How Frontier works

Frontier finds the downstream business entities that may be stale after a dbt SQL change or a source data change, then proves a targeted recompute against a full reference. It reports aggregates and evidence. It does not claim a universal mathematical proof.

Billing is not currently enabled. Frontier is not autonomous self-healing and does not replace dbt, GitHub, Snowflake, orchestration, or your CDC platform.

1. dbt compiles the pull-request base and the PR head.
2. Frontier reads the dbt manifests and compiled SQL (frontier inspect / frontier compare).
3. Frontier classifies semantic SQL changes.
4. Supported changes compile into candidate-key SQL (COMPILED).
5. Candidate discovery executes inside the customer warehouse (EXECUTED).
6. Frontier recomputes only candidate business entities.
7. Base and head targeted results are compared.
8. Aggregates and evidence are uploaded to SaaS (frontier upload).

Warehouse rows and Snowflake credentials do not leave the customer environment. Unsupported or unsafe changes fail closed and require a full rebuild. See SQL-change assessments.

• It does not compile or orchestrate dbt for you outside CI helpers.
• It does not connect SaaS directly to Snowflake.
• It does not treat frontier run as equivalent to frontier prove. Customer CI uses prove.
• Source data changes use a separate CDC workflow. Consuming a stream is not the same as completing validation.
• Scheduled CDC is optional, commented out by default, and not autonomous repair.