Skip to main content

Metadata Tables

Causal exports information that you set up in the Tools UI for use in your own analytics. The following sections describe these metadata tables.

Experiment Table​

The experiment table contains a record for each experiment that is running or has run in the past.

The table contains:

  • experiment_id: UUID that uniquely identifies the experiment.
  • name: Name of the experiment specified in the Causal Tools UI.
  • description: Description of the experiment specified in the Causal Tools UI.
  • state: either 'published' if the experiment is currently running, or 'done' if it has finished.
  • start_ts: Timestamp (UTC) when the experiment was first available to the front-end.
  • finish_ts: Timestamp (UTC) when the experiment finished, or null if the experiment is still running.
  • main_metric_id: Metric_id for the main metric expected to move during the experiment. See Metrics.
  • guardrail_metrics: List of metric_ids for guardrail_metrics assigned to the experiment.
  • audience_id: If this experiment is limited to an audience, its ID. Otherwise, null. See Audiences.
  • rolled_out_ts: If the experiment was rolled out to control, timestamp UTC when that occurred. Otherwise null.
  • rolled_out_variant_id: If the experiment was rolled out to control, the winning variant ID that was rolled out. Otherwise null.

Variant Table​

This table contains a record for each variant in each experiment (including control).

  • experiment_id: The UUID of the experiment.
  • variant_id: The UUID that uniquely defines this variant.
  • name: The name of the variant given in the Causal Tools UI. Note, for control this will always be null.

Variant_value Table​

This table contains a record for each feature value set in a variant. If a variant alters multiple feature values, there will be one row for each feature value.

  • experiment_id: The UUID of the experiment.
  • variant_id: The UUID that uniquely defines variant.
  • feature_name: The name of the feature containing the field that we are overriding in this variant.
  • field_name: The name of the field we are overriding in this variant.
  • value: The value we are setting the field to. For constants, it is a JSON record with as follows: {"constant": value}. For external outputs the record is: {"external": external name}. For Sagemaker model endpoints it is: {"sagemaker": sagemaker endpoint id}.

Audience table​

This table contains a record for each audience that you've saved in Causal.

  • audience_id: The UUID that uniquely identifies the audience.
  • name: Name of the audience specified in the Causal Tools UI.
  • description: Audience description specified in the Causal Tools UI.
  • state: 'published' if the audience is currently available, or 'archived' if it has been retired.
  • last_changed_by: The user who last edited the definition of this audience.
  • last_changed_ts: Timestamp (UTC) when the audience was last changed.
  • definition: The JSON blob that defines the audience in Causal's internal format. You should not rely on the format of this object. It is included here because we think it may be useful for some "grepping" type analysis.

Metric table​

Contains a record for each metric that has been saved in Causal.

  • metric_id: The UUID that uniquely identifies the audience.
  • metric_name: The name of the metric given in the Causal Tools UI.
  • state: 'published' if the metric is currently available, or 'archived' if it has been retired.
  • metric_description: The description of the metric given in the Causal Tools UI.
  • metric_def: The JSON blob that defines the metric in Causal's internal format. You should not rely on the format of this object. It is included here because we think it may be useful for some "grepping" type analysis.
  • external_metric: True if this metric is an external metric