# D3: engineer a separate logging-health monitor

This is an engineering design, not an ordinary Sigma rule or a native endpoint query. A query over records that arrived cannot discover every host that sent no records. Select your own supported inventory, collector metrics, and management-health sources; no stock customer table or normalized schema is assumed here.

1. Keep an authoritative off-host expected-host inventory and bounded maintenance windows. Alert if inventory itself stops refreshing.
2. At an approved five-minute interval, arrange a benign unique syslog marker through a supported platform facility and record its expected sequence/destination independently. Select and document that facility for the installed platform. This pack runs no host command and creates no monitor.
3. Observe independent successful management checks, forwarding receipt, queue/drop/ingestion health, and source identity. A ping or live collector process alone is insufficient.
4. For each expected host, choose the latest whole observation before testing freshness. Never combine newer valid fields with older fields, and never silently discard malformed latest rows. Missing snapshots, malformed observations, conflicting timestamps, and stale evaluator output are monitor incidents on a second delivery path.
5. Age every measurement forward to evaluation time, including transit delay. A starting investigation threshold is marker age over 900 seconds while management and collector evidence each remain within 300 seconds; allow 1800 seconds for initial enrollment and suppress only approved maintenance windows. Tune against actual scheduling delay and source clocks.
6. Correlate unexplained loss with independently retained vCenter, identity, network, and admin activity. Preserve existing records before configuration changes. Route collector/network outages as health incidents; do not label every gap an intrusion.

A healthy heartbeat cannot prove that every operation generated an audit event, so this control does not detect or close CVE-2026-41709 by itself. It also cannot recover an event that was never written. Patch according to current vendor guidance and preserve independent evidence. See the Broadcom advisory and shell-auditing documentation linked in SOURCES.md.
