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

# Fix, verify, compare, and export

> After a BotGauge red-team run: what to do with each result, verify a fix by replaying the attack, rerun errors, compare versions, and export reports.

## What to do with each result

| You see | Do this |
| - | - |
| Confirmed finding, critical or high | Fix first. Use the evidence to reproduce it, then use **Verify fix**. |
| Confirmed finding, medium | Fix next. Add traces or a verifier for stronger evidence if you think severity is understated. |
| Signal | Read the evidence. If real, fix it. If you can add visibility (grey box), rerun to firm it up. |
| Not evaluable or Couldn't verify | Read the reason and unlock it: add the wrapper, provide the system prompt or test data, or fix a missing key. Then rerun. |
| Error | Fix the cause, for example update an expired secret, and use **Rerun errors**, which re-runs only the failed attacks. |
| Held | Keep it. Re-test after significant changes to your agent. |

## Verify fix

After changing your agent, open the finding and choose **Verify fix**. BotGauge replays the exact attack that worked. If the replay no longer breaks the agent, the finding becomes **fixed (verified)**. If it still breaks, it is marked **regressed**.

"Not seen again" in a later run is a good sign but isn't proof of a fix. Only a verified replay counts.

You can also minimize a finding to the shortest attack that still works, which is helpful when you hand it to a developer.

## Accept a risk

A finding can be given a disposition, or an exception with a reason. Critical findings can't be waived.

## Re-run

If you changed the agent's prompt, tools, or behavior, rerun recon, review the new plan, then approve and start. If only some attacks errored, use **Rerun errors**. This starts a continuation that keeps everything unchanged and re-runs only the failures. Only one run can happen at a time per agent.

## Agent health

| Status | Meaning |
| - | - |
| Not tested | No run yet. |
| Needs attention | There are open, demonstrated findings. |
| Healthy | Tested, with a share of attacks held and nothing demonstrated open. |
| Inconclusive | Nothing could be graded. |

Health is never shown as "0% held" when nothing could be graded. It is shown as not available.

## Export and share

You can export a run report as JSON, CSV, or PDF through the API, and the [compliance view](https://www.botgauge.com/docs/red-teaming/compliance.md) as CSV or JSON. The console's **Share report** button is not available yet, so export from the API instead. CSV exports neutralize cells beginning with `=`, `+`, `-`, or `@`, so opening them in a spreadsheet is safe.

Replay attacks to confirm fixesUse Verify fix to confirm a patch and track health in [BotGauge](https://cloud.botgauge.com).

[Verify a fix →](https://cloud.botgauge.com)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.