query,doc_id,rating
"running shoes",sku-1234,3
"running shoes",sku-5678,2
"running shoes",sku-9999,0
"red dress",sku-2222,3

POST products/_rank_eval
{
  "requests": [
    {
      "id": "running_shoes_query",
      "request": {
        "query": { "match": { "name": "running shoes" } }
      },
      "ratings": [
        { "_index": "products", "_id": "sku-1234", "rating": 3 },
        { "_index": "products", "_id": "sku-5678", "rating": 2 },
        { "_index": "products", "_id": "sku-9999", "rating": 0 }
      ]
    },
    {
      "id": "red_dress_query",
      "request": { "query": { "match": { "name": "red dress" } } },
      "ratings": [
        { "_index": "products", "_id": "sku-2222", "rating": 3 }
      ]
    }
  ],
  "metric": {
    "dcg": { "k": 10, "normalize": true }
  }
}

{
  "metric_score": 0.84,
  "details": {
    "running_shoes_query": { "metric_score": 0.91, "unrated_docs": [...] },
    "red_dress_query": { "metric_score": 0.77, "unrated_docs": [...] }
  }
}

import requests, csv

def load_judgments(path):
    by_query = {}
    with open(path) as f:
        for row in csv.DictReader(f):
            by_query.setdefault(row["query"], []).append({
                "_index": "products",
                "_id": row["doc_id"],
                "rating": int(row["rating"]),
            })
    return by_query

def test_search_relevance_baseline():
    judgments = load_judgments("tests/judgments.csv")
    requests_payload = [
        {
            "id": q.replace(" ", "_"),
            "request": { "query": { "match": { "name": q } } },
            "ratings": ratings,
        }
        for q, ratings in judgments.items()
    ]
    body = {
        "requests": requests_payload,
        "metric": { "dcg": { "k": 10, "normalize": true } },
    }
    r = requests.post("http://localhost:9200/products/_rank_eval", json=body)
    result = r.json()

    # Baseline NDCG must not regress vs known-good
    assert result["metric_score"] >= 0.80, f"NDCG@10 regressed: {result['metric_score']}"

def test_no_query_drops_more_than_10_percent():
    current = run_rank_eval()
    baseline = json.loads(Path("tests/baseline.json").read_text())

    for query_id, baseline_score in baseline["details"].items():
        current_score = current["details"][query_id]["metric_score"]
        delta = current_score - baseline_score["metric_score"]
        assert delta >= -0.10, \
            f"Query {query_id} dropped {delta:.2f} (was {baseline_score['metric_score']:.2f}, now {current_score:.2f})"