White paper · 27 August 2026

Alerts with reasons: what an explained alert should contain

Key points

  • Alert fatigue is a design problem: too many alerts with too little context teach people to ignore them.
  • An explained alert states what was seen, what it was compared with, and which signals mattered most.
  • Explanations need testing like any other part of the system.
  • Open platforms make it easier to inspect and audit the rules and models behind an alert.

The problem

Monitoring systems raise alerts when a reading looks unusual. When the cause is not shown, the operator must investigate from scratch, and the cost of a false alarm is the same as the cost of a real one. After enough false alarms, alerts are muted. The value of the system is then lost precisely when it matters.

What an explained alert contains

  1. What was observed. The signal, the asset and the time, in the operator's terms.
  2. What it was compared with. A normal range, a recent baseline, or a similar period, shown on the same chart.
  3. Why it was flagged. For a rule, the threshold crossed. For a model, which inputs contributed most to the score.
  4. How sure the system is. A simple indication, so that borderline cases are handled differently from clear ones.
  5. What to check next. A short, specific suggestion, written with the people who respond.
  6. A record. Who acknowledged it, what was decided, and whether it turned out to be right.

Rules and models together

Simple rules are easy to explain and often enough. Models can detect patterns that rules miss, but they need an explanation layer. The practical approach is to keep both, show which one fired, and compare their performance over time using the records from point 6.

Testing the explanation

Replay past incidents and ask the operators who handled them whether the explanation would have helped. Check that the explanation changes when the input changes in the way you expect. Check that it is stable for near-identical inputs. If operators disagree with it often, treat that as a finding about the explanation or the model.

Why an open foundation helps

When a monitoring platform is built on open-source components, the rules, data model and integrations can be inspected, and the system is easier to move or extend. That does not make an alert right, but it makes the reasoning behind it easier to audit.

Limits

An explanation can be clear and still wrong. Keep human review in the loop for decisions with safety or legal consequences, and measure how often alerts lead to action.

General guidance. It does not report results from any client deployment.

All white papers