"""Evaluate outfit composition *quality* — Scandi fit, contrast axes, color harmony.

Complements diversity eval. Where diversity asks "do we vary across anchors",
quality asks "is each individual outfit good".

Metrics:

    mean_scandi_score    — average outfit-level scandi score over the sample.
                           AC1 floor: 0.55 (recon target — Scandi catalog
                           heavy enough that this should be easy to clear).
    mean_axes_covered    — average count of Bornstein axes the outfit speaks
                           on (any item past ±0.3). AC1 floor: 2.5.
    color_harmony_rate   — fraction of dress→shoes pairs that read
                           harmonious. A pair is "harmonious" if either
                           dominant LAB is neutral OR ΔE2000 < 25. This
                           replaces the raw mean-ΔE metric in the original
                           plan, which penalises the catalog's correct-but-
                           neutral shoe pairings (a black mule with a blue
                           dress = high ΔE but stylistically right).
                           AC7 floor: 0.80.

Usage::

    uv run --no-project --with numpy --with "scikit-image>=0.22,<0.25" \\
           scripts/eval_outfit_quality.py [N=80]
"""

# /// script
# requires-python = ">=3.11"
# dependencies = ["numpy", "scikit-image>=0.22,<0.25"]
# ///

import random
import sqlite3
import sys
from pathlib import Path
from statistics import mean

sys.path.insert(0, str(Path(__file__).resolve().parents[1]))

from taste.colors import delta_e, is_neutral  # noqa: E402
from taste.outfits import _row_dom, compose_outfit  # noqa: E402

DB_PATH = Path.home() / ".taste/taste.db"
SAMPLE_BUCKETS = ("dress", "top", "bottom", "outerwear")

SCANDI_FLOOR = 0.55
AXES_FLOOR = 2.5
HARMONY_FLOOR = 0.80


def main() -> int:
    n = int(sys.argv[1]) if len(sys.argv) > 1 else 80
    rng = random.Random(42)

    conn = sqlite3.connect(str(DB_PATH))
    conn.row_factory = sqlite3.Row

    placeholders = ",".join("?" * len(SAMPLE_BUCKETS))
    candidates = conn.execute(
        f"""
        SELECT url FROM products
        WHERE in_stock = 1 AND score IS NOT NULL
          AND image_url != '' AND canonical_category IN ({placeholders})
        """,
        SAMPLE_BUCKETS,
    ).fetchall()
    if not candidates:
        print("No anchors found — nothing to evaluate.")
        return 1
    sample = rng.sample(candidates, min(n, len(candidates)))

    scandi_per: list[float] = []
    axes_per: list[int] = []
    contrasted_per: list[int] = []
    harmonious_pairs = 0
    total_dress_shoe_pairs = 0
    delta_e_dress_shoes: list[float] = []  # diagnostic only

    for r in sample:
        outfit = compose_outfit(r["url"], conn)
        if not outfit:
            continue
        scandi_per.append(outfit["scandi_score"])
        axes_per.append(len(outfit["axes_covered"]))
        contrasted_per.append(len(outfit.get("axes_contrasted", [])))

        if outfit["anchor_bucket"] != "dress":
            continue
        anchor_dom = _row_dom(outfit["anchor"])
        for c in outfit["complements"]:
            if c["bucket"] != "shoes":
                continue
            shoes_dom = _row_dom(c["product"])
            if anchor_dom is None or shoes_dom is None:
                continue
            total_dress_shoe_pairs += 1
            de = delta_e(anchor_dom, shoes_dom)
            delta_e_dress_shoes.append(de)
            if is_neutral(anchor_dom) or is_neutral(shoes_dom) or de < 25:
                harmonious_pairs += 1

    if not scandi_per:
        print("No outfits composed.")
        return 1

    n_outfits = len(scandi_per)
    avg_scandi = mean(scandi_per)
    avg_axes = mean(axes_per)
    avg_contrasted = mean(contrasted_per)
    harmony_rate = (
        harmonious_pairs / total_dress_shoe_pairs if total_dress_shoe_pairs else 0.0
    )

    print(f"Outfits composed:        {n_outfits}/{len(sample)}")
    print(f"Mean Scandi score:       {avg_scandi:.3f}   (floor {SCANDI_FLOOR:.2f})")
    print(f"Mean axes covered:       {avg_axes:.2f}    (floor {AXES_FLOOR:.2f})")
    print(f"Mean axes contrasted:    {avg_contrasted:.2f}    (diagnostic)")
    if total_dress_shoe_pairs:
        print(
            f"Color harmony rate:      {harmony_rate:.3f}   "
            f"(floor {HARMONY_FLOOR:.2f}, n={total_dress_shoe_pairs})"
        )
        print(f"Mean dress→shoes ΔE2000: {mean(delta_e_dress_shoes):.1f}    (diagnostic)")

    fails: list[str] = []
    if avg_scandi < SCANDI_FLOOR:
        fails.append(f"scandi {avg_scandi:.3f} < {SCANDI_FLOOR}")
    if avg_axes < AXES_FLOOR:
        fails.append(f"axes {avg_axes:.2f} < {AXES_FLOOR}")
    if total_dress_shoe_pairs and harmony_rate < HARMONY_FLOOR:
        fails.append(f"harmony {harmony_rate:.3f} < {HARMONY_FLOOR}")

    if fails:
        print(f"\nFAIL — {'; '.join(fails)}")
        return 1
    print("\nPASS — all quality floors cleared")
    return 0


if __name__ == "__main__":
    sys.exit(main())