Lineage coverage for Atlan's dbt and Snowflake customers fell from 92% to 78% after a release. What went wrong?

Introduction

Lineage is one of Atlan's core promises. For customers who use dbt for transformations and Snowflake as their warehouse, Atlan stitches lineage from source tables through dbt models to downstream dashboards, at column level. In this scenario, the share of Snowflake columns with complete upstream and downstream lineage for dbt-plus-Snowflake customers fell from 92% to 78% within two weeks after a product release. Several customers have opened support tickets saying lineage graphs "stop" at certain models.

Lineage coverage is a trust metric. Engineers use lineage for impact analysis, governance teams use it for compliance, and AI agents increasingly use it through Atlan's MCP server. If lineage silently breaks, people make decisions on incomplete information. The task is to find the root cause.

Step 1

Clarifying Questions (3 minutes)

• How is lineage coverage calculated?

• Hypothetical answer: Share of active Snowflake columns (queried in the last 30 days) that have at least one upstream and, where applicable, one downstream lineage edge.
• Impact: Changes in what counts as "active" could also move the metric.

• Is the drop across all dbt-plus-Snowflake customers or a subset?

• Hypothetical answer: Concentrated in about 30% of those customers; others are unchanged.
• Impact: Something specific to that subset.

• What changed in the release?

• Hypothetical answer: A new version of the SQL parser used to derive lineage from query history and dbt manifests, and an update to the dbt connector supporting newer dbt features.
• Impact: Both are candidates.

• Did anything change on the customer side, such as dbt version upgrades?

• Hypothetical answer: Unknown; we need to check.