Features

Anomaly Detection

DeadPing learns what "normal" looks like for each monitor and automatically flags when something deviates. No thresholds to configure – it adapts to your job's actual behavior.

Anomaly detection in DeadPing automatically learns each monitor's normal behavior from its last 50 pings and flags deviations. It detects three types of anomalies: duration spikes (job running significantly longer or shorter), irregular frequency (pings arriving at unexpected intervals), and error rate surges (sudden increase in non-zero exit codes). No manual thresholds required. Pro plan and above.

Pro tier feature. Anomaly detection is available on the Pro plan and above ($12/mo+).

Three Types of Anomalies

Duration Anomalies

Detects when a job takes significantly longer or shorter than usual. If your nightly backup normally finishes in 45 seconds but suddenly takes 8 minutes, DeadPing flags it with a plain-English explanation like "Job took 10x longer than usual."

  • Warning at 3 standard deviations from baseline
  • Critical at 5 standard deviations
  • Requires exit code and duration tracking via the ping URL

Frequency Anomalies

Detects when pings arrive at unexpected intervals. If a job that usually pings every 5 minutes suddenly has a 30-minute gap, that could indicate a stalled job, a skipped run, or infrastructure issues.

  • Warning at 2.5 standard deviations from baseline
  • Critical at 5 standard deviations
  • Respects your configured schedule and grace period – a ping arriving late but within the grace window is never flagged

Error Rate Anomalies

Detects sudden spikes in non-zero exit codes. If your job goes from a 2% failure rate to 40% in the last 10 runs, something changed.

  • Triggers when recent error rate exceeds 3x the baseline rate
  • Requires at least 2 errors in the recent window to avoid noise
  • Critical severity when jumping from 0% to any errors

How the Baseline Works

DeadPing computes a statistical baseline from the last 50 pings for each monitor. This includes the mean and standard deviation of durations and inter-ping intervals. A minimum of 10 pings is required before anomaly detection kicks in for a new monitor.

The baseline adapts over time as your job's behavior changes. There are no static thresholds to maintain – the system learns what "normal" means for each individual monitor.

Cooldown

To prevent alert storms, each anomaly type has a 1-hour cooldown per monitor. If a duration anomaly is detected at 2:00 PM, the next duration anomaly for that monitor won't fire until 3:00 PM at the earliest. Frequency and error rate anomalies have independent cooldowns.

Alerts

Anomaly alerts are sent to all your configured channels (email, Slack, Discord, Teams, PagerDuty) with amber-themed formatting to distinguish them from red "monitor down" alerts. Each alert includes:

  • What was detected and how it compares to the baseline
  • Severity level (warning or critical)
  • The monitor name and a link to the dashboard

Dashboard

Anomaly events appear in the "Anomalies" tab on each monitor's detail page. Each event shows a plain-English headline, a visual comparison of detected vs. usual values, and an actionable explanation. You can acknowledge events to mark them as reviewed.

The main dashboard also shows an anomaly count for Pro and Business users and a banner when there are unacknowledged anomalies.

MCP Integration

AI assistants connected via MCP can use two anomaly tools:

  • list_anomalies – List recent anomaly events, optionally filtered by monitor or type
  • get_monitor_baseline – View the current statistical baseline (mean, stddev, sample size) for any monitor