Testing the semantic layer

Your metric definitions are now business-critical infrastructure. Treat them like it: test them. A test suite lets you refactor the compiler or tweak a metric and instantly know if any number moved.

What's worth testing

• Known totals. Total revenue is $743,772.80 for the seeded data. Pin it — if it ever changes unexpectedly, a definition drifted.
• Invariants. The sum of revenue across all channels must equal total revenue. Parts should sum to the whole.
• Refusals. An impossible metric×dimension combination (marketing spend by product category) should raise SemanticError, not return data.
• Ratios. AOV should equal revenue ÷ orders_count computed independently.

from semantic_layer import SemanticLayer, SemanticError

L = SemanticLayer()

def test_total_revenue():
    assert L.query(["revenue"])["rows"][0]["revenue"] == 743772.8

def test_channels_sum_to_total():
    total = L.query(["revenue"])["rows"][0]["revenue"]
    by_channel = L.query(["revenue"], ["channel"])["rows"]
    assert round(sum(r["revenue"] for r in by_channel), 2) == total

def test_impossible_combo_is_refused():
    try:
        L.query(["marketing_spend"], ["product_category"])
        assert False, "should have refused"
    except SemanticError:
        pass

Running the tests

pip install pytest
pytest test_semantic_layer.py     # pytest auto-discovers every test_* function

# or, with no new installs — add a __main__ runner to the file, then:
python3 test_semantic_layer.py

For the no-pytest route, the runner is one line at the bottom of the file: if __name__ == "__main__": test_total_revenue(); test_channels_sum_to_total(); test_impossible_combo_is_refused(). Plain asserts either way — and both commands assume your venv is active and you're in the starter-kit root, where semantic_layer/ is importable.