GA4 Standard

Devices

The device a session ran on: category, operating system and its version, browser and its version, one row per combination. When a session's events disagree, the value from its latest event wins. Sessions point here through deviceHash, a FARM_FINGERPRINT of the five values. The table has no date, so a hash repeats once per run (once per day for daily runs). Count distinct hashes. In your own SQL over these tables, deduplicate it before joining, or rows multiply; OWOX Data Marts reports collapse joined fields per source row.

Fields

Event Traffic Sources

The traffic source GA4 records on a single event or page view, from its source, medium, campaign, term and content parameters: UTM tags, or GA4's own attribution such as organic or referral. One row per combination. Events and Pageviews point here through eventTrafficSourceHash. Most hits carry none of these parameters: they share one hash, whose row has every value empty, meaning nothing was recorded on the hit, not a missing record. The table has no date, so a hash repeats once per run (once per day for daily runs). Count distinct hashes. In your own SQL over these tables, deduplicate it before joining, or rows multiply; OWOX Data Marts reports collapse joined fields per source row.

Fields

Events

Every GA4 event other than a page view, one row each, tied to the session it fell in. Page views are in Pageviews. Join to Sessions on date and pseudo_session_id together, because the session number restarts every day. hitId identifies an event: a rerun of the same window gives the same value, a different window does not, and Pageviews numbers its hits separately, so one value can appear in both. time is a timestamp in microseconds. eventTrafficSourceHash points at the source GA4 records on this one event (UTM tags or GA4's own attribution); most events carry none. The table holds no event parameters beyond the event name.

Fields

Relationships

Geo

Where a session came from, as GA4 resolves it from the visitor's IP address: country, region and city, one row per combination. When a session's events disagree, the value from its latest event wins. Sessions point here through geoHash, a FARM_FINGERPRINT of the three values. The table has no date, so a hash repeats once per run (once per day for daily runs). Count distinct hashes. In your own SQL over these tables, deduplicate it before joining, or rows multiply; OWOX Data Marts reports collapse joined fields per source row.

Fields

Pageviews

Every GA4 page_view event, one row each, tied to the session it fell in. Other events are in Events. The table carries no page path: the landing page is on Sessions. Join to Sessions on date and pseudo_session_id together, because the session number restarts every day. hitId identifies a page view: a rerun of the same window gives the same value, a different window does not, and Events numbers its hits separately, so one value can appear in both. time is a timestamp in microseconds. eventTrafficSourceHash points at the source GA4 records on this one page view (UTM tags or GA4's own attribution); most carry none.

Fields

Relationships

Sessions

One row per session as this model cuts it, with its landing page, engagement, purchases and keys to its source, device, location and user. A new row starts at GA4's session_start, after 30 idle minutes in the runbook's setup, at midnight, and on every event that carries a source, medium or gclid. So rows can outnumber GA4's sessions several times; on Google's sample, about six times. Count sessions with the sum of sessions, which tracks GA4's count closely. A bounce rate built per row is not comparable to GA4's. A row is keyed by date and pseudo_sessionId. Times are in microseconds; cost, impressions and clicks are always 0. For deduplicated orders, use Transactions. Events without a user_pseudo_id are lost.

Fields

Relationships

Traffic Sources

The source, medium, campaign, keyword and ad content a session is attributed to, one row per combination. They come from parameters GA4 records on events: UTM tags, or GA4's own attribution such as organic or referral. A session opened by session_start takes the first values its GA4 session carries; one split off mid-visit takes those of the event that split it. With none it reads (direct) / (none). Sessions point here through trafficSourceHash. The table has no date, so a hash repeats once per run (once per day for daily runs). Count distinct hashes. In your own SQL over these tables, deduplicate it before joining, or rows multiply; OWOX Data Marts reports collapse joined fields per source row.

Fields

Transactions

One row per purchase, deduplicated by its transaction ID: when GA4 records the same order more than once within the deduplication window (7 days in the runbook's setup), only the first is kept. Revenue is in the currency the site reported. Per deduplication window, at most one purchase with the ID (not set) and one with an empty ID are kept, so their revenue is understated. Purchases on events without a user_pseudo_id are missing. status is always empty and touchpoint is always WEB; both are slots. Join to Sessions on date and pseudo_session_id for the source and device behind each order. Totals here can differ from the purchase counts on Sessions, which are not deduplicated.

Fields

Relationships

User Traffic Sources

How each user was first acquired: the source, medium and campaign GA4 records for the user's first visit, not for any later session. Missing values read (not set). Sessions point here through userTrafficSourceHash, which matches trafficSourceHash in this table although the names differ; both are a FARM_FINGERPRINT of the three values. The table has no date, so a hash repeats once per run (once per day for daily runs). Count distinct hashes. In your own SQL over these tables, deduplicate it before joining, or rows multiply; OWOX Data Marts reports collapse joined fields per source row.

Fields

Users

The visitors behind sessions, one row per GA4 pseudo user ID: the identifier GA4 gives a browser or app instance, not a person. The same person on two devices, or after clearing cookies, is two users here. clientId, userId, visitorId, phone and email are always empty: slots for identifiers from other sources. Sessions point here through userHash, a FARM_FINGERPRINT of all six identifiers, today the pseudo user ID alone. The table has no date, so a hash repeats once per run (once per day for daily runs). Count distinct hashes. In your own SQL over these tables, deduplicate it before joining, or rows multiply; OWOX Data Marts reports collapse joined fields per source row.

Fields