---
title: "Track dashboard trends"
description: "Choose monitoring charts, inspect changes in traffic, and save a dashboard view for repeated checks."
sidebar:
  label: "Track dashboard trends"
seo:
  title: "Track Dashboard Trends | Judgment How-to"
  description: "Configure monitoring charts for behaviors, automations, usage, and latency, then save and verify a dashboard view."
---

Use the monitoring dashboard to spot changes across production traffic before
opening individual traces. This guide builds a repeatable view using the
standard monitoring charts.

## Open the dashboard and choose a period

You need recorded traffic in the selected project. Behavior charts also need
[judge evaluations](/documentation/judges/behavior-setup); automation charts
need [automation invocations](/documentation/monitoring/automations-and-alerts).
Adding a chart does not enable either workflow.

1. Open **Monitoring > Dashboard**.
2. Start from **Overview** in the view selector.
3. Choose a time window, such as **Past 24 Hours**. For a specific incident,
   use the custom date-range controls, choose **Between**, set **From** and
   **To**, and select **Apply**.

Keep the period consistent while comparing charts. A longer window can help
distinguish a recurring pattern from a single unusual execution.

## Add the charts that answer your question

Select **Add Chart**, choose a **Chart Type** and **Size** (**Full Width** or
**Half Width**), then select **Add Chart**. Available chart families include:

| Chart | Use it to | Useful controls |
| --- | --- | --- |
| **Behaviors** | Track detections of judge-defined outputs | In the chart menu, use **Filter by Category** or **Filter by Customer** |
| **Automations** | See how often automations trigger | Expand the chart to inspect more automations |
| **Behavior Relations** | Compare which behaviors occur together | Choose **Select Behaviors** and **Select comparisons**; compare behaviors with the same trace or session scope |
| **Tool Usage** | Find frequently called or failing tools | Click an error bar to open matching error spans |
| **Model Usage** | Compare recorded model costs and tokens | Switch **Cost** to **Total Tokens** or an input/output token metric |
| **Customer Usage** | Identify customers contributing to usage | Select a customer row and choose a cost or token metric |
| **Latency Analysis** | Locate slow executions or operations | Switch between **Trace Latency**, **LLM Generation**, and **Tool Latency** |

For a tool-failure investigation, start with **Tool Usage** and **Latency
Analysis**. Add **Behaviors** when a judge already detects the failure you
care about. Charts appear according to the current configuration; every family
does not need to be present.

Use a chart's three-dot menu to **Rename**, **Change Size**, **Change Chart
Type**, or **Remove from dashboard**. Drag the handle at the top of a chart
to reorder it. Removing a standard chart removes it from this configuration,
not the underlying traces or evaluations.

## Save the view

Save a named view to reuse your chart layout without changing the project's
shared starting layout:

1. Open the current view selector and choose **Save as new view** after making
   changes, or **New view**.
2. Give it a recognizable name, such as `Tool reliability`, and choose
   **Private** or **Shared** visibility.
3. Submit the form, then select the saved view to verify it.

Private views are available to their creator. Shared views are available to
others in the project; creating or changing one requires the **Developer**
role or higher. Only a saved view's creator can update or delete it. Use
**Save as new view** to adapt someone else's view.

To update your selected view, choose **Save changes** in the view selector.
Saving changes directly to **Overview** updates the project's shared starting
layout and requires an **Admin** or **Owner**. That built-in view cannot be
renamed or deleted.

Saved views retain chart order, sizes, names, and supported chart configuration,
including behavior filters and comparison selections. They do **not** retain
the time window. Temporary controls such as the cost/token metric, selected
customer row, and latency tab also reset; reselect them when needed. A view
saves configuration, not a snapshot of the chart data.

## Verify and investigate

After saving, reload, select the view, and check that the charts, order, and
saved filters return. Set the intended time window again before reading values.

- **Empty chart:** confirm the project and period, then check that the relevant
  spans, customer IDs, evaluations, or automation invocations were recorded.
  Clear chart-specific filters that may exclude the traffic.
- **Loading error:** retry loading the page. A failed request is not a zero
  count or proof that the agent is healthy.
- **Unexpected behavior counts:** confirm the judge's evaluation scope and
  sampling. Detections do not necessarily cover every trace in the project.
- **Cannot save:** check view ownership and your role. Use a private copy when
  you do not have permission to update a shared layout.

For example, click a **Tool Usage** error bar, inspect the resulting **Logs >
Spans** filters and time window, then open a span to examine its parent trace.
Continue with [Search production logs](/documentation/monitoring/logs) to
investigate the evidence behind a trend.