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

# Review Results

> See your agent's latest scores, find which judge failed on which call and why, and compare runs after a change.

After a run finishes, start on the **Overview** tab to see how the agent scored, then open a run to find which judge failed on which call and why.

## Overview

Once a run has finished, the **Overview** tab shows the scores from the newest finished run:

![Overview tab with the Scorecard cards for Accuracy, Experience, Diagnostic flags, and the Default score of 75 out of 100, the What failed most list with Instruction Following and Voice Quality, and the Recent runs list](https://cdn.zeroruntime.ai/zrt/docs-resource/agent-builder/evals-overview.png)

* **Scorecard**: **A · Accuracy** and **X · Experience** as the share of scenarios that passed, **D · Diagnostic** as the number of flags raised, and the run's score with how many calls landed in each band. When there's an earlier run that's fair to compare with, each card shows the change since it.
* **What failed most**: the three scored judges that failed the most tries. Select one to open the run filtered to that judge.
* **Recent runs**: the last three runs. Select **All runs** to open the **Runs** tab.

If a newer run is still going, failed, or was stopped, a banner says so, and the scores stay on the last finished run.

## Runs

The **Runs** tab lists every run of this agent under **Run history**, with its status, when and how it started, and its accuracy, experience, and number of diagnostic flags. Filter by status, or search.

![Runs tab with the Run history table showing a finished run with its status, start time and trigger, accuracy, experience, and flags, the Status filter, the run search, and the Compare button](https://cdn.zeroruntime.ai/zrt/docs-resource/agent-builder/evals-runs-tab.png)

| Status | Meaning |
| - | - |
| **Starting** | The run is queued. |
| **Running** | Tries are being graded. |
| **Finished** | Every try has been graded. |
| **Failed** | The run ended without grading any try, so nothing was scored. |
| **Stopped** | You stopped the run before it finished. |

How a run started is shown next to its start time: **Manual**, **API**, **Auto run · live calls**, or **Auto run · simulation**.

## A run

Select a run to open it. The header shows the run's number and status, and its actions:

* **Stop run** while it's running. See [Stop a run](#stop-a-run).
* **Compare with run #N** when there's an earlier run that's fair to compare with. See [Compare runs](#compare-runs).
* **Re-score** on a finished run. See [Re-score a run](#re-score-a-run).
* **Run again** on a finished run, or **Try again** on a failed or stopped one. See [Run again](#run-again).

Below the header you see how and when the run started, how long it took, the version (branch) it tested, the scorecard, the number of scenarios, tries per scenario, total tries, and the judges.

![Finished run page with the Re-score and Run again buttons, the run details, the Scorecard cards, and the Every try, every judge grid showing pass and fail results for each judge, grouped by Accuracy and Experience](https://cdn.zeroruntime.ai/zrt/docs-resource/agent-builder/eval-run-detail.png)

### Scorecard

The same score cards as on the **Overview** tab, for this run. When each scenario was played more than once, switch between **Best of N tries** and **All N tries**. See [How a run is scored](https://docs.zeroruntime.ai/agent-builder/evaluations/overview.md#how-a-run-is-scored).

### Every try, every judge

A grid with one row per scenario and caller, and one column per judge, grouped by category. Each cell shows one try's verdict:

| Result | Meaning |
| - | - |
| **Pass** | The judge passed the try. |
| **Fail** | The judge failed the try. |
| **Can't tell** | The judge's topic never came up. Counts as not passed. |
| **Error** | The judge couldn't grade the try. Not counted. |
| **Skipped** | The try didn't count, for example because the call was too short. |
| **Not graded** | The judge hasn't graded the try yet. |

Each row also shows the try's score and band. Select a cell or a scenario name to open the try.

When you open the run from **What failed most** or from a judge, the grid shows only that judge and the scenarios where it failed. Select **Show every judge** to see the full grid again.

## A try

A try opens in a side panel. Use **Prev** and **Next** to move through every try in the run.

![Try panel for a live call with the Needs review status, 6 of 8 scored judges passed, the judge's summary of what went wrong, and the Verdicts tab showing failed Instruction Following and Voice Quality verdicts with their reasons and measured values](https://cdn.zeroruntime.ai/zrt/docs-resource/agent-builder/eval-try.png)

The top shows the try's status, its score and band, how many scored judges passed, and the caller persona. A try passes a category only when every judge in that category passes.

* **Verdicts**: every judge's verdict, failures first, with its reason. A judge that quoted a turn shows the quote and the turn number; select **See turn N in transcript** to jump to it. A code check shows the value it measured for each rule.
* **Transcript**: the full conversation.

Select **Open judge** to open the first scored judge that failed, or **Open in Simulation** to open a simulated call in [Simulation](https://docs.zeroruntime.ai/agent-builder/simulation/review-results.md#a-conversation), with its recording.

## Fix and re-run

**1. Find what failed**

Start from **What failed most**, open a failed try, and read the judge's reason and quote.

**2. Fix the agent**

Update the agent in the builder, then [commit](https://docs.zeroruntime.ai/agent-builder/versioning/drafts-and-commits.md) and [deploy](https://docs.zeroruntime.ai/agent-builder/deploy.md) the change.

**3. Run again**

Open the run and select **Run again**. Choose **Simulate again** so the same scenarios are played against your updated agent, and keep everything ticked so the two runs stay comparable.

**4. Compare**

Open the new run and select **Compare with run #N** to see what moved.

## Run again

**Run again** opens the **Run evaluation** panel limited to what the original run used: its branch, its scenarios and callers or its calls, its tries, its judges, and its scorecard. You can untick some, but not add new ones, so the two runs stay comparable.

![Run #1 again panel with Same calls selected, the one live call run #1 graded, only the judges run #1 used, and the same Default scorecard](https://cdn.zeroruntime.ai/zrt/docs-resource/agent-builder/eval-run-again.png)

For a run that graded simulated calls, choose:

* **Simulate again**: plays the same scenarios again on the branch as it's deployed now. Use this to test a change. Only offered when the original run graded **Scenarios**.
* **Same calls**: grades the same simulated calls again. Nothing new is simulated, so this doesn't test a change to the agent.

For a run that graded live calls, **Same calls** grades the same calls again.

## Compare runs

On the **Runs** tab, tick two or three finished runs and select **Compare**. The oldest run is the baseline. You can also select **Compare with run #N** on a run.

Runs can be compared only when they're fair to compare: they graded the same kind of calls (simulated or live), with the same judges at the same versions, the same scorecard, and the same scenarios × tries or the same calls. The agent's version can differ, since that's what you're testing. A run that isn't fair to compare is disabled, and hovering it says why.

> [!TIP]
>
> A run started with **Run again** is fair to compare with the original, as long as you leave every scenario, call, and judge ticked.

The compare page shows:

* **Scorecard change**: how accuracy, experience, and the score changed from the baseline.
* **By judge**: how often each judge passed in each run.
* **By scenario**: whether each scenario passed accuracy (**A**) and experience (**X**) in each run.

Switch between **Best of N** and **All N** to change how a scenario passes.

## Re-score a run

Select **Re-score** on a finished run and pick a scorecard. Each call's verdicts are scored again with it, so the scores and bands change, but the verdicts don't. Judges don't run again, so re-scoring costs nothing.

![Re-score panel with the scorecard picker set to Default and the Manage scorecards button](https://cdn.zeroruntime.ai/zrt/docs-resource/agent-builder/eval-rescore.png)

Re-score with the same scorecard after you edit it to apply your edits to the run.

## Stop a run

Select **Stop run** on the run page, or in the run's **⋮** menu on the **Runs** tab. The confirmation shows how many tries are done.

* Tries already graded keep their results.
* Tries not graded yet are discarded and not charged.


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