Anomaly Detection and Alerting

The platform turns detected data anomalies into action automatically. This page describes the end-to-end path; each step links to its detailed documentation.

1. Detect — rules annotate suspect data

Anomaly detection happens in rules. A validation rule (standard or custom Python) inspects the data flowing through and annotates the datapoints that fail its check — a spike, a gap, a value exceeding contracted capacity, a decreasing register read. Annotations stay attached to the datapoints and are visible on every chart.

Rules can also call deployed machine learning models for model-based detection.

2. Escalate — annotation thresholds create tasks

Every rule that generates annotations should carry an AlertThreshold parameter (see Rule configurations). When the number of annotations produced by the rule crosses the threshold, the platform creates a task so a person can follow up. Rules can also create tasks directly from code for full control over task content — see Task creation.

Tasks land on configurable boards in Task Management, where they are prioritised, assigned and resolved.

3. Act — corrective flows and notifications

Detection can trigger automated processing, not just human follow-up:

• Corrective flows — a flow chains a follow-up flow, optionally selected at runtime by a Decision Tree; estimation and editing rules in later sequences correct what validation rules flagged (the VEE pattern).
• Coordinated reactions — the Collective Trigger starts a downstream flow once related messages have accumulated.
• Email notifications — rules send templated emails to configured recipients, per rule execution or aggregated per flow. See Sending Emails.

Example: consumption spike, end to end

1. Standard_Interval_Spike_Validation annotates datapoints deviating beyond the configured percentage.
2. Its AlertThreshold is set to 1 → a task appears on the Critical_Validation_Errors board.
3. The flow's next sequence runs estimation rules that replace the flagged values, and the corrected series continues downstream.
4. An operator reviews the task, confirms the correction (or adjusts the rule's threshold), and closes it.

Watching the process itself

Besides data anomalies, the platform can also alert on process anomalies — files that fail transformation, flows that error, throughput that drops. Process Monitoring tracks per-process KPIs and sends email alerts when a metric crosses a configured threshold on a schedule you define.