Docs
Your event tables
Every event lands in an Apache Iceberg table in your own cloud account. Any tool that reads Iceberg can query it, with or without Torque Lift.
Where they live
In the bucket the setup stack created in your account, as Iceberg v2 tables in one namespace. That copy is the system of record. The Torque Lift engine reads from it to build your readout, and is never the only copy.
| Table | Holds |
|---|---|
events | Events as they happen, from your app's tracking |
events_import | Events brought in from your past analytics tool, with the same columns |
quarantine | One row for each call that was turned away, with the reason, never the call itself |
Rows are partitioned by received_date and sorted by event and time. message_id is the key that removes duplicates, so a call sent twice lands once.
What a row holds
The shape follows the Segment spec. Changes to these columns only add new ones.
| Column | Type | What it holds |
|---|---|---|
message_id | string | The call's unique ID, and the key that removes duplicates |
type | string | track, identify, page, screen, group or alias |
event | string | The event's name, on a track call |
name | string | The route template, on a page or screen call |
user_id | string | Your user's ID. A row has this or anonymous_id |
anonymous_id | string | The browser or device ID, before sign-in |
previous_id | string | The earlier ID, on an alias call |
timestamp | timestamp | When the event happened, corrected for clock skew |
sent_at | timestamp | When the sender sent it |
received_at | timestamp | When Torque Lift's collector, or the import, received it |
received_date | date | The UTC date of received_at, and the partition |
properties | string | JSON text, declared keys only |
traits | string | JSON text, declared keys only |
context | string | JSON text, from an allowed list of keys, such as the app's version |
source | string | Where the row came from. One of jitsu (live tracking), posthog, mixpanel, segment or amplitude |
plan_version | int | The tracking plan's version when the row was written |
plan_sha | string | The plan's commit, so you can open the exact plan |
What was removed, and how to see it
Five more columns say what happened to a row on the way in. They never hold the removed value.
| Column | What it holds |
|---|---|
_pii_dropped | How many values were removed from this row |
_pii_blocked | Of those, how many were removed as personal data. Empty on older rows |
_dropped_keys | The names, never the values, of properties the plan doesn't declare, up to 20. Empty when none |
_batch_id | Where the row came from in the pipeline, for tracing |
_written_at | When the row was written |
A call that can't be kept at all goes to the turned-away table, with one of a closed list of reasons. Among them: not valid JSON, an unknown call type, no received time, and no user or anonymous ID left after filtering. Your console's Data page shows the counts by day and reason.
Query it
Point any query tool that reads Iceberg at the bucket, with a role your team owns. This counts events by day and name:
-- Events by day and name, from a date you choose
SELECT received_date, event, count(*) AS events
FROM events
WHERE type = 'track'
AND received_date >= DATE '2026-11-13'
GROUP BY received_date, event
ORDER BY received_date, events DESC;The growth tables
Torque Lift builds growth accounting and experiment tables from these events with dbt, in an independent pipeline. Their definitions are in Readouts. They are yours to open too. Handing the stack to your own data team is a permissions change, not a migration.