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

# Runtime investigation

> Learn how Jina validates risky behavior by executing targeted probes in isolated environments.

# Runtime investigation

Runtime investigation is the point where Jina moves from a code hypothesis to behavioral evidence.

Unlike a static review agent, Jina can run targeted probes and simulations against the relevant parts of your application. The goal is not to execute every possible test. It is to validate the flows that the pull request and its blast radius make most important.

## What an investigation can include

* Tracing the source path behind a risky state transition or contract.
* Reviewing existing test intent to understand what is already covered.
* Creating a focused check for an uncovered failure hypothesis.
* Seeding the state needed to reproduce a risky flow.
* Executing the flow and comparing the observed result with the expected behavior.
* Iterating when the first result reveals a deeper dependency or alternate root cause.

## Where probes run

Probes run in ephemeral, isolated sandboxes against staging or preview deployments. Code, traffic, and traces are scoped to the review and the execution environment is torn down when the run completes.

Jina does not require permission to execute probes in production. It can read production signals through explicitly granted integrations, while active investigation remains isolated from the production environment.

## Execution plus judgment

A failed check alone is not automatically a reported issue. Jina reviews the evidence, checks whether the behavior is attributable to the pull request, and filters low-confidence or low-impact results before presenting the final report.

Continue to [Findings and merge score](/concepts/findings-and-merge-score).