---
title: "Alerts & Noise Protection"
id: "1960"
type: "page"
slug: "alerts"
published_at: "2026-09-30T22:50:45+00:00"
modified_at: "2026-10-01T00:22:56+00:00"
url: "https://xedant.com/agents/news/docs/alerts"
markdown_url: "https://xedant.com/agents/news/docs/alerts.md"
excerpt: "The most common reason people abandon news watching is not poor collection but a stream…"
---

# Alerts & Noise Protection

[https://xedant.com/agents/news/docs/alerts.md](https://xedant.com/agents/news/docs/alerts.md)

The most common reason people abandon news watching is not poor collection but a stream of notifications. After a week they stop reading them; after two weeks they turn them off. That is why noise protection here is on by default rather than “if you like”, and every suppressed message can be explained.

## What an alert is

An alert is a rule of the form “events of this importance — into these channels”. It does not decide what is important (that is the materiality rating’s job); it decides whom to wake up about it and how quickly. The same event can carry several alerts: one to your phone, another to the team’s work chat.

## Scope and conditions

Like a rule, an alert has a scope: which monitors, sources and tags it concerns. An empty scope means “everything”. A condition can be added on top — for example, “only events about our direct competitor”. Everything outside the scope and the condition simply never reaches the channels.

## Materiality bands

An alert has several bands — one per materiality band, each with its own set of channels and its own mode. This is the most practical way to separate noise from the important:

- **Noise and “worth knowing”** — into the digest only; they do not arrive as separate messages.
- **Material** — straight into the team’s work chat.
- **Existential** — straight to the phone, without waiting for the digest.

A band’s mode sets how to deliver: **immediately** or **into the digest**. An event without a materiality rating counts as zero — only the “from zero” band catches it.

## Noise protection

Five limiters work from day one — they do not need enabling:

- **No more than 3 alerts per hour** — when twenty events happen, you get three messages, not twenty.
- **A 30-minute pause on repeats** — the same change on the same monitor does not arrive twice in a row.
- **Identical changes — once per day** — for example, an exchange rate that trembles all day.
- **Burst waiting** — when one value jumps back and forth, the product waits and reports once, not on every step.
- **Quiet hours from 23:00 to 08:00** — night messages are **not thrown away** but deferred and arrive in the morning.

## Confirmation by a second source

A lone news item often turns out to be one site’s mistake. So an alert can carry a requirement: do not send the message until several independent sites confirm the news. Until the confirmation is there, the alert waits. When a second site writes about the same thing — the message goes out, and it shows that the confirmation arrived.

When the news gets confirmed even more strongly, you are not woken again: a new reminder arrives only when the count of independent sources has doubled. And when no confirmation came within a week, the message is not sent at all.

## The order of checks

The checks run in a strict order, and it is not a formality — the order decides which limiter gets named as the reason: first the repeat pause, then the identical-change protection, then burst waiting, then the rate limit, and last the quiet hours.

## The suppression journal

Every unsent message has a record with a reason: “repeat pause”, “quiet hours”, “already reported today”, “hourly message limit exceeded”. Silence is always explainable: when you expected a message and did not get it, the reason is two steps away, not “lost somewhere”.

## Digests

A digest is one scheduled roundup instead of a stream of messages. It is what makes news readable: you open one letter in the morning and see everything that happened over the day, arranged by stories, with materiality ratings and links.

- **How it differs from an alert:** an alert is about one event and speed; a digest is about everything over a period, and order.
- **What gets in:** events that passed the threshold; deferred night messages land in the morning digest too.
- **An empty period stays silent:** when nothing important happened in the window, the digest does not arrive — unless you yourself asked for empty ones too.

## What to do on overload

- **Look at the whole check schedule** — understand what is due to happen and when.
- **Pause the schedule** — a global pause stops the checks without resetting the settings.
- **Repeat suppressed deliveries** — when a channel was unlocked later, the piled-up messages can be re-queued.
- **Turn to the agent** — it sees the alerts’ posture and the held-message queue and can release them by hand.

Next → [Delivery Channels](/agents/news/docs/channels)

[← Back to the documentation index](/agents/news/docs)