Rows not attached to a dataset or experiment - Braintrust

Documentation Index

Fetch the complete documentation index at: /docs/llms.txt

Use this file to discover all available pages before exploring further.

Applies to:

• Plan - Plans: Any
• Deployment - Deployments: Any
• Use case - Running EvalAsync with a parent context exported from braintrust.init() when dataset linkage needs to be recorded on the experiment

Summary

Issue: When using EvalAsync with a parent context, the experiment sidebar shows “Rows not attached to a dataset” and the dataset activity tab shows “No experiment that used this dataset was found.”
Cause: Passing parent=parent_ctx to EvalAsync prevents it from creating a new experiment, so dataset linkage is never recorded and the experiment created by braintrust.init() has no dataset parameter set.
Resolution: Pass dataset=dataset to braintrust.init() to register the dataset ID and version on the experiment.

Resolution Steps

If you are using EvalAsync with a parent context, pass dataset to braintrust.init()

Add dataset=dataset to your init() call. This records the dataset ID and current version on the experiment before EvalAsync runs.

dataset = braintrust.init_dataset(
    project_id=self.project_id,
    name=self.config.dataset.name,
)

experiment = braintrust.init(
    project_id=self.project_id,
    experiment=self.config.experiment.name,
    metadata=experiment_metadata,
    dataset=dataset,  # Required to link dataset to this experiment
)

parent_ctx = experiment.export() if experiment else None
await braintrust.EvalAsync(
    ...,
    data=dataset,
    parent=parent_ctx,
)

After this change:

• The experiment sidebar shows the linked dataset name and version.
• The dataset activity tab lists experiments that used the dataset.

If you need to pin to a specific dataset version

Pass version to init_dataset()

dataset = braintrust.init_dataset(
    project_id=self.project_id,
    name=self.config.dataset.name,
    version="<version_string>",
)

The version string is visible on the Datasets page. Experiments automatically pin to the dataset version at run time when dataset is passed to init().