Skip to main content

Save and reuse check metrics

A check metric is one requirement used to evaluate a result, such as “name the source for each key fact.” Save it in Library → Check metrics when you want to reuse it across tasks. Each task gets its own copy, which you can adjust to fit that task.

To decide which requirements belong in one task, see Set acceptance criteria. To understand the results of checking them, start with How checks work.

Open and browse the metric library​

Open Library from the left navigation's … menu, then choose Check metrics. You can also reach it through Manage metric library when choosing metrics for a task.

Use the search box to find a metric by name or requirement. The views are:

ViewWhat you can do
PresetBrowse ready-made checks for quality, text requirements and tool use
CustomCreate, edit and delete metrics you saved
StarredFind preset or custom metrics you marked with a star

Starring a metric makes it easier to find. It does not add the metric to any task or turn on a check.

Create a reusable metric​

  1. Click Create custom metric.
  2. Give it a short Metric name that you will recognize later.
  3. Write the Criterion: what must be true for the requirement to be met.
  4. Choose an Evidence type, then save the metric.

For example:

FieldExample
NameSources are named
CriterionEach key factual claim names its source or includes a source link. Information not yet verified is identified as unverified.
Evidence typeOutput text

This example checks whether sources and uncertainty are stated in the delivery. Verifying that a source actually supports a claim needs the source content or other supporting evidence too.

Choose the evidence that can answer the question​

Evidence typeUse it when the requirement concerns…
Output textThe words, structure or content actually delivered
Tool recordsRecorded tool calls and their results
Artifact or external stateA produced file or a change outside the text reply

Choosing an evidence type says what the check needs. It does not itself provide the evidence or grant access to a file or service. A check that cannot establish the requirement from the available evidence may need your review.

You can also keep a useful metric from a task: in the task's acceptance criteria, use Save as custom metric. The library entry and the task's metric remain separate copies.

Use a metric in a task​

  1. Open the task's Checks tab and find Acceptance criteria.
  2. Click Choose from metric library. Search or browse presets, custom metrics or favorites, then select the metrics you need.
  3. Click Done and review the copies in the task. Fill in any details a rule needs, such as the text to match, a length range or a tool name.
  4. Save the acceptance criteria to apply the changes.

Each task supports up to 10 metrics. Choose requirements that help you decide whether this task's delivery is usable. Selecting a metric or closing the picker does not replace saving the task's criteria.

What happens when you edit or delete one?​

ActionEffect
Edit a custom metric in LibraryChanges the version available to add to tasks in the future; existing task copies stay as they are
Edit the copied metric in a taskChanges that task's draft; save its acceptance criteria to apply it. The library and other tasks stay unchanged
Delete a custom metric from LibraryRemoves it from your library and favorites; copies already used in tasks and existing reports remain
Remove a metric from a taskRemoves it from that task after you save the criteria; the library entry and earlier reports remain

For example, you add “Sources are named” to two tasks, then make the library version more specific. Neither task changes automatically. Update each task's copy if you want that task to use the new wording.

Next steps​