> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rapidreview.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Learnings

> How Rapid Review learns from replies that decline its findings and uses those learnings in later reviews; read this when you want to know what a learning is, why Rapid Review replied "Learning recorded", what its reactions to your reply mean, or whether a learning was used.

## What is a learning?

A learning is a convention, or a fact about your code, that Rapid Review records from a reply that declined one of its findings. Later reviews of the repository take it into account, so the same kind of finding isn't raised again where the reason still holds.

## How does Rapid Review record a learning?

When someone with write access to the repository replies to one of Rapid Review's findings and declines it with a reason, for example that the behaviour is intended, or that it isn't a defect because of how the code works, Rapid Review can record that reason as a learning. The learning is anchored to a sentence of the reply.

Rapid Review announces it in the thread: "Learning recorded for later reviews of this repository:", with the learning, the paths it applies to, and how to correct or remove it.

If the reply repeats a learning recorded before, Rapid Review says so in the thread, and the thread becomes another source of that learning.

A finding that Rapid Review had already found fixed before the reply teaches nothing.

## What do Rapid Review's reactions to my reply mean?

Rapid Review leaves one reaction on each new reply it reads in its threads:

* **Thumbs up:** the reply changed something, such as a learning recorded, corrected, withdrawn, forgotten or kept. Rapid Review's written answer says what.
* **Eyes:** Rapid Review read the reply and nothing changed, as with "Fixed, thanks." or a question.
* **Confused:** reading the reply failed. The learning stays as it was, and you can reply again.

Edits and deletions of replies get no reaction.

## Whose replies does Rapid Review read?

Rapid Review reads replies in its own threads from the repository's owners, members and collaborators. It doesn't read replies from bots, from people without write access, or in repositories that record no learnings, such as repositories in the free tier.

## How are learnings used in reviews?

In standard mode, the reviewer is shown the active learnings that apply to the code it's reviewing, at most five for each part of the change. It's asked not to raise a finding of the same kind where the reason still holds, and to raise it where the code differs in a way that matters.

The check lists each learning shown under "Learnings shown", with a link to the thread it came from.

Budget mode and the free tier don't use learnings, and neither do the parts of a standard review that add only a few lines. Learnings are for later pull requests: on the same pull request, a finding isn't posted again anyway.

## Why does a learning wait for a person?

Some learnings aren't used until a person accepts them, such as a learning from a security finding. Its announcement then says "It waits for a person to accept it before reviews use it."

## Which repositories record learnings?

Every repository that Rapid Review reviews records learnings, except repositories in the free tier.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.