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

# Best practices

> Four habits that keep system knowledge alive, correct, and compounding.

System Knowledge is not a one-time setup you run and forget. It stays valuable only if a few habits keep it accurate as your system and your team change. These are the four that matter most.

<Info>
  The thread running through all four: **correct knowledge is the bottleneck to making agents useful.** Every habit here is really about protecting correctness.
</Info>

## 1. Own the book

The [tribal book](/system-knowledge/tribal-book) is where you do the hard work once and reuse it for months. Read it, correct it, extend it, and treat it as yours, not Pavo's.

<Tip>
  **Why it pays off:** the hour you spend fixing a wrong join or an outdated metric definition is inherited by every future investigation. Get the book right and you stop paying that cost over and over.
</Tip>

## 2. Vet the bench

A generated [knowledge bench](/system-knowledge/knowledge-bench) is a starting point, not the finish line. Make the questions represent the gotchas your team actually knows.

<Tip>
  **Why it pays off:** the bench is what lets you and Pavo's agents optimize the knowledge: they work to close the questions you set. If the bench does not encode your real priorities, a high score means nothing. Early on, get the whole team in a room and build the question set together.
</Tip>

## 3. Use knowledge for work

Knowledge earns its keep when agents use it to do real investigations, and write what they learn back. Point work at the knowledge base and let the [knowledge hub](/system-knowledge/knowledge-hub) capture the results.

<Tip>
  **Why it pays off:** knowledge that is only read goes stale; knowledge that is used and extended stays current. Every task both consumes the base and adds to it.
</Tip>

## 4. Curate what compounds

Parallel agents produce findings faster than anyone can absorb them. Decide what is worth keeping, validate what is load-bearing, and retire the dead ends.

<Tip>
  **Why it pays off:** an uncurated hub is noise; a curated one is the trusted base every future investigation starts from. Curation is what turns volume into compounding.
</Tip>

## The habit loop

<Steps>
  <Step title="Correct">
    Fix the book where it is wrong.
  </Step>

  <Step title="Measure">
    Score the bench to see where knowledge still falls short.
  </Step>

  <Step title="Work">
    Run investigations that use, and extend, the knowledge.
  </Step>

  <Step title="Curate">
    Keep and validate what matters, so the base gets stronger each cycle.
  </Step>
</Steps>

Done consistently, these four turn System Knowledge from a static document into a base that gets more accurate and more valuable every week.

## Next steps

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/system-knowledge/quickstart">
    Run the loop end to end for the first time.
  </Card>

  <Card title="Overview" icon="book-open" href="/system-knowledge/overview">
    Revisit how the four artifacts fit together.
  </Card>
</CardGroup>