The export holds every event, so you can rebuild GA4's numbers in SQL. The traps are always the same:
- Sessions are distinct pairs of
user_pseudo_id and ga_session_id.
- Date ranges over
events_* use _TABLE_SUFFIX BETWEEN '20261001' AND '20261031', which only reads those tables. A filter on event_date still reads every table, and BigQuery bills you for it.
- Parameters: a subquery,
(SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'coupon'), keeps events that lack the parameter (as NULL). CROSS JOIN UNNEST(event_params) with a WHERE key = ... drops them.
- Engaged sessions carry the
session_engaged parameter, sent as the string '1'.
Each task shows four queries. Pick the one that answers it correctly. Write and run queries like these in the BigQuery course's GA4 lessons and its GA4 funnel project.