ADVISORI: synthetic login correlation exercise Purpose Follow one small, reproducible SIEM correlation example. All identities, applications and events are fictional. Addresses are reserved for documentation. No events are sent to a live SIEM, and no sign-in or network access is performed. Files siem-login-example-events-en.csv: 27 events with explicit UTC event timestamps. siem-login-example-expected-en.csv: seven expected decisions and their reasons. Demonstration rule For each success event S, select all failure events F where: F.account = S.account F.source_ip = S.source_ip F.application = S.application S.timestamp_utc minus 5 minutes <= F.timestamp_utc < S.timestamp_utc Mark S as a candidate alert if at least three such failure events exist. Report S.event_id, the qualifying failure count and the contributing event IDs. Replay 1. Parse the CSV using comma separation and UTC event time, not file row order. 2. Apply the rule to each successful login without merging different keys. 3. Compare the result with all seven rows in the expected-results file. 4. Inspect the exact lower boundary and the equal-timestamp case explicitly. Expected candidate alerts: E04 (E01, E02, E03) and E15 (E12, E13, E14). The other five successes must not become candidate alerts under this rule. Interpretation and limits The count and time window are illustrative, not production recommendations. A candidate alert is an investigation prompt, not proof of compromise. This example assumes complete, unique records and consistent UTC event time. It does not model late arrival, ingestion delay, duplicate records, NAT, shared accounts, session behavior or changes of source address. Real implementations must decide how to handle those cases and how to route any alert to an owner. Do not infer an operational detection rate from this small synthetic exercise.