> ## 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.

# Confidence tiers and noisy reviews

> How the confidence tiers low, medium and high decide which findings are shown, and other ways to make reviews quieter; read this when a finding is wrong, reviews are too noisy, there are too many comments, or you want to pick a confidence tier.

## What are confidence tiers?

Rapid Review estimates how likely each finding is to be real. A repository's confidence tier decides which findings are shown by that estimate:

* **`low`** shows the most findings. It's the default.
* **`medium`** shows fewer.
* **`high`** shows the fewest.

<Note>
  A higher tier shows fewer findings, and it drops more real findings along with the noise.
</Note>

## Which confidence tier is my repository in?

A repository is in `low` unless another tier is set for it, in its `.rapid-review.yml` or in its settings with Rapid Review.

## How do I pick a confidence tier?

Set `confidence` in the repository's `.rapid-review.yml`:

```yaml theme={null}
version: 1
confidence: high
```

Any tier can be picked. The file is read at the pull request's merge base, so the change applies to later pull requests once it's merged.

The file's tier isn't used when Rapid Review has set the repository's tier to take precedence over its file; the check's warnings then say so.

## What happens to a finding below the confidence tier?

A finding that falls below the repository's tier isn't posted: it isn't an inline comment, isn't listed in the review and isn't carried to later pushes. A finding Rapid Review couldn't estimate is kept.

## A finding is wrong: what can I do?

* **Reply to the finding with the reason.** When someone with write access replies in the finding's thread that it isn't a problem, or that the behaviour is intended, and says why, Rapid Review can record a learning. Later reviews of the repository in standard mode take it into account.
* **Resolve the thread.** A thread that a person resolved is never raised again on that pull request.

## Reviews are too noisy: what can I change?

* **Pick a higher confidence tier** with `confidence: medium` or `confidence: high` in `.rapid-review.yml`. Fewer findings are shown, and more real ones are dropped with the noise.
* **Leave paths out** with `review.exclude` in `.rapid-review.yml`, or limit reviews to some paths with `review.include`.
* **Write review instructions** in `.rapid-review/instructions.md`, for example about what kinds of issues not to raise.
* **Reply to wrong findings with the reason,** so Rapid Review can record a learning for later reviews.

Already built in: each review posts a limited number of inline comments, most severe first, and the check says how many lower-ranked findings went over the limit. A finding that's already on the pull request, or was resolved on it, isn't posted again.

## Can I hide findings below a certain severity?

A severity floor (`minor`, `major` or `critical`) holds back findings judged below it: they aren't posted, and the check lists them under "Below the severity floor". A floor, for a repository or for a whole account, is set on Rapid Review's side. You can't set one in `.rapid-review.yml`, and the Rapid Review app has no setting for it.

Without a floor, findings of every severity are posted.


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