> ## 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.
# Knowledge hub
> Where new findings from real work are written back, with their evidence, so knowledge compounds instead of evaporating.
The book and the bench capture what was already known when you connected your sources. The **knowledge hub** captures what you learn *after*: every deep dive, every A/B test read, every new definition discovered while doing work. It is where knowledge stops evaporating and starts compounding.
This is the surface you spend the most time on once knowledge is stood up, because it grows every time an agent or a teammate does something.
## What lives in the hub
As work happens, its outputs are written back as durable artifacts:
| Artifact | What it is |
| --------------- | -------------------------------------------------------------------------------------- |
| **Findings** | A specific thing you learned: "retention for new users in this region is low because…" |
| **Reports** | The write-up of an investigation: what you did, the evidence, the plots. |
| **Definitions** | A new or refined definition worth preserving: a metric, a segment, a term. |
| **Insights** | A durable observation about the system that should inform future work. |
## Every finding carries its evidence
A finding in the hub is never just a claim. Each one links back through an **evidence chain** to the task and agent that produced it: the run, the queries, the plots, the reasoning. This is what makes a finding safe to reuse: the next person can see exactly how it was reached and decide whether it still holds.
Treat the evidence chain as the finding's proof of work. A finding you can trace back to its run, you can build on; one you cannot, you have to redo.
## Validation states
Not everything written back is equally settled. Each artifact carries a state, so you always know how much to lean on it:
* **Validated**: checked and trusted.
* **Unvalidated**: produced by work but not yet confirmed.
* **To verify**: flagged as needing a look before anyone relies on it.
You can promote, edit, or retire any artifact, and see which agent and task created it and why it was kept.
## Re-scanning knowledge
Work spreads across many tasks and sessions, and the useful result is easy to lose track of. **Re-scan** pulls findings out of past tasks and back into the hub, so when you remember "last week a run found the five metrics correlated with six-month retention," you can recover it without hunting through a hundred open tabs.
## How knowledge compounds
This is the payoff of the whole module. Because [Agent Studio](/agent-studio/overview) runs agents in parallel, a single effort can launch dozens of investigations at once, and each one produces knowledge. Fifty investigations become fifty findings. Your job shifts from *doing* every investigation to *curating* what the fleet produces: keep what matters, validate what is load-bearing, retire what was a dead end.
The hub rewards curation. A pile of unvalidated findings is noise; a curated, validated set is the base every future investigation starts from. Decide what is worth keeping while the context is fresh.
Each kept finding is available to the next engineer and the next agent, which is how the system gets smarter about itself over time, instead of relearning the same things.
## Next steps
How to curate what compounds without drowning in findings.
See how the hub closes the loop with the book and the bench.