One canonical graph of your analytics layer — compiled from SQL, dbt, and warehouse metadata you already have.

ClearMetric merges warehouse metadata, dbt, and SQL into a single graph with physical bindings, derivation state, and explicit warnings when something cannot be resolved. Use it for lineage, impact analysis, catalog output, structural checks, and OpenLineage export — all from the same compiled graph.

pip install clearmetric-core
cd examples/lineage-demo
cm scan
cm compile --format json > graph.json
cm compile --format catalog > catalog.json
cm impact orders_base.amount --downstream
cm clean
cm contract graph.json

Warehouse metadata is a local INFORMATION_SCHEMA JSON export — not a live connector.

If cm is occupied on your PATH: python -m clearmetric.cli --project-dir . …

Status: early development (0.x). Pin your versions. Docs: clearmetric-labs.github.io/ClearMetric-Core · Architecture: docs/public-architecture.md · Validation

• Lineage + impact — upstream/downstream column traversal; answer what breaks if I rename this column?
• One graph — warehouse, dbt, and SQL merged with physical bindings on lineage nodes
• Catalog — compile --format catalog for table/column/model nodes
• OpenLineage — compile --format openlineage for interop with DataHub, Marquez, etc.
• Cleaner + security floor — structural checks and schema drift warnings; compile fails closed on security policy errors (see docs/limitations.md)

Start with the self-contained examples/lineage-demo project (SQL pipeline + warehouse JSON).

pip install clearmetric-core
cd examples/lineage-demo
cm scan
cm compile --format json > graph.json
cm impact orders_base.amount --downstream
cm clean
cm contract graph.json

Column selections accept orders.amount, column:orders.amount, or column.fct_orders.net_revenue (normalized via clearmetric.core.ids.parse_column_selection).

Adapters normalize sources in; the core merges and binds; validation runs through the cleaner and security floor; emitters shape output.

from pathlib import Path
from clearmetric.compiler import build_graph, check_graph, compile

compiled = compile(Path("./my-project"))          # build + enforce
built = build_graph(Path("./my-project"))           # build pipeline without enforce
report = check_graph(built.artifact, posture=built.project.posture)  # report-only (same path as cm clean)

One install (pip install clearmetric-core) — Python subpackages, not separate PyPI packages.

Static analysis for SQL/dbt lineage; warehouse metadata exports only in the public CLI. ClearMetric does not connect to live warehouses or execute production queries. On star-heavy SQL (SELECT * without schema), ClearMetric flags what it cannot resolve. See docs/limitations.md (pointers to Validation and the resolver spec) and the Docs site.

• Bugs / features: Issues
• Questions: Discussions

Apache 2.0.

python -m pip install -e "packages/clearmetric-core[dev,runtime,release]"
python -m pytest -v packages/clearmetric-core/tests tests/