external_demand/app/hot_content/wxindex_heat_pattern.py

"""微信指数热度模式分析：持续高热、持续上涨、突然暴涨。"""

from __future__ import annotations

import csv
import json
from datetime import date, datetime, timedelta
from pathlib import Path
from typing import Any

from app.hot_content.client import JsonApiClient
from app.hot_content.demand_cache_service import DemandCacheService
from app.hot_content.demand_hive_export import WEIGHT_DIVISOR, build_demand_id
from app.hot_content.demand_quality import (
    TYPE_PHRASE,
    llm_score_senior_fit,
    lookup_quality_scores,
)
from app.hot_content.demand_pool_writer import sync_wxindex_word_rows_to_odps
from app.hot_content.exceptions import HotContentFlowError
from app.hot_content.postprocess_service import ContributionPostprocessService
from app.hot_content.repository import HotContentRepository
from app.hot_content.timezone import SHANGHAI_TZ
from app.hot_content.types import FlowConfig
from app.hot_content.wxindex_trend import (
    HEAT_RISING_ADJACENT_UP_RATIO,
    HEAT_RISING_OVERALL_CHANGE_RATE,
    HEAT_RISING_WINDOW_CHANGE_RATE,
    HEAT_SPIKE_BASELINE_FLOOR,
    HEAT_SPIKE_RATIO,
    extract_sorted_scores,
    is_wxindex_heat_rising_scores,
    is_wxindex_spike_scores,
)
from app.hot_content.wxindex_words import filter_scores_in_ymd_window

WXINDEX_HEAT_MIN_DAYS = 7
WXINDEX_HEAT_MAX_DAYS = 14
WXINDEX_SUSTAINED_HIGH_THRESHOLD = 10_000_000.0
WXINDEX_SPIKE_LOOKBACK_DAYS = 3
WXINDEX_SPIKE_RATIO = HEAT_SPIKE_RATIO
WXINDEX_SPIKE_BASELINE_FLOOR = HEAT_SPIKE_BASELINE_FLOOR
WXINDEX_RISING_OVERALL_CHANGE_RATE = HEAT_RISING_OVERALL_CHANGE_RATE
WXINDEX_RISING_WINDOW_CHANGE_RATE = HEAT_RISING_WINDOW_CHANGE_RATE
WXINDEX_RISING_ADJACENT_UP_RATIO = HEAT_RISING_ADJACENT_UP_RATIO
WXINDEX_WORD_SENIOR_FIT_NORMALIZED_THRESHOLD = 0.6
WXINDEX_WORD_LLM_BATCH_SIZE = 10

PATTERN_SUSTAINED_HIGH = "sustained_high"
PATTERN_RISING = "rising"
PATTERN_SPIKE = "spike"

RETAIN_REASON_SUSTAINED_HIGH = "持续高热度"
RETAIN_REASON_RISING = "热度持续上涨"
RETAIN_REASON_SPIKE = "热度突然暴涨"

PHASE_ANALYSIS_SKIPPED = "analysis_skipped"
PHASE_HEAT_ANALYZED = "heat_analyzed"
PHASE_DEMAND_MATCHED = "demand_matched"
PHASE_SENIOR_FIT_SCORED = "senior_fit_scored"
PHASE_FINALIZED = "finalized"

_HEAT_DONE_PHASES = frozenset(
    {
        PHASE_ANALYSIS_SKIPPED,
        PHASE_HEAT_ANALYZED,
        PHASE_DEMAND_MATCHED,
        PHASE_SENIOR_FIT_SCORED,
        PHASE_FINALIZED,
    }
)

_WXINDEX_RECORD_UNSET_ANALYSIS_FIELDS: dict[str, Any] = {
    "data_start_ymd": None,
    "data_end_ymd": None,
    "data_days": None,
    "is_sustained_high": None,
    "is_rising": None,
    "is_spike": None,
    "retain_reason": None,
    "is_internal_demand_matched": None,
    "matched_demand": None,
    "internal_demand_match_json": None,
    "senior_fit_score": None,
    "demand_senior_fit_json": None,
    "is_final_retained": None,
    "min_score": None,
    "max_score": None,
    "avg_score": None,
    "detail_json": None,
}


def _parse_record_detail_json(record: dict[str, Any]) -> dict[str, Any]:
    detail = record.get("detail_json")
    if isinstance(detail, dict):
        return detail
    if isinstance(detail, str) and detail.strip():
        try:
            parsed = json.loads(detail)
            return parsed if isinstance(parsed, dict) else {}
        except json.JSONDecodeError:
            return {}
    return {}


def _record_phase(record: dict[str, Any]) -> str:
    return str(_parse_record_detail_json(record).get("phase") or "").strip()


def _nullable_bool_from_record(value: Any) -> bool | None:
    if value is None:
        return None
    return bool(value)


def _parse_record_json_field(record: dict[str, Any], field: str) -> Any:
    value = record.get(field)
    if value is None:
        return None
    if isinstance(value, (dict, list)):
        return value
    if isinstance(value, (bytes, bytearray)):
        value = value.decode("utf-8")
    if isinstance(value, str):
        try:
            return json.loads(value)
        except json.JSONDecodeError:
            return None
    return value


def _is_heat_analysis_done(record: dict[str, Any] | None) -> bool:
    if not record:
        return False
    return _record_phase(record) in _HEAT_DONE_PHASES


def _is_analysis_skipped_record(record: dict[str, Any]) -> bool:
    return _record_phase(record) == PHASE_ANALYSIS_SKIPPED


def _is_senior_fit_needed(
    *,
    retain_reason: str | None,
    is_internal_demand_matched: bool | None,
) -> bool:
    if not retain_reason:
        return False
    if retain_reason == RETAIN_REASON_SUSTAINED_HIGH:
        return bool(is_internal_demand_matched)
    return True


def _is_finalized_record(record: dict[str, Any] | None) -> bool:
    if not record:
        return False
    return _record_phase(record) == PHASE_FINALIZED


def _latest_score_ymd(scores: list[dict[str, Any]]) -> str | None:
    ymds = [_score_row_ymd(row) for row in scores if _score_row_ymd(row)]
    return max(ymds) if ymds else None


def _should_rerun_heat_analysis(
    existing_record: dict[str, Any] | None,
    *,
    scores: list[dict[str, Any]],
    fetch_start_ymd: str,
    fetch_end_ymd: str,
    min_days: int,
    max_days: int,
) -> bool:
    if not _is_heat_analysis_done(existing_record):
        return True
    assert existing_record is not None
    latest_ymd = _latest_score_ymd(scores)
    record_end = str(existing_record.get("data_end_ymd") or "").strip()
    if latest_ymd and record_end and latest_ymd > record_end:
        return True
    if _is_analysis_skipped_record(existing_record):
        _, skip_reason = prepare_analysis_scores(
            scores,
            start_ymd=fetch_start_ymd,
            end_ymd=fetch_end_ymd,
            min_days=min_days,
            max_days=max_days,
        )
        if skip_reason is None:
            return True
    return False


def _is_senior_fit_attempt_done(
    item: dict[str, Any],
    existing_record: dict[str, Any] | None,
) -> bool:
    if existing_record:
        phase = _record_phase(existing_record)
        if phase in (PHASE_SENIOR_FIT_SCORED, PHASE_FINALIZED):
            return True
    return item.get("senior_fit_score") is not None


def _senior_fit_passed_from_score(score: Any, *, senior_threshold: float) -> bool:
    if score is None:
        return False
    try:
        return float(score) / 10.0 > senior_threshold
    except (TypeError, ValueError):
        return False


def _rehydrate_result_from_record(record: dict[str, Any]) -> dict[str, Any]:
    detail = _parse_record_detail_json(record)
    retain_reason_raw = record.get("retain_reason")
    retain_reason = (
        str(retain_reason_raw).strip() if retain_reason_raw is not None else None
    ) or None

    def _bool_field(key: str) -> bool | None:
        value = record.get(key)
        if value is None:
            return None
        return bool(value)

    return {
        "skipped": False,
        "skip_reason": None,
        "data_days": record.get("data_days"),
        "data_start_ymd": record.get("data_start_ymd"),
        "data_end_ymd": record.get("data_end_ymd"),
        "fetch_start_ymd": record.get("fetch_start_ymd"),
        "fetch_end_ymd": record.get("fetch_end_ymd"),
        "min_score": record.get("min_score"),
        "max_score": record.get("max_score"),
        "avg_score": record.get("avg_score"),
        "is_sustained_high": _bool_field("is_sustained_high"),
        "is_rising": _bool_field("is_rising"),
        "is_spike": _bool_field("is_spike"),
        "retain_reason": retain_reason,
        "patterns": list(detail.get("patterns") or []),
    }


def _rehydrate_pending_item_from_record(
    item: dict[str, Any],
    record: dict[str, Any],
    *,
    senior_threshold: float,
) -> dict[str, Any]:
    result = _rehydrate_result_from_record(record)
    retain_reason = result.get("retain_reason")
    is_internal_demand_matched = _nullable_bool_from_record(
        record.get("is_internal_demand_matched")
    )
    matched_demand_raw = record.get("matched_demand")
    matched_demand = (
        str(matched_demand_raw).strip() if matched_demand_raw is not None else None
    ) or None
    senior_fit_score = record.get("senior_fit_score")
    senior_fit_passed: bool | None = None
    is_final_retained = _nullable_bool_from_record(record.get("is_final_retained"))
    if senior_fit_score is not None:
        senior_fit_passed = _senior_fit_passed_from_score(
            senior_fit_score,
            senior_threshold=senior_threshold,
        )
    elif not _is_senior_fit_needed(
        retain_reason=retain_reason,
        is_internal_demand_matched=is_internal_demand_matched,
    ):
        senior_fit_passed = False
        if is_final_retained is None:
            is_final_retained = False

    return {
        "meta": item["meta"],
        "name": item["name"],
        "fetch_start_ymd": item["fetch_start_ymd"],
        "fetch_end_ymd": item["fetch_end_ymd"],
        "result": result,
        "retain_reason": retain_reason,
        "record_id": int(record.get("id") or item.get("record_id") or 0),
        "is_internal_demand_matched": is_internal_demand_matched,
        "matched_demand": matched_demand,
        "internal_demand_match_json": _parse_record_json_field(
            record,
            "internal_demand_match_json",
        ),
        "senior_fit_score": senior_fit_score,
        "demand_senior_fit_json": _parse_record_json_field(
            record,
            "demand_senior_fit_json",
        ),
        "senior_fit_passed": senior_fit_passed,
        "is_final_retained": is_final_retained,
    }


def _build_skipped_export_row_from_record(
    *,
    record: dict[str, Any],
    meta: dict[str, Any],
    name: str,
    analyze_ymd: str,
    fetch_start_ymd: str,
    fetch_end_ymd: str,
) -> dict[str, Any]:
    detail = _parse_record_detail_json(record)
    return {
        "analyze_ymd": analyze_ymd,
        "name": name,
        "meta_id": meta.get("id"),
        "fetch_start_ymd": fetch_start_ymd,
        "fetch_end_ymd": fetch_end_ymd,
        "analysis_skipped": True,
        "skip_reason": detail.get("skip_reason") or "",
        "data_days": record.get("data_days") or "",
        "retain_reason": "",
        "is_sustained_high": False,
        "is_rising": False,
        "is_spike": False,
        "is_internal_demand_matched": "",
        "matched_demand": "",
        "senior_fit_score": "",
        "is_final_retained": False,
        "min_score": "",
        "max_score": "",
        "avg_score": "",
    }


def _refresh_wxindex_heat_job_summary(
    summary: dict[str, Any],
    *,
    pending_items: list[dict[str, Any]],
    export_rows: list[dict[str, Any]],
) -> None:
    summary["analyzed"] = len(pending_items)
    summary["skipped"] = sum(1 for row in export_rows if row.get("analysis_skipped"))
    summary["retained"] = sum(1 for item in pending_items if item.get("retain_reason"))
    summary["sustained_high"] = sum(
        1
        for item in pending_items
        if item.get("retain_reason") == RETAIN_REASON_SUSTAINED_HIGH
    )
    summary["rising"] = sum(
        1 for item in pending_items if item.get("retain_reason") == RETAIN_REASON_RISING
    )
    summary["spike"] = sum(
        1 for item in pending_items if item.get("retain_reason") == RETAIN_REASON_SPIKE
    )
    summary["internal_demand_matched"] = sum(
        1 for item in pending_items if item.get("is_internal_demand_matched")
    )
    summary["senior_fit_candidates"] = sum(
        1
        for item in pending_items
        if _is_senior_fit_needed(
            retain_reason=item.get("retain_reason"),
            is_internal_demand_matched=item.get("is_internal_demand_matched"),
        )
    )
    summary["senior_scored"] = sum(
        1 for item in pending_items if item.get("senior_fit_score") is not None
    )
    summary["senior_fit_passed"] = sum(
        1 for item in pending_items if item.get("senior_fit_passed")
    )
    summary["final_retained"] = sum(
        1 for item in pending_items if item.get("is_final_retained")
    )


def resolve_retain_reason(
    *,
    is_sustained_high: bool,
    is_rising: bool,
    is_spike: bool,
) -> str | None:
    """按 2->3->1 优先级确定保留原因（上涨 > 暴涨 > 持续高热度）。"""
    if is_rising:
        return RETAIN_REASON_RISING
    if is_spike:
        return RETAIN_REASON_SPIKE
    if is_sustained_high:
        return RETAIN_REASON_SUSTAINED_HIGH
    return None


def _chunk_words(words: list[str], batch_size: int = WXINDEX_WORD_LLM_BATCH_SIZE) -> list[list[str]]:
    size = max(batch_size, 1)
    return [words[index : index + size] for index in range(0, len(words), size)]


def _empty_senior_fit_result() -> dict[str, Any]:
    return {
        "senior_fit_score": None,
        "demand_senior_fit_json": {"source": "", "items": []},
        "passed": False,
    }


def score_wxindex_words_senior_fit(
    *,
    words: list[str],
    config: FlowConfig,
    senior_threshold: float = WXINDEX_WORD_SENIOR_FIT_NORMALIZED_THRESHOLD,
    batch_size: int = WXINDEX_WORD_LLM_BATCH_SIZE,
) -> dict[str, dict[str, Any]]:
    """批量对微信指数热词执行老年性 LLM 评分（每批最多 batch_size 个词）。"""
    cleaned: list[str] = []
    seen: set[str] = set()
    for raw in words:
        word = str(raw or "").strip()
        if not word or word in seen:
            continue
        seen.add(word)
        cleaned.append(word)

    results: dict[str, dict[str, Any]] = {
        word: _empty_senior_fit_result() for word in cleaned
    }
    for batch in _chunk_words(cleaned, batch_size=batch_size):
        candidates = [{"demand_type": TYPE_PHRASE, "demand_text": word} for word in batch]
        senior_fit_json = llm_score_senior_fit(
            channel_content_id=f"wxindex_words:{','.join(batch[:3])}",
            candidates=candidates,
            model=config.demand_quality_llm_model,
            max_attempts=config.demand_quality_llm_max_attempts,
            retry_sleep_seconds=config.demand_quality_llm_retry_sleep_seconds,
            max_tokens=config.demand_quality_llm_max_tokens,
        )
        senior_fit_json["threshold"] = senior_threshold * 10.0
        for word in batch:
            _, senior_score = lookup_quality_scores(
                demand_type=TYPE_PHRASE,
                demand_text=word,
                event_sense_json=None,
                senior_fit_json=senior_fit_json,
            )
            passed = (
                senior_score is not None
                and senior_score / 10.0 > senior_threshold
            )
            results[word] = {
                "senior_fit_score": senior_score,
                "demand_senior_fit_json": senior_fit_json,
                "passed": passed,
            }
    return results


def score_wxindex_word_senior_fit(
    *,
    word: str,
    config: FlowConfig,
    senior_threshold: float = WXINDEX_WORD_SENIOR_FIT_NORMALIZED_THRESHOLD,
) -> dict[str, Any]:
    """对单个微信指数热词执行老年性 LLM 评分。"""
    target_word = str(word or "").strip()
    if not target_word:
        return _empty_senior_fit_result()

    batch_results = score_wxindex_words_senior_fit(
        words=[target_word],
        config=config,
        senior_threshold=senior_threshold,
        batch_size=1,
    )
    return batch_results.get(target_word, _empty_senior_fit_result())


def build_wxindex_word_hive_row(
    *,
    wxindex_word_record_id: int,
    word: str,
    strategy: str,
    partition_dt: str,
    max_score: float | None,
) -> dict[str, Any]:
    normalized_name = str(word or "").strip()
    weight = 0.0
    if max_score is not None:
        weight = float(max_score) / WEIGHT_DIVISOR
    return {
        "record_id": wxindex_word_record_id,
        "strategy": strategy,
        "demand_id": build_demand_id(
            strategy=strategy,
            demand_name=normalized_name,
            partition_dt=partition_dt,
        ),
        "demand_name": normalized_name,
        "weight": weight,
        "type": TYPE_PHRASE,
        "video_count": None,
        "video_list": [],
        "extend": "{}",
        "dt": partition_dt,
    }


def build_wxindex_word_odps_sync_row(
    *,
    wxindex_word_record_id: int,
    word: str,
    strategy: str,
    partition_dt: str,
    max_score: float | None,
) -> dict[str, Any]:
    normalized_name = str(word or "").strip()
    weight = None
    if max_score is not None:
        weight = float(max_score) / WEIGHT_DIVISOR
    return {
        "partition_dt": partition_dt,
        "strategy": strategy,
        "demand_id": build_demand_id(
            strategy=strategy,
            demand_name=normalized_name,
            partition_dt=partition_dt,
        ),
        "demand_name": normalized_name,
        "demand_type": TYPE_PHRASE,
        "record_id": wxindex_word_record_id,
        "weight": weight,
    }


def prepare_analysis_scores(
    scores: list[dict[str, Any]],
    *,
    start_ymd: str,
    end_ymd: str,
    min_days: int = WXINDEX_HEAT_MIN_DAYS,
    max_days: int = WXINDEX_HEAT_MAX_DAYS,
) -> tuple[list[dict[str, Any]], str | None]:
    """截取目标区间内可用数据，不足 min_days 时返回 skip 原因。"""
    window_scores = filter_scores_in_ymd_window(
        scores,
        start_ymd=start_ymd,
        end_ymd=end_ymd,
    )
    if len(window_scores) > max_days:
        window_scores = window_scores[-max_days:]
    if len(window_scores) < min_days:
        return window_scores, "insufficient_days"
    return window_scores, None


def _score_row_ymd(row: dict[str, Any]) -> str:
    return str(row.get("ymd") or row.get("dt") or "").strip()


def _window_ymd_bounds(
    window_scores: list[dict[str, Any]],
) -> tuple[str | None, str | None]:
    if not window_scores:
        return None, None
    start_ymd = _score_row_ymd(window_scores[0]) or None
    end_ymd = _score_row_ymd(window_scores[-1]) or None
    return start_ymd, end_ymd


def analyze_wxindex_heat_patterns(
    scores: list[dict[str, Any]],
    *,
    start_ymd: str,
    end_ymd: str,
    sustained_threshold: float = WXINDEX_SUSTAINED_HIGH_THRESHOLD,
    min_days: int = WXINDEX_HEAT_MIN_DAYS,
    max_days: int = WXINDEX_HEAT_MAX_DAYS,
    spike_days: int = WXINDEX_SPIKE_LOOKBACK_DAYS,
    spike_ratio: float = WXINDEX_SPIKE_RATIO,
    spike_baseline_floor: float = WXINDEX_SPIKE_BASELINE_FLOOR,
    rising_overall_change_rate: float = WXINDEX_RISING_OVERALL_CHANGE_RATE,
    rising_window_change_rate: float = WXINDEX_RISING_WINDOW_CHANGE_RATE,
    rising_adjacent_up_ratio: float = WXINDEX_RISING_ADJACENT_UP_RATIO,
) -> dict[str, Any]:
    """对目标区间数据判断三种热度模式。"""
    window_scores, skip_reason = prepare_analysis_scores(
        scores,
        start_ymd=start_ymd,
        end_ymd=end_ymd,
        min_days=min_days,
        max_days=max_days,
    )
    if skip_reason:
        data_start_ymd, data_end_ymd = _window_ymd_bounds(window_scores)
        return {
            "skipped": True,
            "skip_reason": skip_reason,
            "data_days": len(window_scores) if window_scores else None,
            "data_start_ymd": data_start_ymd,
            "data_end_ymd": data_end_ymd,
            "patterns": [],
        }

    numeric_scores = extract_sorted_scores(window_scores, max_points=max_days)
    data_start_ymd, data_end_ymd = _window_ymd_bounds(window_scores)

    is_sustained_high = all(score > sustained_threshold for score in numeric_scores)
    is_rising = is_wxindex_heat_rising_scores(
        numeric_scores,
        min_points=min_days,
        overall_change_rate=rising_overall_change_rate,
        window_change_rate_threshold=rising_window_change_rate,
        adjacent_up_ratio=rising_adjacent_up_ratio,
    )
    is_spike = is_wxindex_spike_scores(
        numeric_scores,
        spike_days=spike_days,
        min_points=min_days,
        spike_ratio=spike_ratio,
        baseline_floor=spike_baseline_floor,
    )

    retain_reason = resolve_retain_reason(
        is_sustained_high=is_sustained_high,
        is_rising=is_rising,
        is_spike=is_spike,
    )

    patterns: list[str] = []
    if retain_reason == RETAIN_REASON_SUSTAINED_HIGH:
        patterns.append(PATTERN_SUSTAINED_HIGH)
    elif retain_reason == RETAIN_REASON_RISING:
        patterns.append(PATTERN_RISING)
    elif retain_reason == RETAIN_REASON_SPIKE:
        patterns.append(PATTERN_SPIKE)

    return {
        "skipped": False,
        "skip_reason": None,
        "data_days": len(window_scores),
        "data_start_ymd": data_start_ymd,
        "data_end_ymd": data_end_ymd,
        "fetch_start_ymd": start_ymd,
        "fetch_end_ymd": end_ymd,
        "min_score": min(numeric_scores),
        "max_score": max(numeric_scores),
        "avg_score": sum(numeric_scores) / len(numeric_scores),
        "is_sustained_high": is_sustained_high,
        "is_rising": is_rising,
        "is_spike": is_spike,
        "retain_reason": retain_reason,
        "patterns": patterns,
        "scores": window_scores,
    }


def build_wxindex_word_record_init_payload(
    *,
    meta: dict[str, Any],
    name: str,
    analyze_ymd: str,
    fetch_start_ymd: str,
    fetch_end_ymd: str,
    demand_cache_run_id: int | None = None,
) -> dict[str, Any]:
    """分析前写入追溯记录：仅含 meta 与抓取窗口，分析字段待后续更新。"""
    return {
        "name": name,
        "meta_id": meta.get("id"),
        "analyze_ymd": analyze_ymd,
        "fetch_start_ymd": fetch_start_ymd,
        "fetch_end_ymd": fetch_end_ymd,
        "demand_cache_run_id": demand_cache_run_id,
        **_WXINDEX_RECORD_UNSET_ANALYSIS_FIELDS,
    }


def build_wxindex_word_record_skipped_payload(
    *,
    meta: dict[str, Any],
    name: str,
    analyze_ymd: str,
    fetch_start_ymd: str,
    fetch_end_ymd: str,
    result: dict[str, Any],
    demand_cache_run_id: int | None = None,
) -> dict[str, Any]:
    """分析被跳过时更新追溯记录。"""
    payload = {
        "name": name,
        "meta_id": meta.get("id"),
        "analyze_ymd": analyze_ymd,
        "fetch_start_ymd": fetch_start_ymd,
        "fetch_end_ymd": fetch_end_ymd,
        "demand_cache_run_id": demand_cache_run_id,
        **_WXINDEX_RECORD_UNSET_ANALYSIS_FIELDS,
    }
    payload.update(
        {
            "data_start_ymd": result.get("data_start_ymd"),
            "data_end_ymd": result.get("data_end_ymd"),
            "data_days": result.get("data_days"),
            "detail_json": {
                "phase": "analysis_skipped",
                "skip_reason": result.get("skip_reason"),
                "data_days": result.get("data_days"),
            },
        }
    )
    return payload


def build_wxindex_word_record_payload(
    item: dict[str, Any],
    *,
    analyze_ymd: str,
    demand_cache_run_id: int | None,
    phase: str,
    senior_threshold: float,
    sustained_threshold: float,
    spike_days: int,
    spike_ratio: float,
    spike_baseline_floor: float,
    rising_overall_change_rate: float,
    rising_window_change_rate: float,
    rising_adjacent_up_ratio: float,
) -> dict[str, Any]:
    """根据 pending item 当前状态构建 records 表 payload。"""
    meta = item["meta"]
    name = item["name"]
    fetch_start_ymd = item["fetch_start_ymd"]
    fetch_end_ymd = item["fetch_end_ymd"]
    result = item["result"]
    retain_reason = item.get("retain_reason")

    return {
        "name": name,
        "meta_id": meta.get("id"),
        "analyze_ymd": analyze_ymd,
        "fetch_start_ymd": fetch_start_ymd,
        "fetch_end_ymd": fetch_end_ymd,
        "data_start_ymd": result.get("data_start_ymd"),
        "data_end_ymd": result.get("data_end_ymd"),
        "data_days": result.get("data_days"),
        "is_sustained_high": result.get("is_sustained_high"),
        "is_rising": result.get("is_rising"),
        "is_spike": result.get("is_spike"),
        "retain_reason": retain_reason,
        "is_internal_demand_matched": item.get("is_internal_demand_matched"),
        "matched_demand": item.get("matched_demand"),
        "demand_cache_run_id": demand_cache_run_id,
        "internal_demand_match_json": item.get("internal_demand_match_json"),
        "senior_fit_score": item.get("senior_fit_score"),
        "demand_senior_fit_json": item.get("demand_senior_fit_json"),
        "is_final_retained": item.get("is_final_retained"),
        "min_score": result.get("min_score"),
        "max_score": result.get("max_score"),
        "avg_score": result.get("avg_score"),
        "detail_json": {
            "phase": phase,
            "patterns": list(result.get("patterns") or []),
            "retain_reason": retain_reason,
            "retain_reason_priority": "2->3->1",
            "senior_fit_threshold": senior_threshold,
            "senior_fit_threshold_score": senior_threshold * 10.0,
            "is_final_retained": item.get("is_final_retained"),
            "sustained_threshold": sustained_threshold,
            "spike_days": spike_days,
            "spike_ratio": spike_ratio,
            "spike_baseline_floor": spike_baseline_floor,
            "rising_overall_change_rate": rising_overall_change_rate,
            "rising_window_change_rate": rising_window_change_rate,
            "rising_adjacent_up_ratio": rising_adjacent_up_ratio,
        },
    }


def _persist_wxindex_word_record(
    repository: HotContentRepository,
    payload: dict[str, Any],
    *,
    dry_run: bool,
    skip_db_save: bool,
    verbose: bool,
    action: str,
) -> int:
    name = str(payload.get("name") or "").strip()
    if dry_run or skip_db_save:
        if verbose:
            label = "dry-run" if dry_run else "skip-db-save"
            print(f"[{label}] would {action} wxindex word record word={name}")
        return 0
    return repository.save_wxindex_word_record(payload)


def _persist_pending_item_record(
    repository: HotContentRepository,
    item: dict[str, Any],
    *,
    analyze_ymd: str,
    demand_cache_run_id: int | None,
    phase: str,
    senior_threshold: float,
    sustained_threshold: float,
    spike_days: int,
    spike_ratio: float,
    spike_baseline_floor: float,
    rising_overall_change_rate: float,
    rising_window_change_rate: float,
    rising_adjacent_up_ratio: float,
    dry_run: bool,
    skip_db_save: bool,
    verbose: bool,
) -> int:
    payload = build_wxindex_word_record_payload(
        item,
        analyze_ymd=analyze_ymd,
        demand_cache_run_id=demand_cache_run_id,
        phase=phase,
        senior_threshold=senior_threshold,
        sustained_threshold=sustained_threshold,
        spike_days=spike_days,
        spike_ratio=spike_ratio,
        spike_baseline_floor=spike_baseline_floor,
        rising_overall_change_rate=rising_overall_change_rate,
        rising_window_change_rate=rising_window_change_rate,
        rising_adjacent_up_ratio=rising_adjacent_up_ratio,
    )
    record_id = _persist_wxindex_word_record(
        repository,
        payload,
        dry_run=dry_run,
        skip_db_save=skip_db_save,
        verbose=verbose,
        action=f"{phase} word={item['name']}",
    )
    if record_id:
        item["record_id"] = record_id
    return int(item.get("record_id") or 0)


def _filter_candidates_awaiting_yesterday_score(
    repository: HotContentRepository,
    candidate_items: list[dict[str, Any]],
    *,
    yesterday_ymd: str,
    existing_records: dict[str, dict[str, Any]],
    verbose: bool,
) -> tuple[list[dict[str, Any]], int]:
    """初始化完成后：未完成热度分析且缺少昨日指数数据的词留待下次执行。"""
    names_to_check = [
        item["name"]
        for item in candidate_items
        if not _is_heat_analysis_done(existing_records.get(item["name"]))
    ]
    if not names_to_check:
        return candidate_items, 0

    names_with_yesterday = repository.list_wxindex_word_names_with_dt(
        names_to_check,
        dt=yesterday_ymd,
    )
    ready_items: list[dict[str, Any]] = []
    awaiting_count = 0
    for item in candidate_items:
        name = item["name"]
        if _is_heat_analysis_done(existing_records.get(name)):
            ready_items.append(item)
            continue
        if name in names_with_yesterday:
            ready_items.append(item)
            continue
        awaiting_count += 1
        if verbose:
            print(
                f"await yesterday score word={name} dt={yesterday_ymd}, skip this run"
            )
    return ready_items, awaiting_count


def _init_candidate_wxindex_word_records(
    repository: HotContentRepository,
    candidate_items: list[dict[str, Any]],
    *,
    analyze_ymd: str,
    demand_cache_run_id: int | None,
    dry_run: bool,
    skip_db_save: bool,
    verbose: bool,
) -> None:
    if not candidate_items:
        return
    if dry_run or skip_db_save:
        if verbose:
            label = "dry-run" if dry_run else "skip-db-save"
            print(
                f"[{label}] would batch init wxindex word records "
                f"count={len(candidate_items)} analyze_ymd={analyze_ymd}"
            )
        return

    init_payloads = [
        build_wxindex_word_record_init_payload(
            meta=item["meta"],
            name=item["name"],
            analyze_ymd=analyze_ymd,
            fetch_start_ymd=item["fetch_start_ymd"],
            fetch_end_ymd=item["fetch_end_ymd"],
            demand_cache_run_id=demand_cache_run_id,
        )
        for item in candidate_items
    ]
    record_id_map = repository.init_wxindex_word_records(init_payloads)
    for item in candidate_items:
        item["record_id"] = int(record_id_map.get(item["name"]) or 0)
    if verbose:
        print(
            f"init wxindex word records count={len(candidate_items)} "
            f"analyze_ymd={analyze_ymd}"
        )


def _apply_demand_match_to_item(
    item: dict[str, Any],
    match_result: dict[str, Any],
) -> None:
    item["internal_demand_match_json"] = match_result
    item["is_internal_demand_matched"] = bool(match_result.get("matched"))
    matched_demand = str(match_result.get("matched_demand") or "").strip()
    item["matched_demand"] = matched_demand or None
    if not item.get("is_internal_demand_matched"):
        item["is_final_retained"] = False


def _apply_senior_fit_to_item(
    item: dict[str, Any],
    senior_result: dict[str, Any],
) -> None:
    passed = bool(senior_result.get("passed"))
    item["senior_fit_score"] = senior_result.get("senior_fit_score")
    item["demand_senior_fit_json"] = senior_result.get("demand_senior_fit_json")
    item["senior_fit_passed"] = passed
    item["is_final_retained"] = passed


def build_wxindex_word_stats_payload(
    item: dict[str, Any],
    *,
    analyze_ymd: str,
) -> dict[str, Any]:
    """构建通过热度+老年性筛选的词统计 payload。"""
    meta = item["meta"]
    result = item["result"]
    return {
        "name": item["name"],
        "meta_id": meta.get("id"),
        "analyze_ymd": analyze_ymd,
        "wxindex_word_record_id": item.get("record_id"),
        "retain_reason": item.get("retain_reason"),
        "senior_fit_score": item.get("senior_fit_score"),
        "data_start_ymd": result.get("data_start_ymd"),
        "data_end_ymd": result.get("data_end_ymd"),
        "data_days": result.get("data_days"),
        "min_score": result.get("min_score"),
        "max_score": result.get("max_score"),
        "avg_score": result.get("avg_score"),
        "detail_json": {
            "demand_senior_fit_json": item.get("demand_senior_fit_json"),
            "is_sustained_high": result.get("is_sustained_high"),
            "is_rising": result.get("is_rising"),
            "is_spike": result.get("is_spike"),
        },
    }


def _save_senior_fit_passed_stats(
    repository: HotContentRepository,
    items: list[dict[str, Any]],
    *,
    analyze_ymd: str,
    existing_stats_names: set[str] | None = None,
    dry_run: bool,
    skip_db_save: bool,
    verbose: bool,
) -> tuple[int, int]:
    existing = existing_stats_names or set()
    all_payloads = [
        build_wxindex_word_stats_payload(item, analyze_ymd=analyze_ymd)
        for item in items
        if item.get("senior_fit_passed")
    ]
    resumed = sum(1 for payload in all_payloads if payload["name"] in existing)
    payloads = [payload for payload in all_payloads if payload["name"] not in existing]
    if not payloads:
        return 0, resumed
    if dry_run or skip_db_save:
        if verbose:
            label = "dry-run" if dry_run else "skip-db-save"
            print(f"[{label}] would save wxindex word stats count={len(payloads)}")
        return len(payloads), resumed
    saved = repository.save_wxindex_word_stats_batch(payloads)
    if verbose:
        print(f"saved wxindex word stats count={saved} analyze_ymd={analyze_ymd}")
    return saved, resumed


def _build_export_row_from_item(
    item: dict[str, Any],
    *,
    analyze_ymd: str,
    strategy: str,
) -> dict[str, Any]:
    meta = item["meta"]
    name = item["name"]
    result = item["result"]
    retain_reason = item.get("retain_reason")
    is_internal_demand_matched = item.get("is_internal_demand_matched")
    matched_demand = str(item.get("matched_demand") or "")
    senior_fit_score = item.get("senior_fit_score")
    is_final_retained = bool(item.get("is_final_retained"))

    return {
        "analyze_ymd": analyze_ymd,
        "name": name,
        "meta_id": meta.get("id"),
        "fetch_start_ymd": item["fetch_start_ymd"],
        "fetch_end_ymd": item["fetch_end_ymd"],
        "data_start_ymd": result.get("data_start_ymd"),
        "data_end_ymd": result.get("data_end_ymd"),
        "data_days": result.get("data_days"),
        "analysis_skipped": False,
        "skip_reason": "",
        "is_sustained_high": bool(result.get("is_sustained_high")),
        "is_rising": bool(result.get("is_rising")),
        "is_spike": bool(result.get("is_spike")),
        "retain_reason": retain_reason or "",
        "is_internal_demand_matched": (
            "" if is_internal_demand_matched is None else is_internal_demand_matched
        ),
        "matched_demand": matched_demand,
        "senior_fit_score": senior_fit_score if senior_fit_score is not None else "",
        "is_final_retained": is_final_retained,
        "min_score": result.get("min_score"),
        "max_score": result.get("max_score"),
        "avg_score": result.get("avg_score"),
        "demand_id": (
            build_demand_id(
                strategy=strategy,
                demand_name=name,
                partition_dt=analyze_ymd,
            )
            if is_final_retained and strategy
            else ""
        ),
        "weight": (
            float(result.get("max_score") or 0) / WEIGHT_DIVISOR
            if is_final_retained and result.get("max_score") is not None
            else ""
        ),
    }


def run_wxindex_heat_pattern_daily_job(
    repository: HotContentRepository,
    *,
    config: FlowConfig | None = None,
    api_client: JsonApiClient | None = None,
    today: date | None = None,
    sustained_threshold: float = WXINDEX_SUSTAINED_HIGH_THRESHOLD,
    min_days: int = WXINDEX_HEAT_MIN_DAYS,
    max_days: int = WXINDEX_HEAT_MAX_DAYS,
    spike_days: int = WXINDEX_SPIKE_LOOKBACK_DAYS,
    spike_ratio: float = WXINDEX_SPIKE_RATIO,
    spike_baseline_floor: float = WXINDEX_SPIKE_BASELINE_FLOOR,
    rising_overall_change_rate: float = WXINDEX_RISING_OVERALL_CHANGE_RATE,
    rising_window_change_rate: float = WXINDEX_RISING_WINDOW_CHANGE_RATE,
    rising_adjacent_up_ratio: float = WXINDEX_RISING_ADJACENT_UP_RATIO,
    dry_run: bool = False,
    skip_odps: bool = False,
    skip_db_save: bool = False,
    verbose: bool = False,
) -> dict[str, Any]:
    """定时任务：分析仍在抓取窗口内的词，并写入热度模式结果表。"""
    current = today or datetime.now(SHANGHAI_TZ).date()
    analyze_ymd = current.strftime("%Y%m%d")
    meta_rows = repository.list_active_wxindex_word_meta(today=current)

    demand_name_set: list[str] = []
    demand_cache_run_id: int | None = None
    demand_cache_error: str | None = None
    postprocess_service: ContributionPostprocessService | None = None
    if config is not None:
        try:
            cache = DemandCacheService(config, repository).get_or_create_current_hour_cache()
            demand_name_set = list(cache.get("demand_name_set") or [])
            demand_cache_run_id = int(cache["id"])
            if demand_name_set:
                postprocess_service = ContributionPostprocessService(
                    config,
                    repository,
                    api_client
                    or JsonApiClient(
                        timeout_seconds=config.request_timeout_seconds,
                        verify_ssl=config.https_verify_ssl,
                    ),
                )
        except HotContentFlowError as exc:
            demand_cache_error = str(exc)
            if verbose:
                print(f"demand cache unavailable, skip demand match: {exc}")

    summary: dict[str, Any] = {
        "analyze_ymd": analyze_ymd,
        "meta_count": len(meta_rows),
        "analyzed": 0,
        "skipped": 0,
        "retained": 0,
        "sustained_high": 0,
        "rising": 0,
        "spike": 0,
        "internal_demand_matched": 0,
        "senior_scored": 0,
        "senior_fit_candidates": 0,
        "senior_fit_passed": 0,
        "stats_saved": 0,
        "final_retained": 0,
        "odps_synced": 0,
        "odps_written": 0,
        "demand_cache_run_id": demand_cache_run_id,
        "demand_cache_error": demand_cache_error,
        "demand_name_count": len(demand_name_set),
        "llm_batch_size": WXINDEX_WORD_LLM_BATCH_SIZE,
        "demand_match_batches": 0,
        "senior_fit_batches": 0,
        "records_initialized": 0,
        "awaiting_yesterday_score": 0,
        "heat_resumed": 0,
        "demand_match_resumed": 0,
        "senior_fit_resumed": 0,
        "stats_resumed": 0,
        "finalized_resumed": 0,
        "dry_run": dry_run,
        "skip_odps": skip_odps,
        "skip_db_save": skip_db_save,
    }
    retained_words: list[dict[str, Any]] = []
    export_rows: list[dict[str, Any]] = []
    final_hive_rows: list[dict[str, Any]] = []
    senior_threshold = WXINDEX_WORD_SENIOR_FIT_NORMALIZED_THRESHOLD
    strategy = str(config.hot_demand_pool_strategy or "").strip() if config else ""
    pending_items: list[dict[str, Any]] = []
    candidate_items: list[dict[str, Any]] = []

    for meta in meta_rows:
        name = str(meta.get("name") or "").strip()
        fetch_start_ymd = str(meta.get("fetch_start_ymd") or "").strip()
        fetch_end_ymd = str(meta.get("fetch_end_ymd") or "").strip()
        if not name or not fetch_start_ymd or not fetch_end_ymd:
            summary["skipped"] += 1
            continue

        candidate_items.append(
            {
                "meta": meta,
                "name": name,
                "fetch_start_ymd": fetch_start_ymd,
                "fetch_end_ymd": fetch_end_ymd,
            }
        )

    _init_candidate_wxindex_word_records(
        repository,
        candidate_items,
        analyze_ymd=analyze_ymd,
        demand_cache_run_id=demand_cache_run_id,
        dry_run=dry_run,
        skip_db_save=skip_db_save,
        verbose=verbose,
    )
    summary["records_initialized"] = len(candidate_items)

    existing_records: dict[str, dict[str, Any]] = {}
    existing_stats_names: set[str] = set()
    if not dry_run and not skip_db_save:
        candidate_names = [item["name"] for item in candidate_items]
        existing_records = repository.list_wxindex_word_records_by_analyze_ymd(
            analyze_ymd=analyze_ymd,
            names=candidate_names,
        )
        existing_stats_names = repository.list_wxindex_word_stats_names(
            analyze_ymd=analyze_ymd,
            names=candidate_names,
        )

    yesterday_ymd = (current - timedelta(days=1)).strftime("%Y%m%d")
    candidate_items, awaiting_yesterday = _filter_candidates_awaiting_yesterday_score(
        repository,
        candidate_items,
        yesterday_ymd=yesterday_ymd,
        existing_records=existing_records,
        verbose=verbose,
    )
    summary["awaiting_yesterday_score"] = awaiting_yesterday

    for item in candidate_items:
        if not item.get("record_id"):
            existing = existing_records.get(item["name"])
            if existing:
                item["record_id"] = int(existing.get("id") or 0)

    for item in candidate_items:
        meta = item["meta"]
        name = item["name"]
        fetch_start_ymd = item["fetch_start_ymd"]
        fetch_end_ymd = item["fetch_end_ymd"]
        existing_record = existing_records.get(name)
        scores = repository.list_wxindex_word_scores_in_range(
            name,
            start_ymd=fetch_start_ymd,
            end_ymd=fetch_end_ymd,
        )

        if (
            _is_heat_analysis_done(existing_record)
            and not _should_rerun_heat_analysis(
                existing_record,
                scores=scores,
                fetch_start_ymd=fetch_start_ymd,
                fetch_end_ymd=fetch_end_ymd,
                min_days=min_days,
                max_days=max_days,
            )
        ):
            summary["heat_resumed"] += 1
            if _is_analysis_skipped_record(existing_record):
                if verbose:
                    detail = _parse_record_detail_json(existing_record)
                    print(
                        f"resume skip word={name} reason={detail.get('skip_reason')} "
                        f"days={existing_record.get('data_days')}"
                    )
                export_rows.append(
                    _build_skipped_export_row_from_record(
                        record=existing_record,
                        meta=meta,
                        name=name,
                        analyze_ymd=analyze_ymd,
                        fetch_start_ymd=fetch_start_ymd,
                        fetch_end_ymd=fetch_end_ymd,
                    )
                )
                continue

            pending_item = _rehydrate_pending_item_from_record(
                item,
                existing_record,
                senior_threshold=senior_threshold,
            )
            pending_items.append(pending_item)
            if verbose:
                print(
                    f"resume heat analyzed word={name} "
                    f"retain_reason={pending_item.get('retain_reason') or ''} "
                    f"data_days={pending_item['result'].get('data_days')}"
                )
            continue

        result = analyze_wxindex_heat_patterns(
            scores,
            start_ymd=fetch_start_ymd,
            end_ymd=fetch_end_ymd,
            sustained_threshold=sustained_threshold,
            min_days=min_days,
            max_days=max_days,
            spike_days=spike_days,
            spike_ratio=spike_ratio,
            spike_baseline_floor=spike_baseline_floor,
            rising_overall_change_rate=rising_overall_change_rate,
            rising_window_change_rate=rising_window_change_rate,
            rising_adjacent_up_ratio=rising_adjacent_up_ratio,
        )
        if result.get("skipped"):
            if verbose:
                print(
                    f"skip word={name} reason={result.get('skip_reason')} "
                    f"days={result.get('data_days')}"
                )
            skipped_payload = build_wxindex_word_record_skipped_payload(
                meta=meta,
                name=name,
                analyze_ymd=analyze_ymd,
                fetch_start_ymd=fetch_start_ymd,
                fetch_end_ymd=fetch_end_ymd,
                result=result,
                demand_cache_run_id=demand_cache_run_id,
            )
            _persist_wxindex_word_record(
                repository,
                skipped_payload,
                dry_run=dry_run,
                skip_db_save=skip_db_save,
                verbose=verbose,
                action="update skipped",
            )
            export_rows.append(
                {
                    "analyze_ymd": analyze_ymd,
                    "name": name,
                    "meta_id": meta.get("id"),
                    "fetch_start_ymd": fetch_start_ymd,
                    "fetch_end_ymd": fetch_end_ymd,
                    "analysis_skipped": True,
                    "skip_reason": result.get("skip_reason"),
                    "data_days": result.get("data_days"),
                    "retain_reason": "",
                    "is_sustained_high": False,
                    "is_rising": False,
                    "is_spike": False,
                    "is_internal_demand_matched": "",
                    "matched_demand": "",
                    "senior_fit_score": "",
                    "is_final_retained": False,
                    "min_score": "",
                    "max_score": "",
                    "avg_score": "",
                }
            )
            continue

        retain_reason = str(result.get("retain_reason") or "").strip() or None

        pending_item = {
            "meta": meta,
            "name": name,
            "fetch_start_ymd": fetch_start_ymd,
            "fetch_end_ymd": fetch_end_ymd,
            "result": result,
            "retain_reason": retain_reason,
            "record_id": item.get("record_id", 0),
            "is_internal_demand_matched": None,
            "matched_demand": None,
            "internal_demand_match_json": None,
            "senior_fit_score": None,
            "demand_senior_fit_json": None,
            "senior_fit_passed": None,
            "is_final_retained": None,
        }
        pending_items.append(pending_item)
        _persist_pending_item_record(
            repository,
            pending_item,
            analyze_ymd=analyze_ymd,
            demand_cache_run_id=demand_cache_run_id,
            phase="heat_analyzed",
            senior_threshold=senior_threshold,
            sustained_threshold=sustained_threshold,
            spike_days=spike_days,
            spike_ratio=spike_ratio,
            spike_baseline_floor=spike_baseline_floor,
            rising_overall_change_rate=rising_overall_change_rate,
            rising_window_change_rate=rising_window_change_rate,
            rising_adjacent_up_ratio=rising_adjacent_up_ratio,
            dry_run=dry_run,
            skip_db_save=skip_db_save,
            verbose=verbose,
        )
        if verbose:
            print(
                f"heat analyzed word={name} retain_reason={retain_reason or ''} "
                f"data_days={result.get('data_days')}"
            )

    demand_match_results: dict[str, dict[str, Any]] = {}
    pending_by_name = {item["name"]: item for item in pending_items}
    sustained_high_words = [
        item["name"]
        for item in pending_items
        if item.get("retain_reason") == RETAIN_REASON_SUSTAINED_HIGH
        and item.get("is_internal_demand_matched") is None
    ]
    for item in pending_items:
        if item.get("retain_reason") != RETAIN_REASON_SUSTAINED_HIGH:
            continue
        if item.get("is_internal_demand_matched") is not None:
            summary["demand_match_resumed"] += 1

    def _persist_demand_match_updates(words: list[str]) -> None:
        for word in words:
            pending_item = pending_by_name.get(word)
            if pending_item is None:
                continue
            match_result = demand_match_results.get(word) or {
                "word": word,
                "matched": False,
                "matched_demand": "",
                "match_list": [],
            }
            _apply_demand_match_to_item(pending_item, match_result)
            _persist_pending_item_record(
                repository,
                pending_item,
                analyze_ymd=analyze_ymd,
                demand_cache_run_id=demand_cache_run_id,
                phase="demand_matched",
                senior_threshold=senior_threshold,
                sustained_threshold=sustained_threshold,
                spike_days=spike_days,
                spike_ratio=spike_ratio,
                spike_baseline_floor=spike_baseline_floor,
                rising_overall_change_rate=rising_overall_change_rate,
                rising_window_change_rate=rising_window_change_rate,
                rising_adjacent_up_ratio=rising_adjacent_up_ratio,
                dry_run=dry_run,
                skip_db_save=skip_db_save,
                verbose=verbose,
            )

    if sustained_high_words:
        if postprocess_service is not None and demand_name_set:
            for batch in _chunk_words(sustained_high_words):
                summary["demand_match_batches"] += 1
                if verbose:
                    print(f"demand match batch size={len(batch)} words={batch}")
                demand_match_results.update(
                    postprocess_service.match_words_to_demand_pool(
                        words=batch,
                        demand_name_set=demand_name_set,
                    )
                )
                _persist_demand_match_updates(batch)
        else:
            for word in sustained_high_words:
                demand_match_results[word] = {
                    "word": word,
                    "matched": False,
                    "matched_demand": "",
                    "match_list": [],
                    "skip_reason": "empty_demand_cache",
                }
            _persist_demand_match_updates(sustained_high_words)

    senior_fit_candidate_names: list[str] = []
    for item in pending_items:
        retain_reason = item.get("retain_reason")
        name = str(item.get("name") or "").strip()
        if not retain_reason or not name:
            continue
        if _is_senior_fit_attempt_done(item, existing_records.get(name)):
            summary["senior_fit_resumed"] += 1
            continue
        if retain_reason == RETAIN_REASON_SUSTAINED_HIGH:
            if not item.get("is_internal_demand_matched"):
                continue
        senior_fit_candidate_names.append(name)

    senior_fit_results: dict[str, dict[str, Any]] = {}

    def _persist_senior_fit_updates(words: list[str]) -> None:
        for word in words:
            pending_item = pending_by_name.get(word)
            senior_result = senior_fit_results.get(word)
            if pending_item is None or senior_result is None:
                continue
            _apply_senior_fit_to_item(pending_item, senior_result)
            _persist_pending_item_record(
                repository,
                pending_item,
                analyze_ymd=analyze_ymd,
                demand_cache_run_id=demand_cache_run_id,
                phase="senior_fit_scored",
                senior_threshold=senior_threshold,
                sustained_threshold=sustained_threshold,
                spike_days=spike_days,
                spike_ratio=spike_ratio,
                spike_baseline_floor=spike_baseline_floor,
                rising_overall_change_rate=rising_overall_change_rate,
                rising_window_change_rate=rising_window_change_rate,
                rising_adjacent_up_ratio=rising_adjacent_up_ratio,
                dry_run=dry_run,
                skip_db_save=skip_db_save,
                verbose=verbose,
            )
            if verbose and not pending_item.get("senior_fit_passed"):
                print(
                    f"senior fit rejected word={word} "
                    f"score={pending_item.get('senior_fit_score')}"
                )

    if config is not None and senior_fit_candidate_names:
        for batch in _chunk_words(senior_fit_candidate_names):
            summary["senior_fit_batches"] += 1
            if verbose:
                print(f"senior fit batch size={len(batch)} words={batch}")
            senior_fit_results.update(
                score_wxindex_words_senior_fit(
                    words=batch,
                    config=config,
                    senior_threshold=senior_threshold,
                    batch_size=WXINDEX_WORD_LLM_BATCH_SIZE,
                )
            )
            _persist_senior_fit_updates(batch)

    stats_saved, stats_resumed = _save_senior_fit_passed_stats(
        repository,
        pending_items,
        analyze_ymd=analyze_ymd,
        existing_stats_names=existing_stats_names,
        dry_run=dry_run,
        skip_db_save=skip_db_save,
        verbose=verbose,
    )
    summary["stats_saved"] = stats_saved
    summary["stats_resumed"] = stats_resumed

    for item in pending_items:
        name = item["name"]
        result = item["result"]
        retain_reason = item.get("retain_reason")

        if retain_reason:
            retained_words.append(
                {
                    "name": name,
                    "retain_reason": retain_reason,
                    "data_days": result.get("data_days"),
                    "data_start_ymd": result.get("data_start_ymd"),
                    "data_end_ymd": result.get("data_end_ymd"),
                    "is_internal_demand_matched": item.get("is_internal_demand_matched"),
                    "matched_demand": str(item.get("matched_demand") or ""),
                    "senior_fit_score": item.get("senior_fit_score"),
                    "is_final_retained": bool(item.get("is_final_retained")),
                }
            )

        existing_record = existing_records.get(name)
        if _is_finalized_record(existing_record):
            summary["finalized_resumed"] += 1
            record_id = int(item.get("record_id") or existing_record.get("id") or 0)
            if verbose:
                print(f"resume finalized word={name}")
        else:
            record_id = _persist_pending_item_record(
                repository,
                item,
                analyze_ymd=analyze_ymd,
                demand_cache_run_id=demand_cache_run_id,
                phase="finalized",
                senior_threshold=senior_threshold,
                sustained_threshold=sustained_threshold,
                spike_days=spike_days,
                spike_ratio=spike_ratio,
                spike_baseline_floor=spike_baseline_floor,
                rising_overall_change_rate=rising_overall_change_rate,
                rising_window_change_rate=rising_window_change_rate,
                rising_adjacent_up_ratio=rising_adjacent_up_ratio,
                dry_run=dry_run,
                skip_db_save=skip_db_save,
                verbose=verbose,
            )

        export_rows.append(
            _build_export_row_from_item(
                item,
                analyze_ymd=analyze_ymd,
                strategy=strategy,
            )
        )

        if item.get("is_final_retained") and strategy:
            final_hive_rows.append(
                build_wxindex_word_hive_row(
                    wxindex_word_record_id=record_id,
                    word=name,
                    strategy=strategy,
                    partition_dt=analyze_ymd,
                    max_score=result.get("max_score"),
                )
            )

    if (
        final_hive_rows
        and strategy
        and config is not None
        and not dry_run
        and not skip_odps
    ):
        odps_summary = sync_wxindex_word_rows_to_odps(
            config,
            repository,
            hive_rows=final_hive_rows,
            partition_dt=analyze_ymd,
            strategy=strategy,
        )
        summary["odps_written"] = odps_summary.get("written_count", 0)
        summary["odps_synced"] = odps_summary.get("odps_synced", 0)
        summary["odps_sync"] = odps_summary
    elif dry_run or skip_odps:
        summary["odps_written"] = len(final_hive_rows)
        summary["odps_synced"] = len(final_hive_rows)

    _refresh_wxindex_heat_job_summary(
        summary,
        pending_items=pending_items,
        export_rows=export_rows,
    )
    summary["retained_words"] = retained_words
    summary["export_rows"] = export_rows
    return summary


WXINDEX_HEAT_PATTERN_EXPORT_FIELDS = [
    "analyze_ymd",
    "name",
    "meta_id",
    "fetch_start_ymd",
    "fetch_end_ymd",
    "data_start_ymd",
    "data_end_ymd",
    "data_days",
    "analysis_skipped",
    "skip_reason",
    "is_sustained_high",
    "is_rising",
    "is_spike",
    "retain_reason",
    "is_internal_demand_matched",
    "matched_demand",
    "senior_fit_score",
    "is_final_retained",
    "min_score",
    "max_score",
    "avg_score",
    "demand_id",
    "weight",
]


def write_wxindex_heat_pattern_csv(
    rows: list[dict[str, Any]],
    output_path: str | Path,
    *,
    fieldnames: list[str] | None = None,
) -> Path:
    """将热度分析明细写入本地 CSV。"""
    path = Path(output_path).expanduser()
    if not path.is_absolute():
        path = Path.cwd() / path
    path.parent.mkdir(parents=True, exist_ok=True)
    columns = fieldnames or WXINDEX_HEAT_PATTERN_EXPORT_FIELDS
    with path.open("w", encoding="utf-8-sig", newline="") as handle:
        writer = csv.DictWriter(handle, fieldnames=columns, extrasaction="ignore")
        writer.writeheader()
        for row in rows:
            writer.writerow({column: row.get(column, "") for column in columns})
    return path