"""Reproduce the article's arithmetic on synthetic rankings. Python 3, no packages.

These rankings and relevance labels are authored examples. No retriever,
reranking model or paid API is called. Change the lists to inspect failure cases.
"""
from collections import defaultdict
from math import log2


def reciprocal_rank_fusion(rankings, k=60):
    scores = defaultdict(float)
    for ranking in rankings:
        seen = set()
        for rank, chunk_id in enumerate(ranking, start=1):
            if chunk_id in seen:
                raise ValueError("Each ranking must contain distinct chunk IDs")
            seen.add(chunk_id)
            scores[chunk_id] += 1 / (k + rank)
    return sorted(scores.items(), key=lambda item: (-item[1], item[0]))


def ranking_metrics(order, labels, required, cutoff=2):
    if len(order) != len(set(order)):
        raise ValueError("Duplicate candidates")
    relevant = {item for item, grade in labels.items() if grade > 0}
    selected = order[:cutoff]
    recall = len(set(selected) & relevant) / len(relevant)
    reciprocal_rank = next(
        (1 / rank for rank, item in enumerate(order, 1) if item in relevant), 0
    )
    dcg = sum((2 ** labels[item] - 1) / log2(rank + 1)
              for rank, item in enumerate(selected, 1))
    ideal = sorted(labels.values(), reverse=True)[:cutoff]
    idcg = sum((2 ** grade - 1) / log2(rank + 1)
               for rank, grade in enumerate(ideal, 1))
    return {
        "Recall@2": round(recall, 6),
        "RR": round(reciprocal_rank, 6),
        "nDCG@2": round(dcg / idcg, 6),
        "Required evidence coverage@2": len(set(selected) & required) / len(required),
    }


if __name__ == "__main__":
    lexical = ["A", "B", "C"]
    dense = ["B", "D", "C"]
    print("RRF, k=60 (authored rankings)")
    for chunk_id, score in reciprocal_rank_fusion([lexical, dense]):
        print(chunk_id, f"{score:.6f}")
    print("\nIf the dense route is accidentally counted twice:")
    for chunk_id, score in reciprocal_rank_fusion([lexical, dense, dense]):
        print(chunk_id, f"{score:.6f}")

    # D2 is topically related. D3 and D5 supply the two required facts.
    labels = {"D1": 0, "D2": 1, "D3": 2, "D4": 0, "D5": 2}
    required = {"D3", "D5"}
    before = ["D1", "D2", "D3", "D4", "D5"]
    ideal = ["D3", "D5", "D2", "D1", "D4"]
    print("\nRanking arithmetic (authored order versus ideal order)")
    for name, order in [("Before", before), ("Ideal", ideal)]:
        print(name, ranking_metrics(order, labels, required))
