Down Prediction Alerts
Most monitoring tools only alert you after a service goes down. Down Prediction is a proactive feature that watches for a consistent upward trend in response times and alerts you while the monitor is still technically up — giving you time to investigate and act before an outage occurs.
How It Works
After each check, the monitor maintains an in-memory ring buffer of recent response times. When the buffer is full, it runs a linear regression over the samples:
- Slope is positive — response times are consistently climbing, not just spiking randomly
- R² ≥ 0.6 — the increase is a real trend, not noise (coefficient of determination)
- Current response time ≥ 2× the 30-day baseline — the service has already degraded meaningfully
When all three conditions are met, a degradation warning is sent through all your configured notification channels (Slack, Discord, email, webhooks, etc.).
The analysis runs on every check at negligible cost — it’s pure in-memory math with no extra database queries. Only the notification is rate-limited (default: once per 15 minutes per monitor).
Prerequisites
Down Prediction is available on Growth, Business, and Enterprise plans.
If you’re on a lower plan, you’ll see a locked notice on the monitor settings page with a link to upgrade.
How to Enable
- Go to Monitors in your dashboard
- Open an existing monitor (Edit) or create a new one
- Scroll to the Down Prediction Alerts section
- Toggle Enable Down Prediction Alerts on
- Optionally adjust the Analysis window (see below)
- Save the monitor
Configuring the Analysis Window
The analysis window controls how many recent checks are used to evaluate the trend.
| Setting | Effect |
|---|---|
| 3–5 checks | Very sensitive — fires quickly but may produce more false positives on noisy endpoints |
| 10 checks (default) | Balanced — good for most monitors checking every 1–5 minutes |
| 20–30 checks | Conservative — only fires on sustained, long-running degradation |
A larger window gives you more confidence in the trend but means the alert fires later. For monitors checking every minute, a window of 10 = 10 minutes of trend data before alerting.
What Triggers an Alert
An alert fires when all of the following are true simultaneously:
- The last N response times show a consistent upward slope (positive linear regression)
- The R² value is ≥ 0.6 (the trend is not random scatter)
- The current response time is at least 2× your monitor’s 30-day average baseline
The baseline is automatically computed from your historical aggregation data and refreshed daily.
Alert Content
The notification includes:
- Current response time
- Your 30-day baseline average
- The trend slope (e.g. “+45ms per check”)
- A link to the monitor dashboard
No downtime has occurred when this fires — it is an early warning only.
Notification Cooldown
To avoid repeated pings during a degradation event, alerts are rate-limited to once per 15 minutes per monitor by default. This is not configurable in the current release.