"""Evaluate a specific verse sample with the project's evaluator.

Writes per-verse dimension scores to a JSON file WITHOUT touching the
canonical chapter eval files — the harness used for all A/B experiments
(register calibration, morphology, translator-model comparisons).

Usage:
    uv run python scripts/evaluate_verse_sample.py <out.json> book:ch:v [...]
"""
import json, subprocess, sys
sys.path.insert(0, str(__import__("pathlib").Path(__file__).parent.parent))
from src.context.builder import ContextBuilder
from src.context.glossary import GlossaryManager
from src.data.loader import DataLoader
from src.data.storage import Storage
from src.evaluation.evaluator import (
    EVALUATION_SYSTEM_PROMPT, TranslationEvaluator, parse_evaluation_response,
)
from src.models.verse import VerseReference
from src.pipeline.translator import DEFAULT_MODEL

if len(sys.argv) < 3:
    print(__doc__)
    raise SystemExit(2)
out_path, refs = sys.argv[1], " ".join(sys.argv[2:]).split()
loader = DataLoader("data")
glossary = GlossaryManager(loader.load_glossary())
builder = ContextBuilder(loader, glossary)
storage = Storage("data")
evaluator = TranslationEvaluator(data_dir="data")

results = {}
for r in refs:
    book, ch, v = r.split(":")
    ref = VerseReference(book=book, chapter=int(ch), verse=int(v))
    t = storage.load_translation(ref)
    if not t:
        print(f"{r}: no translation"); continue
    ctx = builder.build_context(ref)
    prompt = evaluator.build_verse_prompt(t.text, ctx)
    full = f"{EVALUATION_SYSTEM_PROMPT}\n\n{prompt}"
    p = subprocess.run(["claude", "--print", "-p", full, "--model", DEFAULT_MODEL],
                       capture_output=True, text=True, timeout=300)
    if p.returncode != 0:
        print(f"{r}: eval failed: {p.stderr[:100]}"); continue
    try:
        scores = parse_evaluation_response(p.stdout.strip())
    except ValueError as e:
        print(f"{r}: parse failed: {e}"); continue
    results[r] = {
        "text": t.text,
        "scores": {k: {"score": s.score, "reasoning": s.reasoning} for k, s in scores.items()},
        "avg": sum(s.score for s in scores.values()) / len(scores),
    }
    print(f"{r}: avg={results[r]['avg']:.1f}")
json.dump(results, open(out_path, "w"), ensure_ascii=False, indent=1)
print(f"saved {len(results)} to {out_path}")