How the platform exposes its internal state.
Counters track totals, gauges track current values, and histograms track distributions. All three carry the same labels: service, region, and shard.
Every request receives a trace id at the edge. The id propagates through every hop and is attached to logs, spans, and any error returned to the caller.
| Level | When to use | Retention |
|---|---|---|
| debug | local development | 1 day |
| info | routine operation | 14 days |
| warn | recoverable anomalies | 30 days |
| error | failed operations | 90 days |
The golden dashboard pairs traffic with latency and errors; every service must appear on it before it takes production traffic.