"""Summarize a synthetic/all-query trace, retaining missing ranks in denominators.

No inference about the private SOP79 run. Python 3.11+, standard library only.
"""
from __future__ import annotations
import csv
import json
import sys


def summarize(rows: list[dict]) -> dict:
    ids = [row["query_id"] for row in rows]
    if not rows or len(ids) != len(set(ids)):
        raise ValueError("provide a nonempty trace with unique query IDs")
    total = len(rows)
    stages = {}
    for stage in ("dense", "sparse", "fused", "rerank"):
        ranks = []
        for row in rows:
            value = row.get(stage + "_rank", "")
            rank = int(value) if value not in (None, "", "-") else None
            if rank is not None and rank < 1:
                raise ValueError("ranks are one-based; use blank for missing")
            ranks.append(rank)
        hits = sum(rank is not None and rank <= 5 for rank in ranks)
        stages[stage] = {"queries": total, "rank_observed": sum(x is not None for x in ranks),
                         "top5_hits": hits, "hit_rate_at_5": hits / total}
    return {"scope": "input rows only; use the full query set", "stages": stages}


if __name__ == "__main__":
    with open(sys.argv[1], newline="", encoding="utf-8") as stream:
        print(json.dumps(summarize(list(csv.DictReader(stream))), indent=2))