| Rank |
Servers are ranked from lowest risk to highest risk. Rank 1 is the preferred server. |
| Risk Score |
Lower is better. The score is not capped, so severely unhealthy servers can have scores greater than 100. |
| Error% |
The observed percentage of order requests that failed during the trailing seven-day window. |
| Error Upper 95% |
The upper Wilson confidence bound for the order error rate. This accounts for statistical uncertainty in the smaller order sample. |
| Timeout% |
The percentage of gRPC probes that did not respond within 2.5 seconds during the trailing seven-day window. |
| Median Latency |
The median latency of successful orders during the trailing seven-day window. |
| P95 Latency |
The nearest-rank 95th-percentile latency of successful orders during the trailing seven-day window. |
| Errors / Orders |
The raw number of failed order requests and total order requests used in the calculation. |
| Timeouts / Probes |
The raw number of timed-out probes and total probes used in the calculation. |
| Risk Formula |
100 × [(60% × Error Upper 95% ÷ 1%) + (20% × Timeout% ÷ 1%) + (15% × P95 Latency ÷ 3 seconds) + (5% × Median Latency ÷ 2 seconds)]. |
| Time Window |
All statistics use the trailing seven days: up to 10,080 order attempts and 756,000 gRPC probes per server. |