New check

Create a New Check

Follow these steps to add a new check to dbt-bouncer.

1. Determine Check Location

• Category: manifest, catalog, or run_results?
• Resource type: model, source, seed, exposure, macro, etc.?
• File: place in the appropriate submodule under src/dbt_bouncer/checks/<category>/.

2. Write the Check

Use the @check decorator, passing the rule code. Everything else is inferred from the function signature:

from dbt_bouncer.check_framework.decorator import check, fail

@check(code="XX000")
def check_model_xxx(model):
    """Check description."""
    if some_condition:
        fail(f"`{model.unique_id}` failed because ...")

3. Decorator API Reference

code is the only argument @check takes. All other metadata is inferred from the function signature:

• code — the check's unique rule code, e.g. MO048. See "Assign a rule code" below.
• name — the function name (must match the name: value in YAML config).
• iterate_over — the first positional parameter (excluding ctx). If there are none, the check is global (runs once with context only).
• params — keyword-only arguments (after *) become user-configurable Pydantic fields.
• ctx — optional; only include in the signature if the function actually uses it.
• Parameter ordering — must be (resource, ctx, *, params). Resource first, ctx second. Putting ctx before the resource breaks iterate_over inference. For context-only checks, use (ctx, *, params).

Simple check (resource only)

@check(code="MO021")
def check_model_description_populated(model):
    """Models must have a populated description."""
    if not model.description or len(model.description.strip()) < 4:
        fail(f"`{model.unique_id}` does not have a populated description.")

Check with params

@check(code="MO038")
def check_model_names(model, *, model_name_pattern: str):
    """Models must have a name matching the supplied regex."""
    import re
    if not re.match(model_name_pattern, model.name, re.IGNORECASE):
        fail(f"`{model.unique_id}` does not match pattern `{model_name_pattern}`.")

Context-only check (no resource iteration)

@check(code="MO044")
def check_model_test_coverage(ctx, *, min_model_test_coverage_pct: float = 100):
    """Set the minimum percentage of models that have at least one test."""
    ...

fail() — raises DbtBouncerFailedCheckError

fail("message")

Assign a rule code

Every check needs a unique rule code: a 2-letter resource prefix plus a 3-digit number, e.g. MO048. Two steps:

1. Pass it to the decorator: @check(code="MO048").

2. Add the matching member to the resource's *RuleCode enum in src/dbt_bouncer/enums.py, keeping alphabetical order:

   class ModelRuleCode(StrEnum):
       CHECK_MODEL_XXX = "MO048"
   

Use the next free number for the prefix — read the enum to find it. Never reuse or renumber a published code; users reference codes in their config.

Prefixes: CA catalog, EX exposure, LI lineage, MA macro, ME metadata, MO model, RR run results, SE seed, SM semantic model, SN snapshot, SO source, TE test, UT unit test.

4. Register the Check

• Add the check to dbt-bouncer-example.yml
• Validate: dbt-bouncer --config-file dbt-bouncer-example.yml
• Ensure alphabetical ordering is maintained

5. Write Tests

Use check_passes / check_fails from dbt_bouncer.testing:

from dbt_bouncer.testing import check_fails, check_passes

def test_pass():
    check_passes("check_model_xxx", model={"name": "valid"}, my_param="value")

def test_fail():
    check_fails("check_model_xxx", model={"name": "invalid"}, my_param="value")

# For context-dependent checks:
def test_with_context():
    check_passes("check_model_xxx",
                 model={"name": "m1"},
                 ctx_models=[{"name": "m1"}, {"name": "m2"}])

• Resource dicts are auto-merged with sensible defaults (no fixture setup needed)
• ctx_* kwargs build the CheckContext automatically
• Include at least one happy path and one unhappy path test
• Ensure __init__.py exists in the test subdirectory

6. Verify

mise run generate-schema
mise run generate-rule-codes-doc
mise run test-unit
prek run --all-files

The rule-codes-doc-check hook fails if a check has no code, if a declared code is unused, or if docs/checks/rule_codes.md has drifted.