# Precision Mode (/docs/knowledge/retrieval/precision-mode)
Most systems only start to understand a document when someone asks about it. Precision Mode does that work when the document arrives. Docklee reads each source in depth, writes down what it says and what it answers, and keeps those findings ready. When a question matches, the engine answers straight away, without writing a new answer.

<Callout type="info" title="Turn on Precision Mode when...">
- People ask the same kinds of questions about the same content.
- Response time matters, such as in chat or voice.
- Your documents relate to, update, or contradict each other.
</Callout>

## What's prepared [#whats-prepared]

For each document, Docklee prepares four kinds of finding:

| Finding | What it holds |
| --- | --- |
| Summary | What the document is about, in one paragraph. |
| Facts | Standalone claims the document makes, each one checkable on its own. |
| Implications | Conclusions that follow from the document but aren't written in it. |
| Questions answered | The questions the document answers, with their answers. |

Docklee then compares each document with the documents related to it. See [Cross-document reasoning](/docs/knowledge/retrieval/cross-document-reasoning).

## How it works [#how-it-works]

1. **Prepare.** As each source is added to the container, including sources you add later, Docklee prepares its findings.
2. **Match.** When a question arrives, the engine checks it against the prepared findings.
3. **Answer.** A close, high-confidence match is returned at once. Anything else goes through the normal search and answer steps.

A prepared answer comes back with its confidence and no citations. In a stream, it arrives whole in the `final` event, with no `token` events. See [Streaming answers](/docs/knowledge/streaming-answers).

## Cost [#cost]

<Callout type="warn">
Precision Mode uses generation credits for each source it prepares. Turning it off keeps what's already prepared, so you can turn it back on later at no extra cost.
</Callout>

## Turn it on [#turn-it-on]

1. Open **Sources** in the [Docklee dashboard](https://app.docklee.com).
2. Open the container.
3. Turn on **Precision Mode** and confirm.

Precision Mode is set per container, so every engine that reads the container's sources benefits. When a source changes, its findings are prepared again.

## Safeguards [#safeguards]

- Documents longer than 50,000 characters are prepared from their first 50,000 characters.
- Findings are checked against a strict format, and malformed output is discarded.
- Only document content is sent for preparation, never user details or metadata.
- Findings are encrypted the same way as the rest of your knowledge.

## Next steps [#next-steps]

Common questions are answered from prepared findings. See how documents are compared with each other next.

<Cards>
  <Card title="Cross-document reasoning" href="/docs/knowledge/retrieval/cross-document-reasoning" description="How documents learn about each other." />
  <Card title="Answer cache" href="/docs/knowledge/answers/answer-cache" description="Instant answers to repeated questions." />
  <Card title="Citations and confidence" href="/docs/knowledge/answers/citations-and-confidence" description="How answers are checked." />
</Cards>
