Investigate the first divergence. A visual filter matters when the model is complete, while a dropping transformation matters when the model is already short. A duplicate scan is not a new arrival; a cancelled booking should not silently enter a valid-booking denominator. The database role checks source records and access, the data engineer checks transformation rules, and the analyst checks metric definitions. End with an explicit evidence boundary: unextracted image contents and unverified arrivals cannot become reported facts.
File formats and lineage add a practical check to the choice: neither a CSV nor JSON suffix proves schema validity. Define record grain, field types and key uniqueness, then reconcile input and output counts. A replayed payment_id must not be summed twice before event identity is checked. The engineer records transformation version and rejected counts, while the analyst defines the measure denominator; operating a database alone cannot determine report meaning.
Review questions
Test your understanding with these questions. Marking and saved results are available with a free account.
Create a free account to mark your answers and save your progress.