LoFP / shared systems such as kiosks or conference room computers may have multiple users authenticating.

FROM logs-okta.system-* METADATA _id, _version, _index
| WHERE
    event.dataset == "okta.system"
    AND (event.action LIKE "user.authentication.*" OR event.action == "user.session.start")
    AND okta.outcome.reason IN ("INVALID_CREDENTIALS", "LOCKED_OUT")
    AND okta.actor.alternate_id IS NOT NULL
// Build user-source context as JSON for enrichment
| EVAL Esql.user_source_info = CONCAT(
    "{\"user\":\"", okta.actor.alternate_id,
    "\",\"ip\":\"", COALESCE(okta.client.ip::STRING, "unknown"),
    "\",\"user_agent\":\"", COALESCE(okta.client.user_agent.raw_user_agent, "unknown"), "\"}"
  )
// FIRST STATS: Aggregate by (IP, user) to get per-user attempt counts
// This prevents skew from outlier users with many attempts
| STATS
    Esql.user_attempts = COUNT(*),
    Esql.user_dt_hashes = COUNT_DISTINCT(okta.debug_context.debug_data.dt_hash),
    Esql.user_source_info = VALUES(Esql.user_source_info),
    Esql.user_agents_per_user = VALUES(okta.client.user_agent.raw_user_agent),
    Esql.devices_per_user = VALUES(okta.client.device),
    Esql.is_proxy = VALUES(okta.security_context.is_proxy),
    Esql.geo_country = VALUES(client.geo.country_name),
    Esql.geo_city = VALUES(client.geo.city_name),
    Esql.asn_number = VALUES(source.as.number),
    Esql.asn_org = VALUES(source.as.organization.name),
    Esql.threat_suspected = VALUES(okta.debug_context.debug_data.threat_suspected),
    Esql.risk_level = VALUES(okta.debug_context.debug_data.risk_level),
    Esql.risk_reasons = VALUES(okta.debug_context.debug_data.risk_reasons),
    Esql.event_actions = VALUES(event.action),
    Esql.first_seen_user = MIN(@timestamp),
    Esql.last_seen_user = MAX(@timestamp)
  BY okta.client.ip, okta.actor.alternate_id
// SECOND STATS: Aggregate by IP to detect credential stuffing pattern
// Now we can accurately measure the distribution of attempts across users
| STATS
    Esql.unique_users = COUNT(*),
    Esql.total_attempts = SUM(Esql.user_attempts),
    Esql.max_attempts_per_user = MAX(Esql.user_attempts),
    Esql.min_attempts_per_user = MIN(Esql.user_attempts),
    Esql.avg_attempts_per_user = AVG(Esql.user_attempts),
    Esql.users_with_single_attempt = SUM(CASE(Esql.user_attempts == 1, 1, 0)),
    Esql.users_with_few_attempts = SUM(CASE(Esql.user_attempts <= 2, 1, 0)),
    Esql.first_seen = MIN(Esql.first_seen_user),
    Esql.last_seen = MAX(Esql.last_seen_user),
    Esql.target_users = VALUES(okta.actor.alternate_id),
    Esql.user_source_mapping = VALUES(Esql.user_source_info),
    Esql.event_action_values = VALUES(Esql.event_actions),
    Esql.user_agent_values = VALUES(Esql.user_agents_per_user),
    Esql.device_values = VALUES(Esql.devices_per_user),
    Esql.is_proxy_values = VALUES(Esql.is_proxy),
    Esql.geo_country_values = VALUES(Esql.geo_country),
    Esql.geo_city_values = VALUES(Esql.geo_city),
    Esql.source_asn_values = VALUES(Esql.asn_number),
    Esql.source_asn_org_values = VALUES(Esql.asn_org),
    Esql.threat_suspected_values = VALUES(Esql.threat_suspected),
    Esql.risk_level_values = VALUES(Esql.risk_level),
    Esql.risk_reasons_values = VALUES(Esql.risk_reasons)
  BY okta.client.ip
// Calculate stuffing signature: most users should have very few attempts
| EVAL Esql.pct_users_few_attempts = Esql.users_with_few_attempts * 100.0 / Esql.unique_users
// Credential stuffing: many users, most with 1-2 attempts each, low max per user
// Stacked stats gives us accurate per-user distribution instead of skewed averages
| WHERE
    Esql.total_attempts >= 25
    AND Esql.unique_users >= 15
    AND Esql.max_attempts_per_user <= 2
    AND Esql.pct_users_few_attempts >= 80.0
| SORT Esql.unique_users DESC
| KEEP Esql.*, okta.client.ip