We publish the operating record because we run by it. Targets are public, actuals are measured, and the delta is whatever it is. No spin, no rounding in our favor.
The engineer who installed your node is on the rotation that carries it. There is no overseas tier-one reading from a script. An alert reaches a human in under three minutes, a fix or a status in the window, and a written post-mortem on anything that breaches target. The table below is the last twelve months, unedited.
| Metric | Target | Actual | Delta |
|---|---|---|---|
| Availability · fleet-wide | 99.95% | 99.984% | +0.034 |
| Mean time to detect | ≤ 60 s | 22 s | −63% |
| Mean time to acknowledge | ≤ 4 min | 2 m 18 s | −42% |
| Mean time to repair | ≤ 90 min | 41 min | −54% |
| Pages routed to a human | 100% | 100% | no script tier |
| Sites visited on-foot · year | 2× / site | 2.4× / site | +20% |
| Customers retained · 36 mo | — | 100% | — |
| Outages caused by us | 0 | 0 | — |
| Outages caused by upstream | — | 3 | logged, post-mortemed |
| Post-mortems published | all breaches | 3 / 3 | public |
Every node, mapped or under NDA, sits on one rotation. The carrier of a unit is on the list that answers for it. No node is orphaned to a queue, and no site is a second-class page.
An alert reaches a person, not an auto-responder, in under three minutes on average. We detect in seconds, acknowledge in minutes, and repair inside the window. The dashboard is the same one the customer can read.
Anything that breaches target gets a written, blameless post-mortem, published to customers. Three upstream outages in twelve months, three post-mortems. We caused none, and we still wrote them up.