Dataset lineage
The Dataset lineage view describes the lineage graph for an onboarded dataset.
Use it to understand which datasource produced a dataset, which raw inputs contributed upstream, and how ingestion and onboarding steps connect the data assets together.
When to use
Generate the view
Generate the Dataset lineage view from a dataset.
The request only needs view_type set to DATASET_LINEAGE.
Output
The response follows the shared Response structure format.
The Dataset lineage view returns multiple named dataframes inside view_data.
Interpret the result
The Dataset lineage view is easiest to read as a graph: start with the dataset node, then follow the edges back to the datasource and raw inputs.
Start with the dataset node
Use nodes to find the row where node_type is DATASET.
This row describes the onboarded dataset and includes dataset identifiers, dataset name, row counts, split counts, masked rows, increment metadata, and size information.
Trace the datasource
Use rows where node_type is DATASOURCE to identify the datasource that produced the dataset.
The datasource node helps connect the onboarded dataset back to the ingestion source. This is useful when auditing which datasource version or branch was used to create a dataset.
Identify raw inputs
Use rows where node_type is RAW to identify raw input nodes.
Raw nodes represent upstream input files or raw data sources that contributed to the datasource, and therefore indirectly to the dataset.
Follow edges
Use edges.link_from and edges.link_to to understand the direction of lineage.
For dataset lineage, the graph commonly includes an ONBOARDING edge from datasource to dataset, and INGESTION edges from raw inputs to the datasource. source_branch and target_branch describe the branch relationship when branch information is available.
Review increments and branches
Use increment_label, increment_index, branch, and source_branch to understand how dataset and datasource versions or increments relate to each other.
This is useful when a dataset has been created from a specific branch or when you need to understand which increment produced the dataset currently being used for training or review.
Check data volume
Use n_rows, n_rows_train, n_rows_test, n_rows_masked, n_input_files, increment_size_mb, increment_mask_size_mb, and total_increment_size_mb to review the size and split of each lineage node.
These fields are useful for verifying that the dataset contains the expected amount of data before training, evaluation, or model review.
Field reference
The Dataset lineage view returns dataframes, so individual columns are not documented as standalone API schema properties.
nodes
The nodes dataframe contains the graph nodes.
edges
The edges dataframe contains the graph edges.