#!/usr/bin/env python3
"""
inspect_poc.py - Wertet buyers_raw.jsonl nach dem POC-Lauf aus.

Zeigt:
  - Feld-Fuellquoten (wie oft ist welches Feld nicht null?)
  - Text- vs Vision-Qualitaet im Vergleich
  - die Multi-Sheet-Gruppierung: welche Person hat mehrere Sheets?
  - Datums-Parsing-Quote (fuer die "letzte 5 Jahre"-Abfragen kritisch)

Aufruf: python inspect_poc.py ./poc_out/buyers_raw.jsonl
"""

import sys
import json
import re
from collections import defaultdict, Counter


def person_key(rec):
    """Gruppierungsschluessel: bevorzugt echten Namen, sonst name_company."""
    name = (rec.get("prospective_buyer") or rec.get("name_company") or "").strip().lower()
    # "Nachname, Vorname" und "Vorname Nachname" grob angleichen
    name = re.sub(r"\s+", " ", name)
    return name or "(unbekannt)"


def main():
    path = sys.argv[1] if len(sys.argv) > 1 else "./poc_out/buyers_raw.jsonl"
    records = []
    with open(path, encoding="utf-8") as f:
        for line in f:
            line = line.strip()
            if line:
                records.append(json.loads(line))

    ok = [r for r in records if "_error" not in r]
    err = [r for r in records if "_error" in r]

    print(f"Datensaetze: {len(records)}  (ok: {len(ok)}, Fehler: {len(err)})\n")

    # Feld-Fuellquoten
    fields = ["name_company", "prospective_buyer", "company", "phone", "cell",
              "email", "address", "state", "how_did_you_hear",
              "interested_in_updates", "types_of_business", "background_experience",
              "date_of_introduction"]
    print("=== Feld-Fuellquoten (nicht-null / erfolgreiche) ===")
    for fld in fields:
        filled = 0
        for r in ok:
            v = r.get(fld)
            if v not in (None, "", [], {}):
                filled += 1
        pct = 100 * filled / len(ok) if ok else 0
        print(f"  {fld:<24} {filled:>4}/{len(ok)}  ({pct:4.0f}%)")

    # Text vs Vision
    print("\n=== Text- vs Vision-Pfad ===")
    for mode in ("text", "vision"):
        sub = [r for r in ok if r.get("extraction_mode") == mode]
        if not sub:
            continue
        # als grobe Qualitaetsmetrik: durchschnittliche Zahl gefuellter Felder
        avg_filled = sum(
            sum(1 for f in fields if r.get(f) not in (None, "", [], {})) for r in sub
        ) / len(sub)
        print(f"  {mode:<7} {len(sub):>4} Docs, im Schnitt {avg_filled:.1f}/{len(fields)} Felder gefuellt")

    # Datums-Parsing
    print("\n=== Date of Introduction ===")
    iso = sum(1 for r in ok if isinstance(r.get("date_of_introduction"), str)
              and re.match(r"\d{4}-\d{2}-\d{2}", r["date_of_introduction"]))
    nonnull = sum(1 for r in ok if r.get("date_of_introduction"))
    print(f"  vorhanden: {nonnull}/{len(ok)}   davon sauber ISO (YYYY-MM-DD): {iso}")

    # Multi-Sheet-Gruppierung
    print("\n=== Personen mit mehreren Sheets ===")
    groups = defaultdict(list)
    for r in ok:
        groups[person_key(r)].append(r)
    multi = {k: v for k, v in groups.items() if len(v) > 1}
    print(f"  distinkte Personen: {len(groups)}, davon mit >1 Sheet: {len(multi)}")
    for name, recs in sorted(multi.items(), key=lambda x: -len(x[1]))[:10]:
        cats = set()
        dates = []
        for r in recs:
            cats.update(r.get("types_of_business") or [])
            if r.get("date_of_introduction"):
                dates.append(r["date_of_introduction"])
        print(f"    {name!r}: {len(recs)} Sheets | Kategorien: {sorted(cats)} | Daten: {sorted(dates)}")

    # Haeufigste Kategorien
    print("\n=== Top 20 Kategorien (roh) ===")
    cats = Counter()
    for r in ok:
        for c in (r.get("types_of_business") or []):
            cats[c.strip().lower()] += 1
    for c, n in cats.most_common(20):
        print(f"  {n:>4}  {c}")


if __name__ == "__main__":
    main()