Source selection and data discipline
Every log collected is both cost and noise. Sources are chosen by which detection scenario they feed; a source serving no scenario is not collected. Identity, endpoint and network layers usually take the first three places.
- Every source maps to at least one detection scenario
- Clock sync and time zones are fixed at the start
- Retention follows what the scenarios require
- Removing a source is planned as deliberately as adding one
Detection rules and tuning
Out-of-the-box rule sets are a starting point, not a destination. Rules are tuned to your own behavior; a rule with high false positives gets narrowed rather than switched off. Every rule carries an owner and a response step.
- No rule goes live before its response step exists
- False-positive rate is tracked per rule
- Rules are retuned as organizational behavior changes
- Any disabled rule is recorded with its reason
Incident flow and escalation
How an alert becomes an incident and reaches the right person is defined as a flow. Who triages, when it escalates, at what threshold your own team steps in — all written before the incident. A flow learned during a crisis is not a flow.
- Escalation is defined by role, never by person
- The threshold for involving your team is written down
- Closure requires a verification step
- Every incident ends with an outcome record
Measurement and improvement
The operation itself is measured: how many alerts arrived, how many became incidents, how many were real, how long closure took. Those four numbers show whether the setup is working.
- A low true-incident rate means the rules need tuning
- Recurring incidents are routed to permanent fixes
- Blind spots are compared against scenario coverage
- Measures use the same definition across periods