View results and manage Tasks

Admins view results on the Task Detail Dashboard, review Submission Verdicts, and manage Tasks in Task Config.

1. View results on the Dashboard

①Metric cards

When you open a Task from the Task list, the Dashboard tab appears first. Admins perform the actions on this page.

Metric	Meaning
`Total Submissions`	Total number of submissions
`Attacks Succeeded`	Objectives attacked successfully at least once / total Objectives
`Jailbreak rate (ASR)`	Share of all attempts in which the attack succeeded

②Views

Below the metric cards, you can view results in two ways.

View	Columns	Use
`Individual`	`#`, `Tester`, `Objectives solved`, `Attempts`, `Successes`, `ASR`	Performance ranking by tester
`By Objective`	`#`, `Objective`, `Attempts`, `Succeeded`, `Jailbreak rate`	How often each Objective was broken

The dots in Objectives solved show success (green) or failure (red) for each Objective.
In Individual, you can sort by Attempts, Successes, and ASR.
Click a row to open the Submission details for that tester or Objective.

1-1Check the metrics
1. Check the total number of submissions in Total Submissions.
2. Check the number of broken Objectives in Attacks Succeeded.
3. Check the Jailbreak rate (ASR).
See Metric cards for what each metric means.
1-2View by tester or Objective
1. See the ranking by tester in Individual.
2. Switch to By Objective to see attack status by Objective.
3. Click the row you want to look at in detail.

2. Review Submissions and Verdicts

①Submission details

The left side of Submission details shows the full conversation between the tester and the model, and the right side shows the Verdict.

Field	Meaning
`Judge Type` · `Judge Model`	Scoring method and model
`Attack Success Threshold`	Number of Rubric items that must be met to count as a success
Final verdict	`Attack Succeeded` / `Attack Blocked`
`Attack Rubric`	`Criteria Met` / `Criteria Not Met` for each criterion
`Judge Reason`	Rationale for the verdict

②Override Verdict

If an admin thinks the automatic verdict is wrong, they change it to Succeeded or Failed in Override Verdict.

2-1Check the conversation and verdict
1. Click a tester row on the Dashboard.
2. Read the Objective and the full conversation on the left.
3. Check the final verdict on the right.
4. Check the rationale in Attack Rubric and Judge Reason.
2-2Change the verdict
1. If the verdict is wrong, click the correct option in Override Verdict.
2. Check the changed verdict.
See Override Verdict.

3. Manage in Task Config

①What you can edit

Manage how a Task runs in the Task Config tab.

Item	What you can do
Task name and description	Edit at any time
Status	Switch between `Active` and `Closed`. Change it with the switch at the top or with `Close Task` in the `⋯` menu.
Add Objectives	Add at any time with `New Objective`
Edit Objectives and Judge	Only when there are no submissions. Once there are submissions, `Cannot edit judge after submission.` appears.
Target Model	Can't be changed after creation
Delete Task	Click `Delete Task` and type `Delete Task` exactly in the confirmation field

When you close a Task, it disappears from testers' Annotation Queue and stops accepting new submissions. Deleted Tasks and submission results can't be restored.

3-1Change the status
1. Click the Task Config tab.
2. Switch between Active and Closed with the switch at the top.
3. You can also close it with Close Task in the ⋯ menu.
4. In the confirmation dialog, click Close Task.
Closed Tasks aren't visible to testers.
3-2Manage Objectives
1. Check the description and Judge on each Objective card.
2. Edit Objectives with no submissions using the pencil icon.
3. If there are submissions, a notice says you can't edit the Judge.
4. Add new goals with New Objective.
See What you can edit for what can be changed.
3-3Delete a Task
1. Click Delete Task at the bottom.
2. Type Delete Task in the confirmation field.
3. Click Delete.
Deletion can't be undone.

Frequently asked questions

Why can't I edit an Objective or Judge?Because the Objective already has submissions.

Cause: If the scoring criteria change midway, results can't be compared, so it locks as soon as there's a single submission.
Solution: Add a new Objective instead.

Can I change the Target Model?No. You can't change it after you create the Task.

Cause: All submission results must be based on the same model.
Solution: To evaluate a different model, create a new Task.

What should I do if the automatic verdict looks wrong?If you're an admin, choose the correct verdict with Override Verdict in Submission details. Otherwise, ask an admin.

Cause: LLM Judges can make mistakes too, so Override Verdict is provided.
Solution: In Submission details, choose the correct verdict with Override Verdict.

I closed a Task. Are the results still there?Yes, the results stay as they are.

Cause: Closing only blocks new submissions and hides the Task from testers' lists. You can still view the Dashboard and Submissions.
Solution: To accept submissions again, switch it back to Active.