"""Probe confidence-related metrics across last N estimates.

Usage: python scripts/probe-confidence-distribution.py [N=100]

Prints distribution of:
  - n_analogs per estimate
  - unique addresses per estimate
  - lots_per_address ratio (helps tune downgrade threshold)
  - confidence label distribution
"""
import sys
from collections import Counter
from statistics import median, quantiles

# Run inside tradein-mvp/backend with DATABASE_URL set
from sqlalchemy import create_engine, text
from app.core.config import settings


def main(limit: int = 100) -> None:
    engine = create_engine(settings.database_url, future=True)
    with engine.connect() as conn:
        rows = conn.execute(
            text("""
                SELECT id, n_analogs, confidence, analogs
                FROM trade_in_estimates
                WHERE n_analogs > 0
                ORDER BY created_at DESC
                LIMIT :limit
            """),
            {"limit": limit},
        ).mappings().all()

    if not rows:
        print("No estimates with analogs.")
        return

    n_analogs_vals: list[float] = []
    unique_vals: list[float] = []
    ratios: list[float] = []
    conf_counter: Counter[str] = Counter()
    for r in rows:
        n = r["n_analogs"]
        analogs = r["analogs"] or []
        unique = len(
            {(a.get("address") or "").strip().lower() for a in analogs if a.get("address")}
        )
        n_analogs_vals.append(n)
        unique_vals.append(unique)
        ratios.append(n / max(unique, 1))
        conf_counter[r["confidence"]] += 1

    def stats(label: str, vals: list[float]) -> None:
        if not vals:
            print(f"{label}: empty")
            return
        vals_sorted = sorted(vals)
        q = quantiles(vals_sorted, n=4) if len(vals_sorted) >= 4 else [vals_sorted[0]] * 3
        print(
            f"{label}: n={len(vals)} min={min(vals):.2f} p25={q[0]:.2f} "
            f"median={median(vals):.2f} p75={q[2]:.2f} max={max(vals):.2f}"
        )

    print(f"\n=== Distribution over last {len(rows)} estimates ===")
    stats("n_analogs       ", n_analogs_vals)
    stats("unique_addresses", unique_vals)
    stats("lots_per_address", ratios)
    print(f"confidence       : {dict(conf_counter)}")
    p75_idx = int(len(ratios) * 0.75)
    print(
        f"\n-> p75 lots_per_address = {sorted(ratios)[p75_idx]:.2f} "
        "(use as downgrade threshold)"
    )


if __name__ == "__main__":
    main(int(sys.argv[1]) if len(sys.argv) > 1 else 100)