import argparse
import csv
import hashlib
import re
from collections import Counter, defaultdict
from datetime import datetime, timezone, timedelta
from pathlib import Path


METAL_ORDER = ["铜", "铝", "金", "银", "锂", "镍", "钴", "稀土", "锡", "钨", "钼", "锑"]
THEME_PRIORITY = ["price", "inventory", "supply", "demand", "cost", "project", "policy", "risk", "company"]
OPINION_WORDS = ["推荐", "建议关注", "相关标的", "有望", "预计", "或将", "看好", "维持", "评级"]
TABLE_WORDS = ["图", "表", "数据来源", "来源：", "根据", "库存", "价格", "产量", "产能", "同比", "环比"]
ADJACENT_OR_NOISE_WORDS = [
    "凤凰光学",
    "骏鼎达",
    "宏达电子",
    "银禧科技",
    "一汽解放",
    "佰维存储",
    "纳芯微",
    "AI服务器",
    "光通信",
    "半固态电池",
    "智能物联",
]


def read_csv(path):
    with path.open("r", encoding="utf-8-sig", newline="") as handle:
        return list(csv.DictReader(handle))


def write_csv(path, fieldnames, rows):
    path.parent.mkdir(parents=True, exist_ok=True)
    with path.open("w", encoding="utf-8-sig", newline="") as handle:
        writer = csv.DictWriter(handle, fieldnames=fieldnames)
        writer.writeheader()
        writer.writerows(rows)


def split_tags(value):
    return [tag for tag in (value or "").split("|") if tag]


def clean_text(value):
    return re.sub(r"\s+", " ", (value or "").strip())


def short_hash(path):
    return hashlib.sha256(path.read_bytes()).hexdigest()


def score_fact(row):
    metals = split_tags(row.get("metal_tags"))
    themes = split_tags(row.get("theme_tags"))
    text = row.get("evidence_text", "")
    score = 0
    if row.get("fact_type") == "indicator_sentence":
        score += 6
    if row.get("fact_type") == "risk_sentence":
        score += 3
    if re.search(r"\d", text):
        score += 4
    for theme in themes:
        if theme in THEME_PRIORITY:
            score += max(1, 10 - THEME_PRIORITY.index(theme))
    for metal in metals:
        if metal in METAL_ORDER:
            score += 2
    if any(word in text for word in TABLE_WORDS):
        score += 2
    if len(text) < 25:
        score -= 2
    return score


def load_pages(path):
    if not path.exists():
        return {}
    pages = defaultdict(list)
    current_page = "unknown"
    for line in path.read_text(encoding="utf-8", errors="ignore").splitlines():
        match = re.match(r"\[\[PAGE\s+(\d+)\]\]", line.strip())
        if match:
            current_page = match.group(1)
            continue
        cleaned = clean_text(line)
        if cleaned:
            pages[current_page].append(cleaned)
    return {page: " ".join(lines) for page, lines in pages.items()}


def find_context(row, project_root):
    converted = project_root / row.get("converted_text_path", "")
    pages = load_pages(converted)
    if not pages:
        return "NO_TEXT_FILE", "", ""
    page_no = row.get("page_no") or "unknown"
    text = clean_text(row.get("evidence_text", ""))
    candidates = []
    if page_no in pages:
        candidates.append((page_no, pages[page_no]))
    candidates.extend((page, body) for page, body in pages.items() if page != page_no)
    needle = text[:80]
    for page, body in candidates:
        idx = body.find(needle)
        if idx >= 0:
            start = max(0, idx - 180)
            end = min(len(body), idx + len(text) + 220)
            status = "TEXT_MATCH_CONTEXT_OK" if page == page_no else "TEXT_MATCH_DIFFERENT_PAGE"
            return status, page, body[start:end]
    tokens = [token for token in re.split(r"\W+", text) if len(token) >= 2]
    hit_count = sum(1 for token in tokens[:12] if token in pages.get(page_no, ""))
    if hit_count >= max(2, min(5, len(tokens) // 2)):
        return "TEXT_MATCH_PARTIAL_CONTEXT", page_no, pages.get(page_no, "")[:700]
    return "NO_TEXT_MATCH", page_no, pages.get(page_no, "")[:700]


def source_expression_type(text):
    if any(word in text for word in OPINION_WORDS):
        return "BROKER_VIEW_OR_ORIGINAL_RECOMMENDATION"
    if re.search(r"\d", text):
        return "NUMERIC_FACT_CANDIDATE"
    return "TEXT_FACT_OR_VIEW_CANDIDATE"


def scope_gate(text):
    compact = clean_text(text)
    if "........" in compact or re.search(r"^\d+(?:\.\d+)+\s+.*\.{5,}\s*\d+", compact):
        return "TOC_OR_INDEX_NOISE"
    if "公司公告" in compact or any(word in compact for word in ADJACENT_OR_NOISE_WORDS):
        return "MARKET_NEWS_OR_ADJACENT_REVIEW"
    return "CORE_INDUSTRY_CANDIDATE"


def next_action(row, review_status, expression_type):
    text = row.get("evidence_text", "")
    scope = scope_gate(text)
    if scope == "TOC_OR_INDEX_NOISE":
        return "EXCLUDE_FROM_CORE_EVIDENCE_AS_TOC_NOISE"
    if scope == "MARKET_NEWS_OR_ADJACENT_REVIEW":
        return "ROUTE_TO_MARKET_OR_ADJACENT_REVIEW_BEFORE_USE"
    themes = set(split_tags(row.get("theme_tags")))
    if review_status in {"NO_TEXT_FILE", "NO_TEXT_MATCH"}:
        return "CHECK_CONVERTED_OR_RAW_SOURCE"
    if expression_type == "BROKER_VIEW_OR_ORIGINAL_RECOMMENDATION":
        return "KEEP_AS_BROKER_VIEW_DO_NOT_REWRITE_AS_ANALYST_RECOMMENDATION"
    if {"price", "inventory", "supply", "demand", "cost", "project"} & themes or re.search(r"\d", text):
        return "ADD_PARAGRAPH_TABLE_OR_EXTERNAL_CROSS_CHECK"
    return "KEEP_DRAFT_OR_USE_AS_CONTEXT"


def upgrade_status(row, review_status, expression_type):
    scope = scope_gate(row.get("evidence_text", ""))
    if scope == "TOC_OR_INDEX_NOISE":
        return "EXCLUDE_FROM_CORE_EVIDENCE"
    if scope == "MARKET_NEWS_OR_ADJACENT_REVIEW":
        return "KEEP_DRAFT_SCOPE_REVIEW"
    if review_status not in {"TEXT_MATCH_CONTEXT_OK", "TEXT_MATCH_PARTIAL_CONTEXT", "TEXT_MATCH_DIFFERENT_PAGE"}:
        return "DATA_GAP_REVIEW"
    if expression_type == "BROKER_VIEW_OR_ORIGINAL_RECOMMENDATION":
        return "KEEP_DRAFT_BROKER_VIEW"
    if row.get("fact_type") == "indicator_sentence":
        return "CANDIDATE_FOR_EVIDENCE_CARD"
    return "KEEP_DRAFT"


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--project-root", default=".")
    parser.add_argument("--fact-files", nargs="+", required=True)
    parser.add_argument("--review-out", required=True)
    parser.add_argument("--manifest-out", required=True)
    parser.add_argument("--summary-out", required=True)
    parser.add_argument("--per-metal", type=int, default=8)
    parser.add_argument("--per-theme", type=int, default=12)
    args = parser.parse_args()

    project_root = Path(args.project_root).resolve()
    facts = []
    for fact_file in args.fact_files:
        path = project_root / fact_file
        if path.exists():
            facts.extend(read_csv(path))

    selected = {}
    for metal in METAL_ORDER:
        rows = [row for row in facts if metal in split_tags(row.get("metal_tags"))]
        for row in sorted(rows, key=score_fact, reverse=True)[: args.per_metal]:
            selected[row.get("fact_id")] = row
    for theme in THEME_PRIORITY:
        rows = [row for row in facts if theme in split_tags(row.get("theme_tags"))]
        for row in sorted(rows, key=score_fact, reverse=True)[: args.per_theme]:
            selected[row.get("fact_id")] = row

    selected_rows = sorted(selected.values(), key=score_fact, reverse=True)
    created_at = datetime.now(timezone(timedelta(hours=8))).isoformat(timespec="seconds")
    review_rows = []
    for row in selected_rows:
        review_status, matched_page, context = find_context(row, project_root)
        expression_type = source_expression_type(row.get("evidence_text", ""))
        scope = scope_gate(row.get("evidence_text", ""))
        review_rows.append(
            {
                "review_id": f"YS-FACT-REV-001-{len(review_rows) + 1:04d}",
                "case_id": row.get("case_id", "ANA-YS-INDUSTRY-001"),
                "batch_id": row.get("batch_id", ""),
                "sub_batch_id": row.get("sub_batch_id", ""),
                "run_id": "RUN-ANA-YS-FACT-REVIEW-001",
                "fact_id": row.get("fact_id", ""),
                "doc_id": row.get("doc_id", ""),
                "fact_type": row.get("fact_type", ""),
                "metal_tags": row.get("metal_tags", ""),
                "theme_tags": row.get("theme_tags", ""),
                "evidence_text": row.get("evidence_text", ""),
                "declared_page_no": row.get("page_no", ""),
                "matched_page_no": matched_page,
                "source_location": row.get("source_location", ""),
                "converted_text_path": row.get("converted_text_path", ""),
                "raw_file_path": row.get("raw_file_path", ""),
                "raw_sha256": row.get("raw_sha256", ""),
                "source_expression_type": expression_type,
                "scope_gate": scope,
                "analyst_review_status": review_status,
                "evidence_upgrade_status": upgrade_status(row, review_status, expression_type),
                "next_action": next_action(row, review_status, expression_type),
                "page_context": context[:900],
                "review_status": "DRAFT_FOR_REVIEW",
                "created_at": created_at,
            }
        )

    review_path = project_root / args.review_out
    manifest_path = project_root / args.manifest_out
    summary_path = project_root / args.summary_out
    fields = [
        "review_id",
        "case_id",
        "batch_id",
        "sub_batch_id",
        "run_id",
        "fact_id",
        "doc_id",
        "fact_type",
        "metal_tags",
        "theme_tags",
        "evidence_text",
        "declared_page_no",
        "matched_page_no",
        "source_location",
        "converted_text_path",
        "raw_file_path",
        "raw_sha256",
        "source_expression_type",
        "scope_gate",
        "analyst_review_status",
        "evidence_upgrade_status",
        "next_action",
        "page_context",
        "review_status",
        "created_at",
    ]
    write_csv(review_path, fields, review_rows)

    manifest_rows = [
        {
            "case_id": "ANA-YS-INDUSTRY-001",
            "batch_id": "BATCH-001+BATCH-003",
            "run_id": "RUN-ANA-YS-FACT-REVIEW-001",
            "artifact_type": "key_fact_review",
            "artifact_path": review_path.relative_to(project_root).as_posix(),
            "row_count": len(review_rows),
            "sha256": short_hash(review_path),
            "review_status": "DRAFT_FOR_REVIEW",
            "created_at": created_at,
        }
    ]
    write_csv(
        manifest_path,
        ["case_id", "batch_id", "run_id", "artifact_type", "artifact_path", "row_count", "sha256", "review_status", "created_at"],
        manifest_rows,
    )

    metal_counts = Counter()
    theme_counts = Counter()
    review_counts = Counter()
    upgrade_counts = Counter()
    action_counts = Counter()
    scope_counts = Counter()
    for row in review_rows:
        for metal in split_tags(row.get("metal_tags")):
            metal_counts[metal] += 1
        for theme in split_tags(row.get("theme_tags")):
            theme_counts[theme] += 1
        review_counts[row.get("analyst_review_status")] += 1
        upgrade_counts[row.get("evidence_upgrade_status")] += 1
        action_counts[row.get("next_action")] += 1
        scope_counts[row.get("scope_gate")] += 1

    lines = [
        "# 关键事实复核包 PASS-001 摘要",
        "",
        "状态：DRAFT_FOR_REVIEW",
        f"生成时间：{created_at}",
        "",
        "## 输入",
        "",
        "- BATCH-001 事实句草表",
        "- BATCH-003 事实句草表",
        "- converted text 页码标记和上下文",
        "",
        "## 输出",
        "",
        f"- 复核表：`{review_path.relative_to(project_root).as_posix()}`",
        f"- manifest：`{manifest_path.relative_to(project_root).as_posix()}`",
        f"- 复核记录数：{len(review_rows)}",
        "",
        "## 复核状态",
        "",
        "| 状态 | 数量 |",
        "|---|---:|",
    ]
    for key, count in review_counts.most_common():
        lines.append(f"| {key} | {count} |")
    lines.extend(["", "## 证据升级状态", "", "| 状态 | 数量 |", "|---|---:|"])
    for key, count in upgrade_counts.most_common():
        lines.append(f"| {key} | {count} |")
    lines.extend(["", "## 下一步动作", "", "| 动作 | 数量 |", "|---|---:|"])
    for key, count in action_counts.most_common():
        lines.append(f"| {key} | {count} |")
    lines.extend(["", "## 范围闸门", "", "| 范围 | 数量 |", "|---|---:|"])
    for key, count in scope_counts.most_common():
        lines.append(f"| {key} | {count} |")
    lines.extend(["", "## 金属覆盖", "", "| 金属 | 数量 |", "|---|---:|"])
    for key, count in metal_counts.most_common():
        lines.append(f"| {key} | {count} |")
    lines.extend(["", "## 主题覆盖", "", "| 主题 | 数量 |", "|---|---:|"])
    for key, count in theme_counts.most_common():
        lines.append(f"| {key} | {count} |")
    lines.extend(
        [
            "",
            "## 边界",
            "",
            "本轮只做 converted text 页码上下文核对和证据升级候选标注。`TEXT_MATCH_CONTEXT_OK` 不等于正式证据通过；关键指标仍需段落、表格定位和必要外部交叉验证。",
        ]
    )
    summary_path.parent.mkdir(parents=True, exist_ok=True)
    summary_path.write_text("\n".join(lines) + "\n", encoding="utf-8")

    print(f"review_rows={len(review_rows)}")
    print(f"review={review_path.relative_to(project_root).as_posix()}")
    print(f"manifest={manifest_path.relative_to(project_root).as_posix()}")
    print(f"summary={summary_path.relative_to(project_root).as_posix()}")


if __name__ == "__main__":
    main()