Catch illegal upstream joins, layer violations, and duplicate transformation logic in CI.
SQL linters check syntax and formatting in individual files, but they do not analyze the transformation graph. tff evaluates transformation models holistically—enforcing layer boundaries, domain contracts, and logic deduplication across the entire data warehouse DAG.
$ tff check
TFF ARCHITECTURE AUDIT
6 models · 4 errors · 4 warnings · 0.05s
STATUS LOCATION RULE COUPLING
──────────────────────────────────────────────────────────────────────────────
WRN models/core/layer_violation.sql duplicate_ctes algorithm
* CTE 'cleaned_users' duplicates transformation logic with 3 other models.
ERR models/core/users.sql banselectstar name
! SELECT * is prohibited. Explicitly name your columns to reduce coupling.
ERR models/core/users.sql nomissingowner metadata
! Model owner should always be specified.
WRN models/marts/finance/finance_stats.sql duplicate_ctes algorithm
* CTE 'cleaned_users' duplicates transformation logic with 3 other models.
ERR models/marts/marketing/marketing_all_users.sql layer_integrity dynamic
! marts/marketing depends on sqlmesh_example.finance_stats (marts/finance)
WRN models/marts/marketing/marketing_all_users.sql duplicate_ctes algorithm
* CTE 'marketing_cleaned_users' duplicates transformation logic.
ERR models/marts/marketing/marketing_type_violation.sql join_type_parity type
! Join condition 'o.user_id = u.user_id' compares integer with text (CoT).
──────────────────────────────────────────────────────────────────────────────
FAIL — 4 errors block merge. Run `tff --fix` for auto-correctable rules.
Run tff inside any existing dbt, SQLMesh, or Dataform repository. No configuration file required—tff automatically infers standard layer conventions (staging → intermediate → core → marts) and immediately audits your DAG:
# Instant zero-install invocation via uvx:
uvx --from "tff-core[dbt]" tff check # for dbt
# uvx --from "tff-core[sqlmesh]" tff check # for SQLMesh
# uvx --from "tff-core[dataform]" tff check # for Dataform
# Or install for your pipeline framework:
pip install "tff-core[dbt]" # for dbt (or: uv add "tff-core[dbt]")
# pip install "tff-core[sqlmesh]" # for SQLMesh
# pip install "tff-core[dataform]" # for Dataform
# Catch layer violations, duplicate CTEs, and circular dependencies
tff check
# Calculate your repository architecture health score (0–100)
tff health| Capability | Specification & Function |
|---|---|
| Layer Boundary Integrity | Enforces directional DAG constraints (staging → intermediate → core → marts) and blocks unauthorized cross-mart coupling. |
| Duplicate CTE Detection | Identifies identical transformation sub-queries across independent models (Connascence of Algorithm) to drive upstream consolidation. |
| Model Health Scoring | Computes an objective repository health metric (0–100) and gates CI builds via tff health --fail-under 80. |
| CI/CD Pipeline Governance | Official GitHub Action (tjirab/tff@v1) diffs PR changes against base branches, gates merges, and generates inline annotations. |
| Automated Remediation | Rewrites positional ordering clauses and provisions missing metadata scaffolds via tff lint --fix. |
| Multi-Engine Support | Native semantic analysis across dbt, SQLMesh, and Google Cloud Dataform projects without warehouse connection overhead. |
Complete specifications, configuration guides, and architectural documentation are available at tff.readthedocs.io:
- Getting Started & Integrations — Guides for dbt, SQLMesh, and Dataform.
- Rules & Checks Reference — Catalog of architectural checks and linter rules.
- CLI Reference Guide — Command-line interface reference, options, and output formats (SARIF, JSON, HTML).
- CI/CD & GitHub Actions — Automated PR governance with
tjirab/tff@v1and pre-commit hooks. - Extending tff — Authoring custom rules, checks, and plugins.
- GitLab Case Study — Architectural audit of 2,200+ models in enterprise dbt pipelines.
- Style Guide & Design System — Functional design principles, typography, and documentation standards.
Distributed under the MIT License.