Sample SLA report
This is the record Perstat keeps over a quarter. Every number below is sample data and adds up; the structure, timestamps, and exclusion logic are exactly what the product records. A one-click report export is the stated next step; today the record lives in the product.
Summary · api.example.com · Q2 2026
| metric | value |
|---|---|
| availability, gross | 99.925% |
| availability, after curated exclusions | 99.994% |
| confirmed incidents | 1 |
| maintenance windows | 1 |
| check interval | 30 s |
| checks executed | 1,572,480 (6 regions, 91 days) |
| regions confirming | na-ewr, eu-fra, as-sgp, sa-gru, af-jnb, oce-syd |
Windows
| window (UTC) | cause | duration | counted against SLA |
|---|---|---|---|
| 2026-05-03 02:41:12 to 02:48:24 | incident, quorum-confirmed | 00:07:12 | yes |
| 2026-05-17 22:00:00 to 23:30:00 | maintenance, announced | 01:30:00 | no, excluded |
The exclusion window carries its announcement timestamp and its real boundaries. An exclusion without a matching announced window would not be accepted here, and an incident’s start is the quorum confirmation, not the first failed check and not a rounded minute.
Incident record
| time (UTC) | event |
|---|---|
| 02:40:42 | check fails, eu-fra |
| 02:41:12 | quorum confirms, na-ewr and as-sgp agree. Incident opens. |
| 02:41:12 | page sent, push |
| 02:44:03 | acknowledged |
| 02:48:24 | recovery confirmed across regions. Incident closes. |
| 02:55:00 | post-mortem opened, 2 action items |
Window boundaries in the SLA table above are 02:41:12 to 02:48:24: confirmation to confirmed recovery. The 30 seconds before confirmation are visible in the record but are not silently stretched or shrunk.
Why this holds up
- Timestamps come from the monitoring system itself, not from a human recollection.
- Exclusions are curated and visible, never silent.
- Availability is floored, never rounded up: an SLA number should be a lower bound you can defend.
- The same record feeds your status page, so the public story and the audit story cannot drift apart.
The math is checkable: 91 days are 131,040 minutes; 7:12 of confirmed incident and 90:00 of announced maintenance make 99.925% gross, and the curated exclusion turns it into 99.994% counted. Hand this to a customer, an auditor, or a regulator, and let them recompute it.