#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
釋字及憲法法庭資料庫 — 一鍵建置腳本
=====================================
在 Windows 上直接執行：

    cd C:\\Users\\JudyLin\\Documents\\釋字及憲法法庭資料庫
    python _pipeline\\建置資料庫.py

它會做完下列全部工作，中途可以隨時中斷，再跑一次會從斷點續做：

  1. 從司法院憲法法庭網站抓取 813 則大法官解釋 + 57 則憲法法庭判決原始 HTML
  2. 逐頁驗證「第N筆/共813筆」與頁內解釋字號，確認編號與內容相符
  3. 抓取全部意見書 PDF（約 1,680 篇）
  4. 逐篇抽出 PDF 文字，中文字比例未達門檻者標示為需 OCR，不留空白也不編造
  5. 產出 Obsidian Markdown 筆記庫、法條反向索引、總索引、爭點速查、年表
  6. 產出單一檔案的離線網頁版資料庫（可直接雙擊開啟）
  7. 產出校驗報告

鐵律：原文區塊逐字照錄，不改標點不重排；每則保留官網網址與抓取時間戳。
"""

import os, sys, re, json, time, html, zipfile, subprocess, collections
from pathlib import Path
from concurrent.futures import ThreadPoolExecutor, as_completed

BASE = "https://cons.judicial.gov.tw"
ROOT = Path(__file__).resolve().parent.parent     # 資料庫根目錄
RAW  = ROOT / "90_原始存檔"
PDFD = ROOT / "90_原始存檔" / "意見書PDF"
PIPE = ROOT / "_pipeline"
WORKERS = 4          # 對官網併發數，請勿調高
DELAY = 0.25         # 每次請求後的禮貌延遲（秒）

# 釋字第813號不符合 id = 310181 + N 的規則，實際 id 由官網「下一筆」連結取得
JIE_SPECIAL = {813: 325335}
JUD_IDS = "111-01:310024,111-02:309998,111-03:340434,111-04:309908,111-05:309913,111-06:309577,111-07:309994,111-08:340546,111-09:339757,111-10:309974,111-11:309958,111-12:340615,111-13:309956,111-14:325412,111-15:310029,111-16:310030,111-17:310021,111-18:309960,111-19:341172,111-20:309995,112-01:309875,112-02:309945,112-03:340469,112-04:310013,112-05:344513,112-06:310005,112-07:345269,112-08:340775,112-09:345006,112-10:325406,112-11:309954,112-12:309916,112-13:325399,112-14:351424,112-15:309832,112-16:310022,112-17:310025,112-18:351004,112-19:309986,112-20:309944,113-01:346741,113-02:309854,113-03:340393,113-04:352146,113-05:339983,113-06:349469,113-07:346206,113-08:351689,113-09:352966,113-10:350658,113-11:349244,114-01:355485,115-01:350743,115-02:343660,115-03:358150,115-04:351300,115-05:343661"


# ---------------------------------------------------------------- 相依套件
def need(mod, pip=None):
    try:
        return __import__(mod)
    except ImportError:
        print(f"[安裝] {pip or mod} …")
        subprocess.run([sys.executable, "-m", "pip", "install", "-q", pip or mod], check=True)
        return __import__(mod)

requests = need("requests")
need("bs4", "beautifulsoup4"); need("lxml")
from bs4 import BeautifulSoup
pypdf = need("pypdf")

UA = {"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/125.0 Safari/537.36"}
PROBE = f"{BASE}/docdata.aspx?fid=100&id=310182"
MODE = None          # 連線策略，由 build_session() 決定
SESSION = None


def _plain():
    s = requests.Session(); s.headers.update(UA); return s


def _truststore():
    """使用 Windows 憑證存放區，解決防毒／公司代理做 TLS 攔截時的憑證不受信任問題。"""
    need("truststore")
    import truststore
    truststore.inject_into_ssl()
    s = requests.Session(); s.headers.update(UA); return s


def _legacy():
    """放寬到舊版 TLS 重新協商與較低安全等級，解決舊式伺服器握手失敗。"""
    import ssl
    from requests.adapters import HTTPAdapter
    from urllib3.poolmanager import PoolManager
    ctx = ssl.create_default_context()
    ctx.options |= getattr(ssl, "OP_LEGACY_SERVER_CONNECT", 0x4)
    try:
        ctx.set_ciphers("DEFAULT@SECLEVEL=1")
    except Exception:
        pass

    class A(HTTPAdapter):
        def init_poolmanager(self, connections, maxsize, block=False, **kw):
            kw["ssl_context"] = ctx
            self.poolmanager = PoolManager(num_pools=connections, maxsize=maxsize, block=block, **kw)

    s = requests.Session(); s.headers.update(UA); s.mount("https://", A()); return s


def _noverify():
    """最後手段：不驗證憑證。僅用於抓取公開政府文件，且每則仍保留官網網址可自行核對。"""
    import urllib3
    urllib3.disable_warnings()
    s = requests.Session(); s.headers.update(UA); s.verify = False; return s


def _curl_available():
    try:
        subprocess.run(["curl", "--version"], capture_output=True, timeout=10, check=True)
        return True
    except Exception:
        return False


def build_session():
    global SESSION, MODE
    errs = []
    for name, fn in (("標準", _plain), ("Windows憑證存放區", _truststore),
                     ("寬鬆TLS", _legacy), ("不驗證憑證", _noverify)):
        try:
            s = fn()
            r = s.get(PROBE, timeout=45)
            r.raise_for_status()
            if "解釋字號" not in r.text:
                raise ValueError("回應內容不含預期欄位")
            SESSION, MODE = s, name
            print(f"      連線策略：{name}")
            if name == "不驗證憑證":
                print("      ⚠ 已停用憑證驗證。原始出處網址仍逐則保留，可自行核對。")
            return
        except Exception as e:
            errs.append(f"{name} → {type(e).__name__}: {str(e)[:160]}")
    if _curl_available():
        try:
            r = subprocess.run(["curl", "-sS", "-L", "--max-time", "60", "-A", UA["User-Agent"], PROBE],
                               capture_output=True, check=True)
            if "解釋字號" in r.stdout.decode("utf-8", "replace"):
                SESSION, MODE = "curl", "curl"
                print("      連線策略：改用系統 curl（走 Windows 憑證存放區）")
                return
            errs.append("curl → 回應內容不含預期欄位")
        except Exception as e:
            errs.append(f"curl → {type(e).__name__}: {str(e)[:160]}")
    else:
        errs.append("curl → 系統未安裝")
    print("\n[連線失敗] 四種策略都無法連上司法院憲法法庭網站：")
    for e in errs:
        print("   -", e)
    print("\n請把上面的錯誤訊息貼回對話，我依實際原因調整。")
    sys.exit(1)


def get(url, binary=False, tries=3):
    for i in range(tries):
        try:
            if MODE == "curl":
                r = subprocess.run(["curl", "-sS", "-L", "--max-time", "90", "-A", UA["User-Agent"], url],
                                   capture_output=True, check=True)
                time.sleep(DELAY)
                return r.stdout if binary else r.stdout.decode("utf-8", "replace")
            r = SESSION.get(url, timeout=60)
            r.raise_for_status()
            time.sleep(DELAY)
            return r.content if binary else r.text
        except Exception:
            if i == tries - 1:
                raise
            time.sleep(2 * (i + 1))


# ---------------------------------------------------------------- 1. 抓取
def targets():
    out = []
    for n in range(1, 814):
        out.append((f"釋字{n:04d}", f"{BASE}/docdata.aspx?fid=100&id={JIE_SPECIAL.get(n, 310181 + n)}"))
    for part in JUD_IDS.split(","):
        k, i = part.split(":")
        out.append((f"憲判{k}", f"{BASE}/docdata.aspx?fid=38&id={i}"))
    return out


def fetch_one(key, url):
    dst = RAW / f"{key}.html"
    if dst.exists() and dst.stat().st_size > 200:
        return key, "cached"
    soup = BeautifulSoup(get(url), "lxml")
    art = soup.select_one("#Print_area")
    if art is None:
        return key, "NO_CONTENT"
    dst.write_text(f"<!--src:{url}-->\n<!--at:{time.strftime('%Y-%m-%dT%H:%M:%S%z')}-->\n" + str(art),
                   encoding="utf-8")
    return key, "ok"


def step_fetch():
    RAW.mkdir(parents=True, exist_ok=True)
    print("[0/6] 測試連線 …")
    build_session()
    tg = targets()
    todo = [(k, u) for k, u in tg if not (RAW / f"{k}.html").exists()]
    print(f"[1/6] 抓取本文：共 {len(tg)} 份，待抓 {len(todo)} 份")
    bad = []
    with ThreadPoolExecutor(WORKERS) as ex:
        futs = {ex.submit(fetch_one, k, u): k for k, u in todo}
        for i, f in enumerate(as_completed(futs), 1):
            try:
                k, st = f.result()
                if st not in ("ok", "cached"):
                    bad.append(f"{k}:{st}")
            except Exception as e:
                bad.append(f"{futs[f]}:{type(e).__name__}")
            if i % 50 == 0:
                print(f"      {i}/{len(todo)}")
    if bad:
        print("      失敗：", bad[:10])
    return bad


# ---------------------------------------------------------------- 2. 解析
SEC_ORDER = ["解釋文", "主文", "理由書", "理由", "判決理由",
             "主筆大法官記載", "大法官就主文所採立場表", "判決摘要", "相關文件"]
ROC = re.compile(r"(?:中華民國)?\s*(\d{2,3})\s*年\s*(\d{1,2})\s*月\s*(\d{1,2})\s*日")


def roc_iso(s):
    m = ROC.search(s or "")
    if not m:
        return "", ""
    return f"{int(m.group(1))+1911:04d}-{int(m.group(2)):02d}-{int(m.group(3)):02d}", m.group(0).replace("中華民國", "").strip()


def blocks(node):
    lis = node.select("ul.paragraphs > li, ul.num-list > li")
    if lis:
        out = []
        for li in lis:
            lab = li.find("label")
            pre = li.find("pre")
            body = (pre.get_text() if pre else li.get_text()).replace("\r\n", "\n").strip("\n").rstrip()
            out.append([lab.get_text(strip=True) if lab else "", body])
        return out
    t = node.get_text("\n").strip()
    return [["", t]] if t else []


def parse_all():
    recs, integrity = [], []
    for p in sorted(RAW.glob("*.html")):
        key = p.stem
        txt = p.read_text(encoding="utf-8")
        src = (re.search(r"<!--src:(.*?)-->", txt) or [None, ""])[1]
        at = (re.search(r"<!--at:(.*?)-->", txt) or [None, ""])[1]
        soup = BeautifulSoup(txt, "lxml")

        F = {}
        for ul in soup.select("ul.flex"):
            t, v = ul.find("li", class_="title"), ul.find("li", class_="text")
            if t and v:
                F[t.get_text(strip=True)] = v

        is_jie = key.startswith("釋字")
        # 完整性驗證：頁面自稱的序號與字號
        if is_jie:
            n = int(key[2:])
            sp = soup.select_one(".btn_list span")
            cnt = re.sub(r"\s", "", sp.get_text()) if sp else ""
            m = re.search(r"第(\d+)筆/共(\d+)筆", cnt)
            zt = F.get("解釋字號").get_text(strip=True) if F.get("解釋字號") else ""
            zm = re.search(r"釋字第(\d+)號", zt)
            if not m or int(m.group(1)) != n:
                integrity.append(f"{key} 頁面序號不符：{cnt}")
            if not zm or int(zm.group(1)) != n:
                integrity.append(f"{key} 字號不符：{zt[:20]}")

        num_node = F.get("解釋字號") or F.get("判決字號")
        num_txt = re.sub(r"\s+", " ", num_node.get_text()).strip() if num_node else key
        mm = re.match(r"(釋字第\d+號|\d+年憲判字第\d+號)\s*(?:【(.*?)】)?", num_txt)
        pub_node = F.get("解釋公布院令") or F.get("判決日期") or F.get("公布日期")
        pub_txt = re.sub(r"\s+", " ", pub_node.get_text()).strip() if pub_node else ""
        iso, roc = roc_iso(pub_txt)
        z_node = F.get("解釋爭點") or F.get("案由")

        secs = [[L, blocks(F[L])] for L in SEC_ORDER if L in F and blocks(F[L])]
        fnode = F.get("意見書、抄本等文件") or F.get("意見書")
        ops = []
        if fnode:
            for a in fnode.select("div.lawFileList a"):
                s = a.find("strong")
                title = (s.get_text() if s else a.get_text()).strip()
                if "意見書" not in title:
                    continue
                href = a.get("href", "")
                dl = re.search(r"[?&]id=(\d+)", href)
                ops.append({"t": title, "h": BASE + href if href.startswith("/") else href,
                            "id": dl.group(1) if dl else ""})
        lnode = F.get("相關法令")
        laws = list(dict.fromkeys(p2.get_text(strip=True) for p2 in lnode.select("div.lawLinkList a p"))) if lnode else []
        meta = [[L, re.sub(r"\s+", " ", F[L].get_text()).strip()] for L in ("原分案號", "聲請人")
                if L in F and F[L].get_text(strip=True)]

        recs.append({"k": key, "ty": "釋字" if is_jie else "憲判",
                     "z": mm.group(1) if mm else num_txt, "n": (mm.group(2) if mm and mm.group(2) else ""),
                     "dt": iso, "roc": roc, "decree": re.sub(r".*?日\s*", "", pub_txt).strip(),
                     "m": meta, "q": re.sub(r"\s+", " ", z_node.get_text()).strip() if z_node else "",
                     "s": secs, "o": ops, "l": laws, "u": src, "at": at})
    recs.sort(key=lambda r: (r["ty"] != "釋字", r["k"]))
    return recs, integrity


# ---------------------------------------------------------------- 3. 意見書 PDF
def step_pdfs(recs):
    PDFD.mkdir(parents=True, exist_ok=True)
    jobs = []
    for r in recs:
        for i, o in enumerate(r["o"], 1):
            safe = re.sub(r'[\\/:*?"<>|]', "_", o["t"])[:110]
            jobs.append((f"{r['k']}_{i:02d}_{safe}.pdf", o["h"]))
    todo = [(n, u) for n, u in jobs if not (PDFD / n).exists()]
    print(f"[3/6] 意見書 PDF：共 {len(jobs)} 篇，待抓 {len(todo)} 篇")
    bad = []

    def one(name, url):
        data = get(url, binary=True)
        if len(data) < 500:
            raise ValueError("too small")
        (PDFD / name).write_bytes(data)

    with ThreadPoolExecutor(WORKERS) as ex:
        futs = {ex.submit(one, n, u): n for n, u in todo}
        for i, f in enumerate(as_completed(futs), 1):
            try:
                f.result()
            except Exception as e:
                bad.append(f"{futs[f]}:{type(e).__name__}")
            if i % 100 == 0:
                print(f"      {i}/{len(todo)}")
    if bad:
        print("      失敗：", bad[:10])
    return bad


CJK = re.compile(r"[\u4e00-\u9fff]")


def pdf_text(path):
    try:
        rd = pypdf.PdfReader(str(path))
        t = "\n".join((pg.extract_text() or "") for pg in rd.pages)
    except Exception:
        return "", 0.0
    t = re.sub(r"\n{3,}", "\n\n", t.replace("\r\n", "\n")).strip()
    flat = re.sub(r"\s", "", t)
    q = (len(CJK.findall(flat)) / len(flat)) if len(flat) >= 80 else 0.0
    return t, q


def step_extract(recs):
    print("[4/6] 抽取意見書文字 …")
    store, stats = {}, {"ok": 0, "低品質": 0, "缺檔": 0}
    for r in recs:
        for i, o in enumerate(r["o"], 1):
            safe = re.sub(r'[\\/:*?"<>|]', "_", o["t"])[:110]
            p = PDFD / f"{r['k']}_{i:02d}_{safe}.pdf"
            if not p.exists():
                stats["缺檔"] += 1
                o["x"] = ""
                continue
            t, q = pdf_text(p)
            if q >= 0.25:
                o["x"] = t
                stats["ok"] += 1
            else:
                o["x"] = ""
                o["low"] = True
                stats["低品質"] += 1
    print("      ", stats)
    return stats


# ---------------------------------------------------------------- 5. 產出
# ---------------------------------------------------------------- 大法官姓名
# 官網有兩種寫法：「林大法官俊益」（姓＋大法官＋名）與「詹森林大法官」（全名＋大法官）。
# 先用寬鬆規則掃出候選，剔除碎片後建立名冊，再以名冊回頭比對，避免正則誤切。
_RAW = re.compile(r"([\u4e00-\u9fff]{1,2})大法官([\u4e00-\u9fff]{1,3}?)"
                  r"(?=提出|提|加入|及|與|均|、|，|。|）|\)|之|等|參與|共同|協同|不同|意見|部分|一部|主筆|$)")
_LEAD = "及與含和號決書判由第之（(、，"
_TAIL = "提加均出入共之"


def _clean_name(n):
    while n and n[0] in _LEAD:
        n = n[1:]
    if n.endswith("主筆"):
        n = n[:-2]
    while n and n[-1] in _TAIL:
        n = n[:-1]
    return n


def _roster(recs):
    cand = collections.Counter()

    def scan(t):
        for m in _RAW.finditer(t or ""):
            n = _clean_name(m.group(1) + m.group(2))
            if 2 <= len(n) <= 4:
                cand[n] += 1
        for m in re.findall(r"([\u4e00-\u9fff]{2,4})大法官主筆", t or ""):
            n = _clean_name(m)
            if 2 <= len(n) <= 4:
                cand[n] += 1

    for r in recs:
        for o in r["o"]:
            scan(o["t"])
        for L, ps in r["s"]:
            if L == "主筆大法官記載":
                for _, tx in ps:
                    scan(tx)
    # 剔除「是其他姓名之真子字串」的碎片，例如「太郎」之於「呂太郎」
    return sorted({k for k in cand if not any(k != o and k in o for o in cand)})


def _sitting(main_text):
    """釋字第1至79號的解釋文末尾附有出席大法官名單，第80號以後官網不再附。
    名單以全形空白對齊，單字姓名會被拆成兩段（如「林　彬」），需合併還原。"""
    i = main_text.find("大法官會議主席")
    if i < 0:
        return "", []
    blk = re.split(r"\n\s*\n", main_text[i:])[0]
    m = re.search(r"大法官會議主席\s*院\s*長\s*([^\s　]+)", blk)
    chair = re.sub(r"大法官$", "", m.group(1)) if m else ""
    j = blk.find("大法官", 7)
    if j < 0:
        return chair, []
    toks = [x for x in re.split(r"[\s　]+", blk[j + 3:]) if x]
    out, k = [], 0
    while k < len(toks):
        if len(toks[k]) == 1 and k + 1 < len(toks) and len(toks[k + 1]) == 1:
            out.append(toks[k] + toks[k + 1]); k += 2
        else:
            out.append(toks[k]); k += 1
    return chair, [x for x in out if 2 <= len(x) <= 3 and x != "大法官"]


def derive_judges(recs):
    """為每則補上 j 欄位：主筆、意見書提出者、加入者、意見書類型、出席名單。"""
    roster = _roster(recs)
    TYPES = ["部分協同部分不同意見書", "部分不同部分協同意見書", "部分協同意見書",
             "部分不同意見書", "一部協同一部不同意見書", "協同意見書", "不同意見書", "意見書"]

    def find_names(t):
        flat = re.sub(r"^憲法法庭\d+年憲判字第\d+號判決", "", t or "").replace("大法官", "")
        got = sorted((flat.index(n), n) for n in roster if n in flat)
        return flat, got

    for r in recs:
        pro, add, tys = [], [], []
        for o in r["o"]:
            flat, got = find_names(o["t"])
            ip = flat.find("提出")
            if got:
                if ip < 0:
                    p2, a2 = [n for _, n in got], []
                else:
                    p2 = [n for i, n in got if i < ip]
                    a2 = [n for i, n in got if i > ip]
                    if not p2:
                        p2, a2 = [got[0][1]], [n for n in a2 if n != got[0][1]]
            else:
                p2, a2 = [], []
            o["p"], o["a"] = p2, a2
            for x in p2:
                if x not in pro:
                    pro.append(x)
            for x in a2:
                if x not in add:
                    add.append(x)
            ty = next((t for t in TYPES if t in o["t"]), "")
            o["ty"] = ty
            if ty and ty not in tys:
                tys.append(ty)
        w = []
        for L, ps in r["s"]:
            if L == "主筆大法官記載":
                for _, tx in ps:
                    f = tx.replace("大法官", "")
                    for n in roster:
                        if n in f and n not in w:
                            w.append(n)
        mt = "\n".join(px[1] for L, ps in r["s"] if L in ("解釋文", "主文") for px in ps)
        chair, sit = _sitting(mt)
        r["j"] = {"w": w, "p": pro, "a": add, "t": tys, "c": chair, "s": sit}
    return roster


def wikilaw(raw):
    return re.sub(r"\s*[（(][\d./]+[）)]\s*$", "", raw).strip()


def write_notes(recs):
    (ROOT / "10_釋字").mkdir(exist_ok=True)
    (ROOT / "20_憲判").mkdir(exist_ok=True)
    (ROOT / "30_法條").mkdir(exist_ok=True)
    for r in recs:
        sub = "10_釋字" if r["ty"] == "釋字" else "20_憲判"
        fn = f"釋字第{int(r['k'][2:]):04d}號" if r["ty"] == "釋字" else r["z"]
        y = ["---", f'類型: "{r["ty"]}"', f'字號: "{r["z"]}"', f'案名: "{r["n"]}"',
             f'公布日期: "{r["dt"]}"', f'公布日期_民國: "{r["roc"]}"', f'院令字號: "{r["decree"]}"',
             f'爭點: "{r["q"][:400].replace(chr(34), chr(39))}"',
             "相關法條: [" + ", ".join('"' + wikilaw(x).replace('"', "'") + '"' for x in r["l"]) + "]",
             f'意見書篇數: {len(r["o"])}', f'原始出處: "{r["u"]}"', f'抓取時間: "{r["at"]}"', "---", ""]
        md = y + [f'# {r["z"]}' + (f'【{r["n"]}】' if r["n"] else ""), "",
                  "> [!info] 原始出處",
                  f'> 司法院憲法法庭網站 · [官網原文]({r["u"]}) · 公布 {r["roc"] or r["dt"]} · 抓取於 {r["at"][:10]}', ""]
        for L, v in r["m"]:
            md.append(f"> **{L}**：{v}")
        if r["m"]:
            md.append("")
        if r["q"]:
            md += [f'## {"解釋爭點" if r["ty"]=="釋字" else "案由"}', "", r["q"], ""]
        for L, ps in r["s"]:
            md += [f"## {L}", ""]
            for lab, body in ps:
                md += ([f"**（{lab}）**"] if lab else []) + [body, ""]
        if r["o"]:
            md += [f'## 意見書（{len(r["o"])} 篇）', ""]
            for i, o in enumerate(r["o"], 1):
                md += [f'### {i}. {o["t"]}', "", f'> 原始檔：[官網 PDF]({o["h"]})', ""]
                if o.get("ai"):
                    a = o["ai"]
                    md += ["> [!warning] 白話解說（AI 生成，非原文，不得引用）",
                           f'> **立場**：{a["one"]}',
                           f'> **核心主張**：{a["main"]}',
                           f'> **與多數意見的差異**：{a["diff"]}']
                    if a.get("struct"):
                        md.append(f'> **論證結構**：{a["struct"]}')
                    if a.get("guide"):
                        md.append(f'> **段落指引**：{a["guide"]}')
                    md.append("")
                if o.get("x"):
                    md += [o["x"], ""]
                else:
                    md += ["> [!failure] 本篇 PDF 未能取得可用文字層，請點上方連結查閱官網原檔。", ""]
        if r["l"]:
            md += ["## 相關法條", ""]
            for x in r["l"]:
                v = re.search(r"[（(]([\d./]+)[）)]\s*$", x)
                md.append(f"- [[{wikilaw(x)}]]" + (f"（版本 {v.group(1)}）" if v else ""))
            md.append("")
        ai = r.get("ai")
        if ai:
            md += ["---", "", "## 白話解說（AI 生成，非官方原文）", "",
                   "> [!warning] 以下由 AI 依上方原文撰寫，僅供理解輔助。撰寫論文時請引用原文區塊，不得引用本區塊。", "",
                   f'**一句話結論**：{ai["one"]}', "",
                   "**爭點拆解**", "", ai["issue"], "",
                   "**結論與理由**", "", ai["holding"], ""]
            if ai.get("place"):
                md += ["**在體系中的位置**", "", ai["place"], ""]
        (ROOT / sub / f"{fn}.md").write_text("\n".join(md).rstrip() + "\n", encoding="utf-8")

    # 法條反向索引
    lawmap = collections.defaultdict(list)
    for r in recs:
        for x in r["l"]:
            lawmap[wikilaw(x)].append(r)
    used = {}
    for law, rs in lawmap.items():
        safe = re.sub(r'[\\/:*?"<>|\r\n]', "_", law).strip()
        if not safe:
            continue
        # 檔名上限：Linux 為 255 bytes，中文一字 3 bytes，保守截到 70 字並去重
        if len(safe) > 70:
            safe = safe[:70].rstrip() + "…"
        if used.get(safe) not in (None, law):
            safe = safe[:66] + "_%02d" % len(used)
        used[safe] = law
        body = ["---", '類型: "法條節點"', f'法條: "{law}"', f"被引次數: {len(rs)}", "---", "",
                f"# {law}", "", f"本庫中共 {len(rs)} 則解釋／判決將本條列為相關法令。", ""]
        for r in sorted(rs, key=lambda x: x["dt"]):
            fn = f"釋字第{int(r['k'][2:]):04d}號" if r["ty"] == "釋字" else r["z"]
            body.append(f'- [[{fn}\\|{r["z"]}]]　{r["n"]}　（{r["dt"]}）')
        (ROOT / "30_法條" / f"{safe}.md").write_text("\n".join(body) + "\n", encoding="utf-8")
    return len(lawmap)


def write_index(recs):
    (ROOT / "00_索引").mkdir(exist_ok=True)
    jie = [r for r in recs if r["ty"] == "釋字"]
    jud = [r for r in recs if r["ty"] == "憲判"]
    L = ["# 總索引", "", f"- 大法官解釋：{len(jie)} 則", f"- 憲法法庭判決：{len(jud)} 則", "",
         "## 大法官解釋", "", "| 字號 | 案名 | 公布日期 | 意見書 | 爭點 |", "|---|---|---|---|---|"]
    for r in jie:
        L.append(f'| [[釋字第{int(r["k"][2:]):04d}號\\|{r["z"]}]] | {r["n"]} | {r["dt"]} | {len(r["o"])} | {r["q"][:60]} |')
    L += ["", "## 憲法法庭判決", "", "| 字號 | 案名 | 公布日期 | 意見書 |", "|---|---|---|---|"]
    for r in jud:
        L.append(f'| [[{r["z"]}]] | {r["n"]} | {r["dt"]} | {len(r["o"])} |')
    (ROOT / "00_索引" / "總索引.md").write_text("\n".join(L) + "\n", encoding="utf-8")

    Q = ["# 爭點速查", ""] + [f'- **{r["z"]}**　{r["q"]}' for r in recs]
    (ROOT / "00_索引" / "爭點速查.md").write_text("\n".join(Q) + "\n", encoding="utf-8")

    by = collections.defaultdict(list)
    for r in recs:
        by[r["dt"][:4] or "____"].append(r)
    Y = ["# 年表", ""]
    for y in sorted(by):
        Y.append(f"## {y}（{len(by[y])} 則）")
        Y += [f'- {r["z"]}　{r["n"]}' for r in sorted(by[y], key=lambda x: x["dt"])]
        Y.append("")
    (ROOT / "00_索引" / "年表.md").write_text("\n".join(Y) + "\n", encoding="utf-8")


CSS = """*{box-sizing:border-box}body{margin:0;font:15px/1.9 "Noto Sans TC","Microsoft JhengHei",system-ui,sans-serif;color:#1a1a1a;background:#f6f5f3}
header{background:#6b1220;color:#fff;padding:13px 22px;display:flex;align-items:baseline;gap:16px;flex-wrap:wrap}
header h1{font-size:17px;margin:0;font-weight:600;letter-spacing:.5px}header .sub{font-size:12px;opacity:.82}
header .dash{margin-left:auto;color:#ffd9a8;font-size:12.5px;text-decoration:none;border:1px solid #a34a54;
 border-radius:5px;padding:4px 12px;white-space:nowrap}
header .dash:hover{background:#fff;color:#6b1220;border-color:#fff}
.hx{border:1px solid #e2dad2;border-radius:6px;padding:10px 14px;margin:0 0 12px;background:#fbfaf9;font-size:13px}
.hx b{color:#6b1220}
.hx .ch{margin:8px 0 0;padding-left:14px;border-left:2px solid #ddd}
.hx .ch div{margin:1px 0;color:#444}
.hx .ch em{font-style:normal;color:#999;font-size:12px;margin-right:6px}
.hx a{color:#6b1220}
.wrap{display:flex;height:calc(100vh - 50px)}
.side{width:346px;min-width:280px;border-right:1px solid #ddd;background:#fff;display:flex;flex-direction:column}
.ctl{padding:12px;border-bottom:1px solid #eee;display:flex;flex-direction:column;gap:8px}
#q{width:100%;padding:9px 11px;border:1px solid #ccc;border-radius:5px;font-size:14px;font-family:inherit}
.row{display:flex;gap:9px;align-items:center;font-size:12px;color:#555;flex-wrap:wrap}
select{padding:5px;border:1px solid #ccc;border-radius:4px;font-family:inherit;font-size:12px}
#list{overflow:auto;flex:1}
.it{padding:10px 13px;border-bottom:1px solid #f0f0f0;cursor:pointer;border-left:3px solid transparent}
.it:hover{background:#faf7f5}.it.on{background:#f5ebeb;border-left-color:#6b1220}
.it b{font-size:13.5px}.nm{font-size:12px;color:#6b1220;margin-left:6px}
.mt{font-size:11px;color:#999}.qq{font-size:12px;color:#666;margin-top:3px;line-height:1.6}
#doc{flex:1;overflow:auto;padding:32px 46px 80px;background:#fff}
.hd{display:flex;align-items:flex-start;gap:16px;justify-content:space-between}
#doc h1{font-size:23px;margin:0 0 8px;color:#6b1220;line-height:1.5}
#cp{flex:none;background:#fff;border:1px solid #6b1220;color:#6b1220;border-radius:4px;padding:5px 12px;font-size:12px;cursor:pointer;font-family:inherit}
#cp:hover{background:#6b1220;color:#fff}
#doc h2{font-size:16px;margin:32px 0 10px;padding-left:10px;border-left:4px solid #6b1220}
#doc h3{font-size:14px;margin:22px 0 6px;color:#444}
.src{font-size:12px;color:#777;padding-bottom:10px;border-bottom:1px solid #eee}.src a{color:#6b1220}
.src2{font-size:12px;color:#666;padding:8px 0;border-bottom:1px solid #f0f0f0}
.bd{white-space:pre-wrap;margin:0 0 14px;text-align:justify;max-width:54em}
.op{white-space:pre-wrap;margin:0 0 18px;max-width:54em;background:#fbfaf9;border-left:3px solid #ddd;padding:10px 14px;font-size:14px}
.pn{display:inline-block;min-width:2em;color:#a05;font-size:12px;vertical-align:top}
.warn{background:#fff4f4;border-left:3px solid #c33;padding:8px 12px;font-size:13px;margin-bottom:14px}
.laws .lw{display:inline-block;background:#f1ede9;border:1px solid #e2dad2;border-radius:3px;padding:2px 9px;margin:0 6px 6px 0;font-size:13px}
mark{background:#ffe9a8;padding:0 1px}
.ai{background:#fffaf0;border:1px solid #f0dfc0;border-left:4px solid #d79a2b;border-radius:4px;padding:14px 18px;margin:18px 0;max-width:54em}
.ai .cap{font-size:12px;color:#8a6114;font-weight:600;margin-bottom:8px}
.ai h4{margin:14px 0 4px;font-size:13.5px;color:#7a5410}
.ai p{margin:0 0 6px;white-space:pre-wrap}
.ai .one{font-size:15px;font-weight:600;color:#5a3f0c}
.chips{display:flex;flex-wrap:wrap;gap:6px;align-items:center;margin:14px 0 4px;padding-bottom:12px;border-bottom:1px solid #eee}
.chips .lb{font-size:12px;color:#888;margin-right:2px}
.chip{font-size:12.5px;padding:4px 12px;border:1px solid #d6cec6;border-radius:14px;cursor:pointer;background:#fff;color:#888;user-select:none}
.chip:hover{border-color:#6b1220}
.chip.on{background:#6b1220;border-color:#6b1220;color:#fff}
.chip.mini{padding:3px 9px;font-size:11.5px;border-radius:4px;color:#6b1220;border-color:#c9b3b3}
.ck{margin-right:7px;vertical-align:-1px;cursor:pointer}
.cmpbar{background:#f3e9e9;border-bottom:1px solid #e0cfcf;padding:8px 12px;font-size:12px;display:flex;gap:6px;align-items:center;flex-wrap:wrap}
.cmpbar .names{width:100%;color:#7a5c5c;line-height:1.6;max-height:3.4em;overflow:auto}
.cmpbar b{color:#6b1220}
.cmpbar button{font-size:11.5px;padding:3px 10px;border:1px solid #6b1220;background:#fff;color:#6b1220;border-radius:4px;cursor:pointer;font-family:inherit}
.cmpbar button:hover{background:#6b1220;color:#fff}
.cmptbl{border-collapse:collapse;width:max-content;min-width:100%;margin-top:14px}
.cmptbl th,.cmptbl td{border:1px solid #e0dad4;vertical-align:top;padding:9px 12px;text-align:left}
.cmptbl th.rh{background:#f7f4f1;font-size:13px;width:110px;min-width:110px;position:sticky;left:0;z-index:2;color:#6b1220}
.cmptbl thead th{background:#6b1220;color:#fff;font-size:13px;min-width:330px;max-width:430px}
.cmptbl thead th.rh{background:#6b1220;color:#fff}
.cmptbl td{min-width:330px;max-width:430px;font-size:13.5px;line-height:1.8}
.cmpcell{max-height:340px;overflow:auto;white-space:pre-wrap}
.cmpwrap{overflow:auto}
#advBtn{width:100%;padding:6px;border:1px solid #6b1220;background:#fff;color:#6b1220;border-radius:4px;cursor:pointer;font-family:inherit;font-size:12.5px}
#advBtn:hover{background:#6b1220;color:#fff}
.mask{position:fixed;inset:0;background:rgba(0,0,0,.42);display:flex;align-items:flex-start;justify-content:center;z-index:99;padding:34px 16px;overflow:auto}
.modal{background:#fff;border-radius:7px;width:min(880px,100%);box-shadow:0 12px 40px rgba(0,0,0,.3)}
.modal h3{margin:0;padding:14px 20px;background:#6b1220;color:#fff;font-size:15px;border-radius:7px 7px 0 0}
.grid{display:grid;grid-template-columns:1fr 1fr;gap:12px 18px;padding:18px 20px}
.grid .f{display:flex;flex-direction:column;gap:4px}
.grid .f.wide{grid-column:1/-1}
.grid label{font-size:12px;color:#6b1220;font-weight:600}
.grid input,.grid select{padding:7px 9px;border:1px solid #ccc;border-radius:4px;font-family:inherit;font-size:13px}
.grid .hint{font-size:11px;color:#999;font-weight:400}
.mfoot{padding:12px 20px;border-top:1px solid #eee;display:flex;gap:8px;justify-content:flex-end;align-items:center}
.mfoot button{padding:7px 18px;border-radius:4px;cursor:pointer;font-family:inherit;font-size:13px;border:1px solid #6b1220}
.mfoot .go{background:#6b1220;color:#fff}.mfoot .cl{background:#fff;color:#6b1220}
.conds{padding:8px 12px;border-bottom:1px solid #eee;display:flex;flex-wrap:wrap;gap:5px;align-items:center;background:#fcfaf9}
.conds .lb{font-size:11.5px;color:#888}
.cond{font-size:11.5px;background:#6b1220;color:#fff;border-radius:12px;padding:3px 8px 3px 10px;display:inline-flex;gap:6px;align-items:center}
.cond i{font-style:normal;cursor:pointer;opacity:.75}.cond i:hover{opacity:1}
.cond b{font-weight:600;opacity:.8}
.conds .clr{font-size:11.5px;color:#6b1220;cursor:pointer;text-decoration:underline;margin-left:auto}
.rng{display:flex;gap:6px;align-items:center}.rng input{width:100%}
.jump{display:flex;flex-wrap:wrap;gap:6px;align-items:center;margin:0 0 6px}
.jump .lb{font-size:12px;color:#888;margin-right:2px}
.jump a{font-size:12.5px;padding:4px 11px;border:1px solid #c9b3b3;border-radius:4px;color:#6b1220;text-decoration:none;background:#fff}
.jump a:hover{background:#6b1220;color:#fff;border-color:#6b1220}
.opbox{border:1px solid #e6dfd9;border-radius:5px;margin:0 0 8px;max-width:54em;overflow:hidden}
.ophd{display:flex;gap:8px;align-items:flex-start;padding:9px 12px;background:#faf7f5;cursor:pointer;font-size:13.5px}
.ophd:hover{background:#f3ebe8}
.ophd .tw{flex:1}
.ophd .ar{color:#6b1220;font-size:12px;flex:none;margin-top:2px}
.ophd .tg{font-size:11px;color:#fff;background:#a08;border-radius:3px;padding:1px 6px;flex:none;margin-top:2px}
.opbody{padding:10px 14px;border-top:1px solid #eee;white-space:pre-wrap;font-size:14px}
.opbar{display:flex;gap:8px;align-items:center;margin:6px 0 12px;flex-wrap:wrap}
.opbar button{font-size:11.5px;padding:4px 11px;border:1px solid #6b1220;background:#fff;color:#6b1220;border-radius:4px;cursor:pointer;font-family:inherit}
.opbar button:hover{background:#6b1220;color:#fff}
.opbar .n{font-size:11.5px;color:#888}
@media(max-width:760px){.wrap{flex-direction:column;height:auto}.side{width:100%;height:46vh}#doc{padding:20px}.grid{grid-template-columns:1fr}}"""

APP = r"""(function(){
const D=JSON.parse(document.getElementById('data').textContent);
const $=s=>document.querySelector(s); let cur=null,hits=D;
const esc=s=>(s||'').replace(/[&<>]/g,c=>({'&':'&amp;','<':'&lt;','>':'&gt;'}[c]));
const FIELDS=[['all','全部欄位'],['id','字號・案名'],['issue','爭點・案由'],['main','解釋文・主文'],
              ['reason','理由書・理由'],['op','意見書'],['law','相關法條'],['ai','白話解說'],['misc','其他區塊'],['judge','大法官']];
const SUB=['id','issue','main','reason','op','law','ai','misc','judge'];
const SECS=[['issue','爭點'],['main','解釋文・主文'],['reason','理由書・理由'],
            ['op','意見書'],['law','相關法條'],['ai','白話解說'],['misc','其他區塊']];
const MAIN=['解釋文','主文'], REASON=['理由書','理由','判決理由'];
const OPTYPES=['協同意見書','不同意見書','部分協同意見書','部分不同意見書','部分協同部分不同意見書','部分不同部分協同意見書','一部協同一部不同意見書'];
let ON={issue:1,main:1,reason:1,op:1,law:1,ai:1,misc:1};
let CMP=[];
let ADV={};   // 進階條件
function secs(r,names){ return r.s.filter(x=>names.indexOf(x[0])>=0); }
function fld(r,f){
  if(!r._f) r._f={};
  if(r._f[f]!==undefined) return r._f[f];
  let v='';
  if(f==='id') v=r.z+' '+r.n;
  else if(f==='issue') v=r.q;
  else if(f==='main') v=secs(r,MAIN).map(x=>x[1].map(p=>p[1]).join('\n')).join('\n');
  else if(f==='reason') v=secs(r,REASON).map(x=>x[1].map(p=>p[1]).join('\n')).join('\n');
  else if(f==='op') v=r.o.map(o=>o.t+'\n'+(o.x||'')+'\n'+(o.ai?o.ai.one+' '+o.ai.main+' '+o.ai.diff+' '+(o.ai.struct||'')+' '+(o.ai.guide||''):'')).join('\n');
  else if(f==='law') v=r.l.join('　');
  else if(f==='ai') v=r.ai?[r.ai.one,r.ai.issue,r.ai.holding,r.ai.place].join('\n'):'';
  else if(f==='misc') v=r.s.filter(x=>MAIN.indexOf(x[0])<0&&REASON.indexOf(x[0])<0).map(x=>x[0]+'\n'+x[1].map(p=>p[1]).join('\n')).join('\n');
  else if(f==='judge') v=r.j?[].concat(r.j.w,r.j.p,r.j.a,r.j.s||[],r.j.c?[r.j.c]:[]).join('　'):'';
  if(f==='id'||f==='issue'||f==='op'){ if(v.indexOf('大法官')>=0) v=v+'\n'+v.split('大法官').join(''); }
  r._f[f]=v; return v;
}
function terms(q){
  return (q||'').split(/\s+/).filter(Boolean).map(function(t){
    if(t.indexOf('大法官')<0) return t;
    var v=t.split('大法官').join('');
    return v.length>=2 ? v : t;
  });
}
function hit(r,f,ts){
  if(!ts.length) return true;
  if(f==='all') return ts.every(t=>SUB.some(x=>fld(r,x).indexOf(t)>=0));
  const v=fld(r,f); return ts.every(t=>v.indexOf(t)>=0);
}
function judgeHit(r){
  if(!ADV.judge) return true;
  const ts=terms(ADV.judge); if(!ts.length) return true;
  const j=r.j||{w:[],p:[],a:[]};
  const pool = ADV.jrole==='w'?j.w : ADV.jrole==='p'?j.p : ADV.jrole==='a'?j.a : ADV.jrole==='s'?(j.s||[]) : [].concat(j.w,j.p,j.a,j.s||[],j.c?[j.c]:[]);
  const v=pool.join('　');
  return ts.every(t=>v.indexOf(t)>=0);
}
function advHit(r){
  for(const f of ['id','issue','main','reason','op','law','ai','misc']){
    if(ADV[f] && !hit(r,f,terms(ADV[f]))) return false;
    if(ADV['not_'+f]){ const v=fld(r,f); if(terms(ADV['not_'+f]).some(t=>v.indexOf(t)>=0)) return false; }
  }
  if(!judgeHit(r)) return false;
  if(ADV.not_judge){ const v=fld(r,'judge'); if(terms(ADV.not_judge).some(t=>v.indexOf(t)>=0)) return false; }
  if(ADV.otype){ const ts=(r.j&&r.j.t)||[]; if(ts.indexOf(ADV.otype)<0) return false; }
  if(ADV.y1 && +r.dt.slice(0,4) < +ADV.y1) return false;
  if(ADV.y2 && +r.dt.slice(0,4) > +ADV.y2) return false;
  if(ADV.nop && r.o.length < +ADV.nop) return false;
  return true;
}
function mark(t,ts){ let h=esc(t); if(!ts.length)return h;
  const all=[]; ts.forEach(x=>{ all.push(x); if(x.length>=2) all.push(x.slice(0,1)+'大法官'+x.slice(1)); });
  all.forEach(x=>{h=h.replace(new RegExp(x.replace(/[.*+?^${}()|[\]\\]/g,'\\$&'),'g'),'<mark>$&</mark>');}); return h; }
function allTerms(){
  let ts=terms($('#q').value.trim());
  ['id','issue','main','reason','op','law','ai','misc','judge'].forEach(f=>{ if(ADV[f]) ts=ts.concat(terms(ADV[f])); });
  return ts;
}
function snip(r,f,ts){
  if(!ts.length) return r.q.slice(0,88);
  const order = f==='all' ? ['issue','id','main','reason','ai','law','op','misc','judge'] : [f];
  for(const x of order){ const v=fld(r,x); const i=v.indexOf(ts[0]);
    if(i>=0){ const a=Math.max(0,i-30); return (a?'…':'')+v.slice(a,a+96).replace(/\n/g,' ')+'…'; } }
  return r.q.slice(0,88);
}
function cite(r){ return (r.ty==='釋字'?'司法院'+r.z+'解釋':'憲法法庭'+r.z+'判決')+(r.n?'【'+r.n+'】':''); }

// ---------------- 進階篩選面板 ----------------
const ADVF=[['id','字號・案名'],['issue','爭點・案由'],['main','解釋文・主文'],['reason','理由書・理由'],
            ['op','意見書內容'],['law','相關法條'],['ai','白話解說'],['misc','其他區塊']];
function openAdv(){
  const m=document.createElement('div'); m.className='mask'; m.id='mask';
  let g=ADVF.map(f=>'<div class="f"><label>'+f[1]+'</label>'
    +'<div class="rng"><input data-a="'+f[0]+'" value="'+esc(ADV[f[0]]||'')+'" placeholder="含（留空不限）">'
    +'<input data-a="not_'+f[0]+'" value="'+esc(ADV['not_'+f[0]]||'')+'" placeholder="不含" style="background:#fff8f8"></div></div>').join('');
  g+='<div class="f"><label>大法官姓名 <span class="hint">可輸入「林俊益」或「林大法官俊益」</span></label>'
    +'<div class="rng"><input data-a="judge" value="'+esc(ADV.judge||'')+'" placeholder="含（留空不限）">'
    +'<input data-a="not_judge" value="'+esc(ADV.not_judge||'')+'" placeholder="不含" style="background:#fff8f8"></div></div>';
  g+='<div class="f"><label>大法官身分</label><select data-a="jrole">'
    +[['','不限（提出、加入、主筆或出席）'],['p','提出意見書'],['a','加入意見書'],['w','主筆（憲判）'],['s','出席（釋字1–79）']]
      .map(o=>'<option value="'+o[0]+'"'+(ADV.jrole===o[0]?' selected':'')+'>'+o[1]+'</option>').join('')+'</select></div>';
  g+='<div class="f"><label>意見書類型</label><select data-a="otype"><option value="">不限</option>'
    +OPTYPES.map(o=>'<option value="'+o+'"'+(ADV.otype===o?' selected':'')+'>'+o+'</option>').join('')+'</select></div>';
  g+='<div class="f"><label>意見書篇數（至少）</label><input data-a="nop" type="number" min="0" value="'+esc(ADV.nop||'')+'" placeholder="留空表示不限"></div>';
  g+='<div class="f wide"><label>公布年（西元）</label><div class="rng"><input data-a="y1" type="number" placeholder="自" value="'+esc(ADV.y1||'')+'">'
    +'<span style="color:#999">至</span><input data-a="y2" type="number" placeholder="至" value="'+esc(ADV.y2||'')+'"></div></div>';
  m.innerHTML='<div class="modal"><h3>進階篩選　各欄可同時指定，條件之間為「且」</h3><div class="grid">'+g+'</div>'
    +'<div class="mfoot"><span style="flex:1;font-size:11.5px;color:#999">同一欄內以空白分隔多個詞，亦為「且」</span>'
    +'<button class="cl" id="advClear">全部清除</button><button class="cl" id="advCancel">取消</button><button class="go" id="advGo">送出</button></div></div>';
  document.body.appendChild(m);
  m.addEventListener('click',e=>{ if(e.target===m) m.remove(); });
  $('#advCancel').onclick=()=>m.remove();
  $('#advClear').onclick=()=>{ ADV={}; m.remove(); search(); };
  $('#advGo').onclick=function(){
    const n={};
    m.querySelectorAll('[data-a]').forEach(function(el){ const v=(el.value||'').trim(); if(v) n[el.dataset.a]=v; });
    if(!n.judge) delete n.jrole;
    ADV=n; m.remove(); search();
  };
  m.querySelector('input').focus();
}
function condLabel(k){
  const f=ADVF.find(x=>x[0]===k); if(f) return f[1];
  if(k.slice(0,4)==='not_'){ const b=condLabel(k.slice(4)); return b+'（不含）'; }
  return {judge:'大法官',jrole:'身分',otype:'意見書類型',nop:'意見書篇數≥',y1:'公布年自',y2:'公布年至'}[k]||k;
}
function condVal(k,v){
  if(k==='jrole') return {p:'提出',a:'加入',w:'主筆',s:'出席'}[v]||v;
  return v;
}
function conds(){
  const c=$('#conds'); const ks=Object.keys(ADV).filter(k=>ADV[k]!=='' && !(k==='jrole'&&!ADV.judge));
  ks.sort();
  if(!ks.length){ c.style.display='none'; return; }
  c.style.display='flex';
  c.innerHTML='<span class="lb">篩選條件</span>'
    + ks.map(k=>'<span class="cond"><b>'+condLabel(k)+'</b>'+esc(condVal(k,ADV[k]))+'<i data-x="'+k+'" title="移除">×</i></span>').join('')
    + '<span class="clr" id="condClr">全部清除</span>';
  c.querySelectorAll('[data-x]').forEach(function(e){ e.onclick=function(){ delete ADV[e.dataset.x]; if(e.dataset.x==='judge') delete ADV.jrole; search(); }; });
  $('#condClr').onclick=function(){ ADV={}; search(); };
}

function search(){ const q=$('#q').value.trim(),ty=$('#ty').value,yr=$('#yr').value,op=$('#op').checked,f=$('#fd').value;
  const ts=terms(q);
  hits=D.filter(r=>{ if(ty&&r.ty!==ty)return false; if(op&&!r.o.length)return false;
    if(yr){const y=+r.dt.slice(0,4); if(yr==='a'&&y>=1990)return false; if(yr==='b'&&(y<1990||y>=2005))return false; if(yr==='c'&&(y<2005||y>=2015))return false; if(yr==='d'&&y<2015)return false;}
    if(!hit(r,f,ts)) return false;
    return advHit(r); });
  $('#cnt').textContent=hits.length+' 筆';
  const mts=allTerms();
  $('#list').innerHTML=hits.slice(0,500).map(r=>'<div class="it'+(cur&&cur.k===r.k?' on':'')+'" data-k="'+r.k+'">'
    +'<b><input type="checkbox" class="ck" data-k="'+r.k+'"'+(CMP.indexOf(r.k)>=0?' checked':'')+' title="加入比較">'+esc(r.z)+'</b>'
    +(r.n?'<span class="nm">'+esc(r.n)+'</span>':'')
    +'<div class="mt">'+esc(r.dt)+(r.o.length?' · 意見書 '+r.o.length:'')+'</div>'
    +'<div class="qq">'+mark(snip(r,f,mts),mts)+'</div></div>').join('')
    +(hits.length>500?'<div class="mt" style="padding:12px">僅列前 500 筆</div>':'');
  conds(); bar();
}
function bar(){
  const b=$('#cmpbar');
  if(!CMP.length){ b.style.display='none'; return; }
  b.style.display='flex';
  b.innerHTML='<b>比較清單 '+CMP.length+' 案</b><span style="flex:1"></span>'
    +'<button id="doCmp">產生比較表</button><button id="cpCmp">複製為表格</button><button id="clrCmp">清空</button>'
    +'<div class="names">'+CMP.map(k=>{const r=D.find(x=>x.k===k);return r?r.z:k;}).join('、')+'</div>';
  $('#doCmp').onclick=showCompare; $('#clrCmp').onclick=function(){CMP=[];search(); if(cur)show(cur.k);};
  $('#cpCmp').onclick=copyCompare;
}
function chips(){
  return '<div class="chips"><span class="lb">顯示區塊</span>'
    + SECS.map(x=>'<span class="chip'+(ON[x[0]]?' on':'')+'" data-s="'+x[0]+'">'+x[1]+'</span>').join('')
    + '<span class="chip mini" data-all="1">全選</span><span class="chip mini" data-none="1">全不選</span></div>';
}
function bindChips(after){
  document.querySelectorAll('.chips .chip').forEach(function(c){
    c.onclick=function(){
      if(c.dataset.all){ SECS.forEach(x=>ON[x[0]]=1); }
      else if(c.dataset.none){ SECS.forEach(x=>ON[x[0]]=0); }
      else { ON[c.dataset.s]=ON[c.dataset.s]?0:1; }
      after();
    };
  });
}
function judgeLine(r){
  const j=r.j; if(!j) return '';
  const bits=[];
  if(j.c) bits.push('<b>主席</b> '+esc(j.c));
  if(j.s&&j.s.length) bits.push('<b>出席大法官</b> '+esc(j.s.join('、')));
  if(j.w&&j.w.length) bits.push('<b>主筆</b> '+esc(j.w.join('、')));
  if(j.p&&j.p.length) bits.push('<b>提出意見書</b> '+esc(j.p.join('、')));
  if(j.a&&j.a.length) bits.push('<b>加入</b> '+esc(j.a.join('、')));
  return bits.length?'<div class="src2">'+bits.join(' ｜ ')+'</div>':'';
}
function judgeText(r){
  const j=r.j; if(!j) return '';
  const b=[];
  if(j.c) b.push('主席：'+j.c);
  if(j.s&&j.s.length) b.push('出席：'+j.s.join('、'));
  if(j.w&&j.w.length) b.push('主筆：'+j.w.join('、'));
  if(j.p&&j.p.length) b.push('提出意見書：'+j.p.join('、'));
  if(j.a&&j.a.length) b.push('加入：'+j.a.join('、'));
  return b.join('\n');
}
function show(k){ const r=D.find(x=>x.k===k); if(!r)return; cur=r; history.replaceState(null,'','#'+k);
  const ts=allTerms();
  let h='<div class="hd"><h1>'+esc(r.z)+(r.n?'【'+esc(r.n)+'】':'')+'</h1><button id="cp">複製引用</button></div>';
  h+='<div class="src">司法院憲法法庭網站 · <a href="'+r.u+'" target="_blank" rel="noopener">官網原文</a> · 公布 '+esc(r.roc||r.dt)+(r.decree?' · '+esc(r.decree):'')+' · 抓取於 '+esc((r.at||'').slice(0,10))+'</div>';
  if(r.m&&r.m.length) h+='<div class="src2">'+r.m.map(x=>'<b>'+esc(x[0])+'</b> '+esc(x[1])).join(' ｜ ')+'</div>';
  h+=judgeLine(r);
  h+=chips();
  const jump=[];
  let body='';
  if(ON.issue&&r.q){ const t1=(r.ty==='釋字'?'解釋爭點':'案由'); jump.push(['s_issue',t1]);
    body+='<h2 id="s_issue">'+t1+'</h2><div class="bd">'+mark(r.q,ts)+'</div>'; }
  r.s.forEach(function(p,ix){
    const key = MAIN.indexOf(p[0])>=0?'main':(REASON.indexOf(p[0])>=0?'reason':'misc');
    if(!ON[key]) return;
    const id='s_sec'+ix; jump.push([id,p[0]]);
    body+='<h2 id="'+id+'">'+esc(p[0])+'</h2>';
    p[1].forEach(function(b){ body+='<div class="bd">'+(b[0]?'<span class="pn">'+esc(b[0])+'</span>':'')+mark(b[1],ts)+'</div>'; }); });
  if(ON.op&&r.o.length){ jump.push(['s_op','意見書']);
    body+='<h2 id="s_op">意見書（'+r.o.length+' 篇）</h2>';
    body+='<div class="opbar"><button id="opAll">全部展開</button><button id="opNone">全部收合</button>'
        + '<button id="opCmp">比較所選意見書</button><span class="n" id="opN">已選 0 篇</span></div>';
    r.o.forEach(function(o,i){
      body+='<div class="opbox" data-i="'+i+'">'
        +'<div class="ophd"><input type="checkbox" class="opck" data-i="'+i+'" onclick="event.stopPropagation()">'
        +'<span class="ar" data-ar="'+i+'">▸</span>'
        +(o.ty?'<span class="tg">'+esc(o.ty)+'</span>':'')
        +'<span class="tw">'+(i+1)+'. '+mark(o.t,ts)+'</span>'
        +'<a href="'+o.h+'" target="_blank" rel="noopener" style="font-size:11.5px;flex:none" onclick="event.stopPropagation()">官網PDF</a></div>'
        +'<div class="opbody" data-b="'+i+'" style="display:none">'
        + (o.ai ? '<div class="ai" style="margin:0 0 12px"><div class="cap">⚠ 白話解說（AI 生成，非原文，不得引用）</div>'
             +'<p class="one">'+mark(o.ai.one,ts)+'</p>'
             +'<h4>核心主張</h4><p>'+mark(o.ai.main,ts)+'</p>'
             +'<h4>與多數意見的差異</h4><p>'+mark(o.ai.diff,ts)+'</p>'
             +(o.ai.struct?'<h4>論證結構</h4><p>'+mark(o.ai.struct,ts)+'</p>':'')
             +(o.ai.guide?'<h4>段落指引</h4><p>'+mark(o.ai.guide,ts)+'</p>':'')+'</div>' : '')
        + (o.x ? mark(o.x,ts) : '<span style="color:#c33">本篇 PDF 未能取得可用文字層，請點上方連結查閱官網原檔。</span>')
        +'</div></div>'; }); }
  if(ON.law&&r.l.length){ jump.push(['s_law','相關法條']);
    body+='<h2 id="s_law">相關法條</h2><div class="laws">'+r.l.map(x=>'<span class="lw">'+mark(x,ts)+'</span>').join('')+'</div>'; }
  if(ON.ai&&r.ai){ jump.push(['s_ai','白話解說']);
    body+='<h2 id="s_ai">白話解說</h2><div class="ai"><div class="cap">⚠ 本區塊由 AI 依上方原文撰寫，非官方原文。撰寫論文請引用原文，不得引用本區塊。</div>'
    +'<p class="one">'+mark(r.ai.one,ts)+'</p>'
    +'<h4>爭點拆解</h4><p>'+mark(r.ai.issue,ts)+'</p>'
    +'<h4>結論與理由</h4><p>'+mark(r.ai.holding,ts)+'</p>'
    +(r.ai.place?'<h4>在體系中的位置</h4><p>'+mark(r.ai.place,ts)+'</p>':'')+'</div>'; }
  if(r.hx){ jump.push(['s_hx','案件歷程']);
    const dash='研究儀表板.html#his/'+encodeURIComponent(r.k);
    body+='<h2 id="s_hx">案件歷程</h2><div class="hx">'
      +'<b>'+r.hx.n+'</b> 筆確定終局裁判與前審紀錄，構成 <b>'+r.hx.nc+'</b> 條獨立審級鏈'
      +(r.hx.m?'，另有 '+r.hx.m+' 筆未命中':'')+'。'
      +'　<a href="'+dash+'" target="_blank" rel="noopener">在研究儀表板展開 →</a>'
      +r.hx.c.map(function(c){ return '<div class="ch">'+c.map(function(s){
          return '<div'+(s[2]?' style="margin-left:'+(s[2]*18)+'px"':'')+'>'
                 +'<em>'+esc(s[0])+'</em>'+esc(s[1])+'</div>'; }).join('')+'</div>'; }).join('')
      +'<div style="color:#999;font-size:12px;margin-top:8px">'
      +'串接自司法院終結案件資料（DisposedCase 多年期），非官網解釋文內容。'
      +'該資料集不含當事人姓名，故無法以個人為單位追蹤。</div></div>'; }
  if(jump.length>1) h+='<div class="jump"><span class="lb">跳至</span>'
      + jump.map(x=>'<a href="#'+x[0]+'" data-j="'+x[0]+'">'+esc(x[1])+'</a>').join('')+'</div>';
  h+=body;
  const dv=$('#doc'); dv.innerHTML=h; dv.scrollTop=0;
  dv.querySelectorAll('[data-j]').forEach(function(a){
    a.onclick=function(e){ e.preventDefault(); const el=document.getElementById(a.dataset.j);
      if(el) dv.scrollTop = el.offsetTop - dv.offsetTop - 8; }; });
  bindOpinions(r,ts);
  bindChips(function(){show(k);});
  $('#cp').onclick=function(){navigator.clipboard.writeText(cite(r)+'，'+r.u).then(function(){$('#cp').textContent='已複製';setTimeout(function(){$('#cp').textContent='複製引用';},1500);});};
  search(); }

function bindOpinions(r,ts){
  const dv=$('#doc'); const sel=new Set();
  function setN(){ const e=document.getElementById('opN'); if(e) e.textContent='已選 '+sel.size+' 篇'; }
  function toggle(i,force){
    const b=dv.querySelector('[data-b="'+i+'"]'); const a=dv.querySelector('[data-ar="'+i+'"]');
    if(!b) return;
    const openIt = force===undefined ? b.style.display==='none' : force;
    b.style.display = openIt?'block':'none';
    if(a) a.textContent = openIt?'▾':'▸';
  }
  dv.querySelectorAll('.ophd').forEach(function(hd){
    hd.onclick=function(){ toggle(hd.parentNode.dataset.i); };
  });
  dv.querySelectorAll('.opck').forEach(function(c){
    c.onchange=function(){ if(c.checked) sel.add(+c.dataset.i); else sel.delete(+c.dataset.i); setN(); };
  });
  const all=document.getElementById('opAll'), none=document.getElementById('opNone'), cmp=document.getElementById('opCmp');
  if(all) all.onclick=function(){ r.o.forEach(function(_,i){ toggle(i,true); }); };
  if(none) none.onclick=function(){ r.o.forEach(function(_,i){ toggle(i,false); }); };
  if(cmp) cmp.onclick=function(){
    const idx=[...sel].sort((a,b)=>a-b);
    if(idx.length<1){ alert('請先勾選要比較的意見書'); return; }
    showOpCompare(r,idx,ts);
  };
}
function opCell(o,key){
  if(key==='ty')   return o.ty||'（未標類型）';
  if(key==='who')  return ((o.p&&o.p.length)?'提出：'+o.p.join('、'):'（未標作者）')+((o.a&&o.a.length)?'\n加入：'+o.a.join('、'):'');
  if(key==='ai')   return o.ai?('【立場】'+o.ai.one+'\n\n【核心主張】'+o.ai.main+'\n\n【與多數意見的差異】'+o.ai.diff
      +(o.ai.struct?'\n\n【論證結構】'+o.ai.struct:'')+(o.ai.guide?'\n\n【段落指引】'+o.ai.guide:'')):'（尚未產生白話解說）';
  if(key==='x')    return o.x||'（本篇無可用文字層）';
  return '';
}
function showOpCompare(r,idx,ts){
  const OPROWS=[['ty','意見書類型'],['who','提出／加入'],['ai','白話解說'],['x','全文']];
  const os=idx.map(i=>r.o[i]);
  let h='<div class="hd"><h1>意見書比較　'+esc(r.z)+'（'+os.length+' 篇）</h1><button id="cp">複製為表格</button></div>';
  h+='<div class="src">同一則之意見書並排對照。全文逐字照錄，白話解說列為 AI 撰寫，不得引用。'
    +' <a href="#" id="backCase">← 回到本則內文</a></div>';
  h+='<div class="cmpwrap"><table class="cmptbl"><thead><tr><th class="rh">欄位</th>'
    + os.map((o,n)=>'<th>'+(idx[n]+1)+'. '+esc(o.t.slice(0,40))+(o.t.length>40?'…':'')+'</th>').join('')+'</tr></thead><tbody>';
  OPROWS.forEach(function(f){
    h+='<tr><th class="rh">'+f[1]+'</th>'+os.map(function(o){
      const t=opCell(o,f[0]);
      return '<td><div class="cmpcell">'+mark(t,ts)+'</div></td>'; }).join('')+'</tr>';
  });
  h+='<tr><th class="rh">官網PDF</th>'+os.map(o=>'<td><a href="'+o.h+'" target="_blank" rel="noopener">開啟原檔</a></td>').join('')+'</tr>';
  h+='</tbody></table></div>';
  const dv=$('#doc'); dv.innerHTML=h; dv.scrollTop=0;
  document.getElementById('backCase').onclick=function(e){ e.preventDefault(); show(r.k); };
  document.getElementById('cp').onclick=function(){
    const e2=t=>(t||'').replace(/\|/g,'\\|').replace(/\n/g,'<br>');
    let md='| 欄位 | '+os.map((o,n)=>(idx[n]+1)+'. '+o.t).join(' | ')+' |\n|---|'+os.map(()=>'---').join('|')+'|\n';
    OPROWS.forEach(function(f){ md+='| '+f[1]+' | '+os.map(o=>e2(opCell(o,f[0]))).join(' | ')+' |\n'; });
    md+='| 官網PDF | '+os.map(o=>o.h).join(' | ')+' |\n';
    md='### '+cite(r)+' 意見書比較\n\n'+md;
    navigator.clipboard.writeText(md).then(function(){ const b=document.getElementById('cp'); b.textContent='已複製'; setTimeout(function(){b.textContent='複製為表格';},1500); });
  };
}
function cellText(r,key){
  if(key==='issue') return r.q||'';
  if(key==='law')   return r.l.join('\n');
  if(key==='ai')    return r.ai?('【一句話】'+r.ai.one+'\n\n【爭點拆解】'+r.ai.issue+'\n\n【結論與理由】'+r.ai.holding+(r.ai.place?'\n\n【體系定位】'+r.ai.place:'')):'';
  if(key==='op')    return r.o.length?r.o.map((o,i)=>(i+1)+'. '+o.t).join('\n'):'（無意見書）';
  const names = key==='main'?MAIN:(key==='reason'?REASON:null);
  const list = names? secs(r,names) : r.s.filter(x=>MAIN.indexOf(x[0])<0&&REASON.indexOf(x[0])<0);
  return list.map(x=>x[0]+'\n'+x[1].map(p=>(p[0]?'（'+p[0]+'）':'')+p[1]).join('\n')).join('\n\n');
}
function cmpRows(){ return SECS.filter(x=>ON[x[0]]); }
function showCompare(){
  const rs=CMP.map(k=>D.find(x=>x.k===k)).filter(Boolean);
  if(!rs.length) return;
  const ts=allTerms();
  let h='<div class="hd"><h1>多案比較（'+rs.length+' 案）</h1><button id="cp">複製為表格</button></div>';
  h+='<div class="src">比較的欄位由下方「顯示區塊」控制，取消勾選即可把該列從表中拿掉。原文逐字照錄，白話解說列不得引用。</div>';
  h+=chips();
  h+='<div class="cmpwrap"><table class="cmptbl"><thead><tr><th class="rh">欄位</th>'
    + rs.map(r=>'<th>'+esc(r.z)+(r.n?'<br><span style="font-weight:400;font-size:12px">【'+esc(r.n)+'】</span>':'')+'</th>').join('')+'</tr></thead><tbody>';
  h+='<tr><th class="rh">公布日期</th>'+rs.map(r=>'<td>'+esc(r.roc||r.dt)+'</td>').join('')+'</tr>';
  h+='<tr><th class="rh">大法官</th>'+rs.map(r=>'<td><div class="cmpcell">'+(judgeText(r)?mark(judgeText(r),ts):'<span style="color:#aaa">（無）</span>')+'</div></td>').join('')+'</tr>';
  cmpRows().forEach(function(f){
    h+='<tr><th class="rh">'+f[1]+'</th>'+rs.map(function(r){
      const t=cellText(r,f[0]);
      return '<td><div class="cmpcell">'+(t?mark(t,ts):'<span style="color:#aaa">（無）</span>')+'</div></td>'; }).join('')+'</tr>';
  });
  h+='</tbody></table></div>';
  const dv=$('#doc'); dv.innerHTML=h; dv.scrollTop=0;
  bindChips(showCompare);
  $('#cp').onclick=copyCompare;
}
function copyCompare(){
  const rs=CMP.map(k=>D.find(x=>x.k===k)).filter(Boolean);
  if(!rs.length) return;
  const esc2=t=>(t||'').replace(/\|/g,'\\|').replace(/\n/g,'<br>');
  let md='| 欄位 | '+rs.map(r=>cite(r)).join(' | ')+' |\n|---|'+rs.map(()=>'---').join('|')+'|\n';
  md+='| 公布日期 | '+rs.map(r=>r.roc||r.dt).join(' | ')+' |\n';
  md+='| 大法官 | '+rs.map(r=>esc2(judgeText(r))).join(' | ')+' |\n';
  cmpRows().forEach(function(f){ md+='| '+f[1]+' | '+rs.map(r=>esc2(cellText(r,f[0]))).join(' | ')+' |\n'; });
  md+='| 原始出處 | '+rs.map(r=>r.u).join(' | ')+' |\n';
  navigator.clipboard.writeText(md).then(function(){
    const b=document.getElementById('cp')||document.getElementById('cpCmp');
    if(b){const o=b.textContent;b.textContent='已複製';setTimeout(function(){b.textContent=o;},1500);} });
}
$('#fd').innerHTML=FIELDS.map(f=>'<option value="'+f[0]+'">'+f[1]+'</option>').join('');
['#q','#ty','#yr','#op','#fd'].forEach(function(s){$(s).addEventListener(s==='#q'?'input':'change',search);});
$('#advBtn').onclick=openAdv;
$('#list').addEventListener('click',function(e){
  const ck=e.target.closest('.ck');
  if(ck){ const k=ck.dataset.k; const i=CMP.indexOf(k); if(i>=0)CMP.splice(i,1); else CMP.push(k); bar(); e.stopPropagation(); return; }
  const it=e.target.closest('.it'); if(it)show(it.dataset.k);
});
search(); show(decodeURIComponent(location.hash.slice(1))||'釋字0001');
})();"""


def write_html(recs, lite=False, filename=None):
    """建置並寫檔。只要 HTML 字串請用 build_html。"""
    doc = build_html(recs, lite=lite)
    out = ROOT / (filename or "釋字及憲法法庭資料庫.html")
    out.write_text(doc, encoding="utf-8")
    return out, len(doc)


def build_html(recs, lite=False):
    if lite:
        import copy
        recs = [{**r, "o": [{k2: v2 for k2, v2 in o.items() if k2 != "x"} for o in r["o"]]} for r in recs]
    nj = sum(1 for r in recs if r["ty"] == "釋字")
    nd = len(recs) - nj
    no = sum(len(r["o"]) for r in recs)
    data = json.dumps(recs, ensure_ascii=False).replace("<", "\\u003c")
    doc = ("<!DOCTYPE html><html lang=\"zh-Hant\"><head><meta charset=\"utf-8\">"
           "<meta name=\"viewport\" content=\"width=device-width,initial-scale=1\">"
           "<title>釋字及憲法法庭資料庫</title><style>" + CSS + "</style></head><body>"
           "<header><h1>釋字及憲法法庭資料庫</h1><span class=\"sub\">大法官解釋 " + str(nj) +
           " 則 · 憲法法庭判決 " + str(nd) + " 則 · 意見書 " + str(no) +
           " 篇 · 逐字取自司法院憲法法庭網站" + ("（輕量版：意見書僅列連結）" if lite else "") + "</span>"
           "<a class=\"dash\" href=\"研究儀表板.html\" target=\"_blank\" rel=\"noopener\">研究儀表板 →</a></header>"
           "<div class=\"wrap\"><div class=\"side\"><div class=\"ctl\">"
           "<input id=\"q\" placeholder=\"快速檢索…\">"
           "<div class=\"row\"><label style=\"color:#666\">範圍</label><select id=\"fd\"></select>"
           "<select id=\"ty\"><option value=\"\">全部</option><option value=\"釋字\">釋字</option>"
           "<option value=\"憲判\">憲判</option></select></div>"
           "<div class=\"row\"><select id=\"yr\"><option value=\"\">不限年代</option><option value=\"a\">1949–1989</option>"
           "<option value=\"b\">1990–2004</option><option value=\"c\">2005–2014</option>"
           "<option value=\"d\">2015 以後</option></select>"
           "<label><input type=\"checkbox\" id=\"op\"> 僅有意見書</label><span id=\"cnt\"></span></div>"
           "<button id=\"advBtn\">進階篩選（可多欄同時指定）</button></div>"
           "<div id=\"conds\" class=\"conds\" style=\"display:none\"></div>"
           "<div id=\"cmpbar\" class=\"cmpbar\" style=\"display:none\"></div><div id=\"list\"></div></div><div id=\"doc\"></div></div>"
           "<script id=\"data\" type=\"application/json\">" + data + "</" + "script><script>" + APP + "</" + "script></body></html>")
    return doc


def write_report(recs, integrity, fetch_bad, pdf_bad, stats, nlaw, htmlsize):
    jie = sorted(int(r["k"][2:]) for r in recs if r["ty"] == "釋字")
    missing = sorted(set(range(1, 814)) - set(jie))
    prob = list(integrity)
    if missing:
        prob.append(f"釋字缺號 {len(missing)} 則：{missing[:30]}")
    if len([r for r in recs if r['ty'] == '憲判']) != 57:
        prob.append(f"憲判數量 {len([r for r in recs if r['ty']=='憲判'])}，與官網列表 57 則不符")
    for r in recs:
        if not r["s"]:
            prob.append(f'{r["z"]} 無內文區塊')
        if not r["dt"]:
            prob.append(f'{r["z"]} 無公布日期')
    prob += [f"抓取失敗 {x}" for x in fetch_bad] + [f"PDF 失敗 {x}" for x in pdf_bad]

    rep = ["# 校驗報告", "", f"建置時間：{time.strftime('%Y-%m-%d %H:%M:%S')}", "",
           f'- 大法官解釋：{len(jie)} / 813',
           f'- 憲法法庭判決：{len([r for r in recs if r["ty"]=="憲判"])} / 57',
           f'- 意見書篇數：{sum(len(r["o"]) for r in recs)}',
           f'- 意見書文字擷取：成功 {stats["ok"]}、低品質需 OCR {stats["低品質"]}、缺檔 {stats["缺檔"]}',
           f'- 法條節點：{nlaw}',
           f'- 網頁版大小：{htmlsize/1048576:.2f} MB', "",
           "## 逐頁完整性驗證", "",
           "每則釋字皆比對官網頁面自身標示之「第N筆／共813筆」與頁內「解釋字號」是否等於本庫指派之編號。", ""]
    if prob:
        rep += [f"## 不合格項目（{len(prob)}）", ""] + [f"- {x}" for x in prob[:200]]
    else:
        rep += ["## 全部通過", "", "- 編號齊全、序號與字號相符、內文與日期非空、抓取與 PDF 無失敗"]
    (ROOT / "00_索引").mkdir(exist_ok=True)
    (ROOT / "00_索引" / "校驗報告.md").write_text("\n".join(rep) + "\n", encoding="utf-8")
    return prob


def main():
    try:
        sys.stdout.reconfigure(encoding="utf-8")
    except Exception:
        pass
    print("=" * 60)
    print("釋字及憲法法庭資料庫 — 建置開始")
    print("根目錄：", ROOT)
    print("=" * 60)
    fetch_bad = step_fetch()
    print("[2/6] 解析與完整性驗證 …")
    recs, integrity = parse_all()
    print(f"      {len(recs)} 筆；完整性問題 {len(integrity)} 項")
    if len(recs) < 800:
        print(f"\n[中止] 只解析到 {len(recs)} 筆，遠少於應有的 870 筆。")
        print("為避免產生殘缺的筆記庫與網頁，這裡停住，不覆寫任何成品。")
        print("請先解決抓取問題後再執行一次；已抓到的檔案會保留，不會重抓。")
        sys.exit(1)
    pdf_bad = step_pdfs(recs)
    stats = step_extract(recs)
    print("[5/6] 產出 Markdown 筆記庫 …")
    nlaw = write_notes(recs)
    write_index(recs)
    print("[6/6] 產出離線網頁版 …")
    out, size = write_html(recs)
    prob = write_report(recs, integrity, fetch_bad, pdf_bad, stats, nlaw, size)
    json.dump(recs, open(PIPE / "資料庫.json", "w", encoding="utf-8"), ensure_ascii=False)
    print("=" * 60)
    print("完成。")
    print("  網頁版：", out)
    print("  筆記庫：", ROOT / "10_釋字", "／", ROOT / "20_憲判")
    print("  校驗報告：", ROOT / "00_索引" / "校驗報告.md")
    print("  不合格項目：", len(prob))
    print("=" * 60)


if __name__ == "__main__":
    main()