eval/test_api.py

"""API contract tests — run against the real precomputed artifacts.

Covers every route: response shape, parameter validation, 404s, cross-endpoint
consistency (the recommendation endpoint must agree with the raw forecast it is
derived from), and action-list invariants.
"""

import sys
from pathlib import Path

import pytest
from fastapi.testclient import TestClient

sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
from app.main import app

KNOWN = ("CA_1", "FOODS_3_090")


@pytest.fixture(scope="module")
def client():
    # context manager triggers the startup event that loads the parquet artifacts
    with TestClient(app) as c:
        yield c


class TestHealth:
    def test_health(self, client):
        r = client.get("/health")
        assert r.status_code == 200
        assert r.json() == {"status": "ok"}


class TestSkus:
    def test_shape(self, client):
        r = client.get("/api/skus").json()
        assert len(r["stores"]) == 10
        assert len(r["items"]) == 3049
        assert KNOWN[0] in r["stores"] and KNOWN[1] in r["items"]


class TestForecast:
    def test_arrays_aligned(self, client):
        r = client.get(f"/api/forecast/{KNOWN[0]}/{KNOWN[1]}")
        assert r.status_code == 200
        b = r.json()
        n = len(b["dates"])
        assert n == 28
        assert len(b["p10"]) == len(b["p50"]) == len(b["p90"]) == len(b["actual"]) == n
        assert len(b["history_dates"]) == len(b["history_actual"]) == 56

    def test_quantiles_ordered(self, client):
        b = client.get(f"/api/forecast/{KNOWN[0]}/{KNOWN[1]}").json()
        for lo, mid, hi in zip(b["p10"], b["p50"], b["p90"]):
            assert lo <= mid <= hi

    def test_unknown_series_404(self, client):
        assert client.get("/api/forecast/XX_9/NOPE").status_code == 404


class TestRecommendation:
    def test_defaults(self, client):
        b = client.get(f"/api/recommendation/{KNOWN[0]}/{KNOWN[1]}").json()
        assert b["service_level"] == 0.95
        assert b["lead_time_days"] == 7
        assert b["suggested_order_qty"] >= 0

    def test_inventory_reduces_order(self, client):
        base = f"/api/recommendation/{KNOWN[0]}/{KNOWN[1]}"
        q0 = client.get(base, params={"current_inventory": 0}).json()
        q100 = client.get(base, params={"current_inventory": 100}).json()
        assert q100["suggested_order_qty"] <= q0["suggested_order_qty"]
        # reorder point is independent of what's on the shelf
        assert q100["reorder_point"] == q0["reorder_point"]

    def test_higher_service_level_bigger_cushion(self, client):
        base = f"/api/recommendation/{KNOWN[0]}/{KNOWN[1]}"
        lo = client.get(base, params={"service_level": 0.90}).json()
        hi = client.get(base, params={"service_level": 0.99}).json()
        assert hi["safety_stock"] > lo["safety_stock"]

    def test_agrees_with_forecast(self, client):
        """Recommendation must be derivable from the served forecast (no drift
        between endpoints)."""
        fc = client.get(f"/api/forecast/{KNOWN[0]}/{KNOWN[1]}").json()
        rec = client.get(f"/api/recommendation/{KNOWN[0]}/{KNOWN[1]}").json()
        expected_avg = sum(fc["p50"][:7]) / 7
        assert rec["avg_daily_demand"] == pytest.approx(expected_avg, rel=1e-3)

    @pytest.mark.parametrize("bad", [{"service_level": 0}, {"service_level": 1.5},
                                     {"lead_time_days": 0}, {"current_inventory": -5}])
    def test_validation_422(self, client, bad):
        r = client.get(f"/api/recommendation/{KNOWN[0]}/{KNOWN[1]}", params=bad)
        assert r.status_code == 422


class TestActionList:
    def test_counts_sum_to_total(self, client):
        b = client.get(f"/api/action_list/{KNOWN[0]}").json()
        assert b["total_items"] == 3049
        assert sum(b["counts"].values()) == b["total_items"]

    def test_sorted_by_order_size(self, client):
        items = client.get(f"/api/action_list/{KNOWN[0]}?limit=50").json()["items"]
        qtys = [i["suggested_order_qty"] for i in items]
        assert qtys == sorted(qtys, reverse=True)

    def test_zero_inventory_means_everything_ordered(self, client):
        b = client.get(f"/api/action_list/{KNOWN[0]}?current_inventory=0").json()
        assert b["counts"].get("order_now", 0) + b["counts"].get("order_soon", 0) >= 3000

    def test_cached_response_identical(self, client):
        url = f"/api/action_list/{KNOWN[0]}?current_inventory=42"
        assert client.get(url).json() == client.get(url).json()


class TestBacktest:
    def test_summary_shape(self, client):
        b = client.get("/api/backtest_summary").json()
        for policy in ("naive", "recommended"):
            p = b["policies"][policy]
            assert 0 <= p["stockout_rate_pct"] <= 100
            assert p["avg_holding_units"] >= 0
        assert len(b["top_skus"]) == 20