Security Operations Evidence Correlation Engine

An incident-analysis workflow that correlates host, identity, and network events into timelines for faster security operations decisions.

Impact

Built timeline-driven analysis from fragmented security events.

Impact

Prioritized alerts by evidence density and blast-radius signals.

Impact

Documented clear escalation paths and containment recommendations.

Deliverables

  • Incident timeline
  • Correlation rules
  • Responder summary

Technical Overview

Individual alerts rarely explain an incident by themselves. The project needed a method for joining identity events, endpoint telemetry, DNS, proxy logs, and ticket context into a defensible sequence of activity.

Architecture

flowchart LR identity_logs[Identity logs] --> normalized_events[Normalize] endpoint_alerts[Endpoint alerts] --> normalized_events network_logs[DNS and proxy logs] --> normalized_events normalized_events --> correlation_rules[Correlation rules] correlation_rules --> incident_timeline[Incident timeline] incident_timeline --> response_decision[Containment decision]

Model

C(h)=1eEh(1ce)C(h) = 1 - \prod_{e \in E_h}(1 - c_e)

where C(h) is host compromise confidence and c_e is the confidence contribution of each event on host h.

Implementation Sketch

index=security earliest=-24h
| eval entity=coalesce(host, user, src_ip)
| stats values(alert) as alerts, min(_time) as first_seen, max(_time) as last_seen by entity
| where mvcount(alerts) >= 3
| sort first_seen

Engineering Approach

  • Normalize event time, host identifiers, users, source IPs, and alert names into a shared schema.
  • Join weak signals into higher-confidence incident hypotheses.
  • Record both confirmed and unconfirmed assumptions so responders can act without losing uncertainty.

Results

  • Turned alert clusters into concise incident narratives that identify scope, likely entry point, and next action.
  • Separated duplicate telemetry from meaningful progression such as initial access, discovery, lateral movement, and exfiltration risk.
  • Improved handoff quality by using evidence tables and timestamped assumptions.

What This Demonstrates

  • Security operations work improves when analysis is structured around timelines, not alert volume.
  • Correlation rules need confidence labels because not every joined signal is causal.
  • Responder-ready writing is part of the technical deliverable.