Metrics & alerts

The Metrics page is a live telemetry dashboard for every flag evaluation your workspace serves. Use it to watch volume, performance and errors — and to get alerted when something crosses a threshold.

Choosing a time range

A range selector at the top switches between 24h, 7d, 30d and 90d. It controls both the window of data shown and how finely it is bucketed (hourly for 24h, daily for longer ranges).

KPI cards

The cards at the top summarise the selected range:

• Total evaluations — how many flag evaluations were served.
• Unique users — distinct user ids seen (your monthly-active-user figure).
• Error rate — the share of evaluations that failed.
• Latency — p50, p95 and p99 response times in milliseconds.

Charts

Below the cards:

• Evaluations over time — evaluation volume across the range.
• Top flags — the busiest flags by evaluation count, with a trend sparkline.
• Latency distribution — a percentile breakdown of response times.
• Status codes — counts of 200, 400, 401, 429 and 500 responses.
• Regions — geographic distribution of traffic, when your SDK reports it.
• Error timeline — which days saw errors, for spotting incidents.

Alerts

Alerts watch a metric and notify you when it crosses a threshold — so you don't have to keep the dashboard open.

Creating an alert

1. In the alerts section, click New alert.
2. Enter a name, choose the metric to watch, and set a threshold value.
3. Click Create. The alert is now active.

Creating alerts requires the admin workspace role.

Alert states

• OK — the metric is within its threshold.
• Triggered — the threshold has been crossed.
• Snoozed — a triggered alert you have temporarily silenced.

Thresholds are re-evaluated whenever the alerts list loads, so states stay current.

Snoozing

If an alert is triggered and you are already on it, snooze it to suppress further notifications for a while. The alert keeps evaluating; snoozing only quiets it.

Suggested workflow

• Skim the KPI cards daily — error rate and p95 latency are the most telling.
• Use the error timeline to correlate spikes with your own deploys or incidents.
• Set alerts on error rate and latency so regressions reach you without anyone watching the dashboard.