Blog
Self-hosted monitoring, and what its numbers really mean.
Why an MTTR can quietly measure your check interval. When a percentile should stay blank instead of being approximated. How nine failed samples become one issue, and where a read-only collector's boundary actually sits. Written from the source, with the limits stated.
Your MTTR is only as honest as its state machine
How soft failures, unresolved outages, and unknown time change the number
Read →8 August 2026Nine failed samples should not become nine issues
What RootTrace groups automatically, and what it deliberately leaves separate
Read →8 August 2026Turn CPU profile time into a monthly capacity estimate
Why RootTrace refuses to price wall-clock samples as CPU
Read →8 August 2026In 48% of Mandiant cases, detection came from outside
What request-context security signals can add, and where they can still fail
Read →8 August 2026“Read-only” monitoring needs a precise boundary
No remediation executor does not mean no local writes
Read →8 August 2026How an offline licence check actually works
Local ES256 verification, the free-tier fallback, and the revocation tradeoff
Read →8 August 2026A reproducible self-hosted monitoring quickstart
The exact Compose stack, its limits, and what teardown really removes
Read →