> ## 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.

# Our Approach

> Learn how runtime execution, a connected context graph, and an adversarial agent swarm help Jina find deeper issues with fewer false positives.

# Our Approach

Most AI review tools read a diff once. Jina investigates the running system until it has evidence your team can act on.

## Run the code

* Executes focused tests, services, and API or UI checks.
* Recreates the state and conditions behind risky flows.
* Writes targeted probes when existing tests miss the path.

**Benefit:** Catch integration, state, and behavior bugs that static review cannot prove.

## Trace the real blast radius

* Connects code, infrastructure, runtime behavior, and past issues.
* Follows dependencies across services, contracts, and product flows.
* Finds non-obvious paths that ordinary code search misses.

**Benefit:** Focus testing on what is actually at risk instead of reviewing every file equally.

## Keep investigating

* Runs on Jina's long-horizon harness, not a single prompt.
* Uses a swarm of agents to plan, investigate, challenge, and review.
* Feeds evidence from each round into the next.
* Replans when it finds another dependency or failure path.

**Benefit:** Go deeper than a one-shot prompt or a single sandbox pass.

## What your team gets

* Deeper runtime bugs caught before merge.
* Fewer false positives and less review noise.
* Clear root cause, impact, and evidence for every issue.
* A merge score your team can use to make the call.

Next, follow a review from context to merge score in [How Jina works](/overview/how-jina-works).