Skip to main content

IBM Verify Events

Warning

Use IBM Verify Counts instead. These tables hold one row per event and are kept for 3 years. At full volume they will grow by hundreds of millions of rows a month. Query them only when no summary table answers the question, and always filter on the partitions.

The IBM Verify event stream, in the ibm_verify_events_raw schema. An hourly Glue job pulls events from IBM Verify through a proxy in the CanadaLogin platform account. The proxy hashes every personal identifier before the data leaves that account. See the streaming job and its privacy design, both in an internal repo.

Data starts at launch, 2026-04-22, except adaptive_access, which starts on 2026-09-09.

Tables

Table Description
ibm_verify_events_raw.authentication_activity One row per IBM Verify authentication event.
ibm_verify_events_raw.application_usage One row per IBM Verify SSO event, when someone reaches an application.
ibm_verify_events_raw.mfa_activity One row per IBM Verify multi-factor authentication event.
ibm_verify_events_raw.adaptive_access One row per IBM Verify adaptive access evaluation.
No matching items

Users

The user ID is subject in authentication_activity and mfa_activity, and userid in application_usage and adaptive_access. Both hold the same HMAC-SHA256 digest of the IBM Verify user ID, so they join across tables and stay stable over time. Count users as distinct values of it.

Exclude null, "", and the literal "UNKNOWN". IBM Verify logs "UNKNOWN" when it cannot tie an event to an account, such as a password attempt on a username that does not exist. The summary tables keep it and count it as one user, to match IBM’s own figures.

username, origin (the client IP address), and mfadevice are hashed the same way. Countries and regions are not hashed.

Time and partitions

time is a UTC string, for example 2026-09-15 00:00:05 UTC. The partition columns year, month, and day are strings for the UTC day.

Always filter on the partitions, which lets Athena skip every other day. Filtering only on time reads the whole table.

Reports use ET days, which end at 04:00 or 05:00 UTC. Read the following UTC day too, then convert:

date(with_timezone(CAST(substr(time, 1, 19) AS timestamp), 'UTC') AT TIME ZONE 'America/Toronto')

Distinct-user counts do not add up across days. See Why Unique Users Cannot Be Added Up.