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

# Detector packs — which checks run on your project

> Detector packs choose which silent-failure checks run on your project. Owners and admins see why each pack is on, answer the questions Zespan asks, confirm what your tools do, and re-check recent traces.

**Detector packs** is in **Settings → Detector packs**. Each pack checks one kind of failure, for example a tool that moved money twice. A pack's findings appear on [Findings](https://docs.zespan.com/dashboard/findings.md) and in [Issues](https://docs.zespan.com/dashboard/issues.md).

> [!NOTE]
>
> Only owners and admins open this page. Everyone else sees a read-only note, and no request is made for them.

## The packs and why each is on

The packs list shows one row per pack: its name, what it checks, its **Mode** and how many **Checks** it holds (and how many of them are in preview).

| Mode | What it does |
| - | - |
| **Off** | The pack does not run |
| **Preview** | The pack runs, but its findings stay out of Issues until you turn it on |
| **On** | The pack runs and its findings appear in Issues |

A tag beside the mode says who chose it: **Chosen by Zespan** (from the tools in your traces, with a sentence that names them), **Chosen by you**, or **Default**. Open **Override** to force a pack **Off**, **Preview** or **On**. Your choice then stays. Turning a pack on can ask you to confirm first.

> [!NOTE]
>
> Packs are picked once enough tool calls arrive. A confirmation can change a pack only once enough tool calls have arrived; before that the outcome says no pack changed.

### The voice pack

The **voice** pack checks [LiveKit voice agents](https://docs.zespan.com/sdk/integrations/livekit.md#detectors). It is **Off** by default, and all four of its checks are in preview: they run and record findings, but they stay out of Issues and fix verification until Zespan releases them. They read stored numbers, never what was said.

| Check | Looks for |
| - | - |
| **Dead air** | A turn where the caller waited more than 3 seconds, or the agent never replied |
| **Repeated barge-in** | A call where at least 3 turns, and at least 30% of them, ended interrupted |
| **Tool loop in a call** | One tool called 5 or more times in a row in a turn, or returning an error 3 or more times |
| **Speech-to-text confidence drop** | At least 3 transcripts under 0.6 confidence, and at least 25% of those that reported one |

A check that does not have the number it needs says it can't tell and names the missing field, never a pass.

## Questions

When the agent layout is on, a **Questions** card sits above the packs list. Zespan chooses the questions, their order and their options from your own tool calls. Each one is about a tool or a field:

| Question | Buttons |
| - | - |
| Is this tool one that moves money (or deletes, sends a message, and so on)? | **Yes, it moves money** (or **Yes**), **No**, **Not sure** |
| Which fields identify one record? | A checkbox for each field, **Save**, **Not sure** |

* **Yes** confirms the tool's role. **No** rejects the suggestion. Each sends one request, even if you press twice.
* **Not sure** sends nothing. The question stays for later.
* **Save** needs at least one field. With none picked, nothing is sent.
* After an answer, the card's buttons go away and one line stays, in Zespan's words from what the server returned: *Saved. The Payments pack is now on.* It names a pack only when that pack changed. If none changed it says *Saved. No pack changed: packs are picked once enough tool calls arrive.* If the pack status could not be read it says *Saved. Pack status could not be read.* Keyboard focus moves to that line, and the line stays until your next answer.
* The same answer also refreshes **Confirm your tools**, so its row shows the new status. In the other direction, confirming a tool or a field in those cards refreshes the questions, so a question you just settled goes away.
* If a refresh fails after an answer, the page stays as it was with a note and **Retry**, and the outcome line keeps focus.

While the questions are loading, the card says *Reading your questions…* and does not say there are none. When the read worked and nothing is open it says *No questions right now.* If the read fails it says so and offers **Retry**. If a refresh fails after the questions were shown, they stay with a note that this is what was read earlier. If part of the read fails, the card says that too, and never reads as "none".

The same questions can also appear as an **Answer** item in [Needs you](https://docs.zespan.com/dashboard/needs-you.md#answer-items), and on the live step of setup.

## Confirming tools and fields

**Confirm your tools** tells a pack what each of your tools does. A confirmed role is used only by that pack. Pick the pack at the top right of the card.

* **Confirm** accepts the suggested role. **Change** picks another role. **Not this** rejects the suggestion.
* A role that can do harm (for example one that moves money) opens a dialog. It shows what confirming changes and an optional **Only when** row, for example `amount_inr > 100000`. Leave it empty to treat every call as risky.
* A tool whose shape changed since you confirmed it shows **Changed since you confirmed**.

**Fields** shows where in your tool calls each value a pack reads can be found. Edit the path and press **Confirm**.

## Re-check recent traces

**Re-check recent traces** reads traces sent before setup, or with older timestamps, from the last 1, 3 or 7 days, oldest first. It can run once an hour. Each run checks part of the window; run it again later to continue. Findings appear within a few minutes.

## Detector precision

When finding labels are on, owners and admins also see **Detector precision**. It counts how often your own Confirm and Reject labels said a detector was right. See [Findings](https://docs.zespan.com/dashboard/findings.md#detector-precision).

## Next steps

* [Findings](https://docs.zespan.com/dashboard/findings.md): what the packs produce, and how to label it
* [Needs you](https://docs.zespan.com/dashboard/needs-you.md): where unanswered questions wait for you
* [Annotation queues](https://docs.zespan.com/dashboard/annotation-queues.md): label sample runs to settle a cause
* [Navigation](https://docs.zespan.com/dashboard/navigation.md): where Detector packs sits in the sidebar


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