ummaya 0.2.2 → 0.2.4

package/src/ummaya/tools/search.py

@@ -40,6 +40,13 @@ _ADMIN_LOCATION_RE = re.compile(
     r"(?:[가-힣]{2,}(?:시|군|구|동|읍|면)\b|[가-힣0-9]{2,}(?<!으)(?:로|길)\b)"
 )
 _EMERGENCY_RE = re.compile(r"(응급|응급실|응급의료|\bemergency\b|\ber\b)", re.IGNORECASE)
+_IMPLICIT_EMERGENCY_RE = re.compile(
+    r"(사람이\s*(?:쓰러|쓰러졌|쓰러져)|의식(?:을)?\s*(?:잃|없)|"
+    r"갑자기\s*쓰러|쓰러진\s*사람|위급|심정지|호흡(?:이)?\s*없|"
+    r"collapsed|unconscious|cardiac\s*arrest)",
+    re.IGNORECASE,
+)
+_AED_RE = re.compile(r"(\bAED\b|자동심장충격기|자동제세동기|제세동기)", re.IGNORECASE)
 _TRAFFIC_HAZARD_RE = re.compile(
     r"(교통사고|사고\s*위험|사고다발|위험\s*(?:구간|도로|지점)|어린이보호구역|보호구역|"
     r"도로\s*구간|accident|hazard|hotspot)",
@@ -50,6 +57,604 @@ _TRAFFIC_HAZARD_SPECIFIC_RE = re.compile(
     r"도로\s*구간|행정동코드|adm_cd|hazard|hotspot)",
     re.IGNORECASE,
 )
+_KMA_ANALYSIS_CHART_RE = re.compile(
+    r"(분석일기도|지상일기도|보조일기도|WthrChartInfoService|getSurfaceChart|"
+    r"getAuxillaryChart|synoptic\s+chart)",
+    re.IGNORECASE,
+)
+_KMA_GIMHAE_AIRPORT_RE = re.compile(r"(김해(?:공항)?|Gimhae|RKPK)", re.IGNORECASE)
+_KMA_GIMPO_AIRPORT_RE = re.compile(r"(김포(?:공항)?|Gimpo|RKSS)", re.IGNORECASE)
+_KMA_AIRPORT_NAME_RE = re.compile(
+    r"(공항|\bairport\b|\bRK[A-Z]{2}\b|station\s*\d{2,3})",
+    re.IGNORECASE,
+)
+_KMA_AIRPORT_AVIATION_RE = re.compile(
+    r"(AMOS|METAR|SPECI|RVR|항공기상|공항기상|활주로|runway|aviation|"
+    r"비행기|항공편|비행편|이륙|착륙|결항|지연|운항|뜰\s*만|뜨나|뜰\s*수|"
+    r"flight|take\s*off|landing|delay|cancel)",
+    re.IGNORECASE,
+)
+_KMA_EXPLICIT_METAR_RE = re.compile(r"(\bMETAR\b|\bSPECI\b|해독자료)", re.IGNORECASE)
+_KMA_RUNWAY_AREA_RE = re.compile(
+    r"(AMOS|활주로|RVR|runway|시정|visibility|공항기상관측|매분)",
+    re.IGNORECASE,
+)
+_KMA_ANALYSIS_DATA_RE = re.compile(
+    r"(분석자료|이미\s*분석|고해상도\s*격자|객관분석|AWS\s*객관|지도\s*자료|"
+    r"일기도|분석일기도|비구름|바람\s*흐름|날씨\s*흐름|공식\s*기상자료|전국\s*날씨|"
+    r"synoptic|weather\s*chart|"
+    r"objective\s*analysis|high[-\s]?resolution|grid)",
+    re.IGNORECASE,
+)
+_KMA_LIFESTYLE_WEATHER_RE = re.compile(
+    r"(날씨|현재\s*기상|실황|관측|예보|기온|습도|풍속|지금\s*비|"
+    r"비\s*(?:와|오|올|내리)|우산|강수|소나기|산책|퇴근|"
+    r"current\s+weather|forecast|rain|umbrella|precipitation|temperature)",
+    re.IGNORECASE,
+)
+_HIRA_MEDICAL_DETAIL_RE = re.compile(
+    r"((병원|의료기관|의원).*(상세|진료과|진료과목|진료시간|주차)|"
+    r"(상세|진료시간|주차|응급실).*(병원|의료기관|의원)|ykiho|detail)",
+    re.IGNORECASE,
+)
+_MOIS_EMERGENCY_CALL_BOX_RE = re.compile(
+    r"(안전\s*비상벨|비상벨|긴급\s*신고함|긴급신고함|방범벨|"
+    r"emergency\s+call\s+box)",
+    re.IGNORECASE,
+)
+_GYERYONG_ASSISTIVE_CHARGER_RE = re.compile(
+    r"((전동보장구|전동\s*휠체어|보장구|장애인).*(충전|충전소|충전장소)|"
+    r"(충전|충전소|충전장소).*(전동보장구|전동\s*휠체어|보장구|장애인)|"
+    r"계룡시?.*(충전소|충전\s*장소))",
+    re.IGNORECASE,
+)
+_MOF_OCEAN_WATER_QUALITY_RE = re.compile(
+    r"(해양\s*수질|해양수질|수질\s*자동\s*측정|용존산소|\bpH\b|"
+    r"water\s+quality|ocean\s+water)",
+    re.IGNORECASE,
+)
+_PPS_SHOPPING_RE = re.compile(
+    r"(종합\s*쇼핑몰|쇼핑몰|계약\s*물품|물품\s*조회|shopping\s*mall)",
+    re.IGNORECASE,
+)
+_PPS_BID_RE = re.compile(
+    r"(입찰|나라장터|조달청|\bbid\b|procurement|tender)",
+    re.IGNORECASE,
+)
+_KCUE_ACADEMY_INFO_RE = re.compile(
+    r"(대학알리미|대학정보공시|학교구분코드|schl[_\s-]?div[_\s-]?cd|KCUE)",
+    re.IGNORECASE,
+)
+_KCUE_REGIONAL_FINANCE_RE = re.compile(
+    r"(지역별\s*(등록금|재정)|등록금\s*(현황|지역별)?|tuition|finance)",
+    re.IGNORECASE,
+)
+_KCUE_REGIONAL_FOREIGN_STUDENT_RE = re.compile(
+    r"(외국인\s*유학생|유학생\s*현황|foreign\s+student|international\s+student)",
+    re.IGNORECASE,
+)
+_KCUE_TOOL_IDS = frozenset(
+    {
+        "kcue_finance_regional_tuition",
+        "kcue_student_regional_foreign",
+    }
+)
+_KMA_ANALYSIS_MAP_RE = re.compile(
+    r"(일기도|분석일기도|지도\s*자료|비구름|바람\s*흐름|날씨\s*흐름|전국\s*날씨|"
+    r"synoptic|weather\s*chart)",
+    re.IGNORECASE,
+)
+_KMA_ANALYSIS_POINT_RE = re.compile(
+    r"(주변|근처|특정지점|좌표|위도|경도|\blat\b|\blon\b|공항\s*주변)",
+    re.IGNORECASE,
+)
+_KMA_URL_AIR_TOOL_IDS = frozenset(
+    {
+        "kma_apihub_url_air_amos_minute",
+        "kma_apihub_url_air_metar_decoded",
+    }
+)
+_KMA_ANALYSIS_TOOL_IDS = frozenset(
+    {
+        "kma_apihub_url_high_resolution_grid_point",
+        "kma_apihub_url_aws_objective_analysis_grid",
+        "kma_apihub_url_analysis_weather_chart_image",
+    }
+)
+_KMA_LIFESTYLE_WEATHER_TOOL_IDS = frozenset(
+    {
+        "kma_current_observation",
+        "kma_ultra_short_term_forecast",
+        "kma_short_term_forecast",
+    }
+)
+_LOCATION_TOOL_IDS = frozenset(
+    {
+        "locate",
+        "kakao_address_search",
+        "kakao_keyword_search",
+        "kakao_coord_to_region",
+        "juso_adm_cd_lookup",
+        "sgis_adm_cd_lookup",
+    }
+)
+
+
+def _is_kma_analysis_point_query(query: str, *, is_analysis_map_query: bool) -> bool:
+    return bool(_KMA_ANALYSIS_POINT_RE.search(query)) and not is_analysis_map_query
+
+
+def _is_airport_aviation_query(query: str) -> bool:
+    return bool(
+        (
+            _KMA_AIRPORT_NAME_RE.search(query)
+            or _KMA_GIMHAE_AIRPORT_RE.search(query)
+            or _KMA_GIMPO_AIRPORT_RE.search(query)
+        )
+        and _KMA_AIRPORT_AVIATION_RE.search(query)
+    )
+
+
+def _is_lifestyle_weather_query(query: str, *, is_airport_aviation_query: bool) -> bool:
+    return bool(
+        _KMA_LIFESTYLE_WEATHER_RE.search(query)
+        and not is_airport_aviation_query
+        and not _is_emergency_chain_query(query)
+        and not _KMA_ANALYSIS_DATA_RE.search(query)
+        and not _TRAFFIC_HAZARD_RE.search(query)
+        and not _MOF_OCEAN_WATER_QUALITY_RE.search(query)
+    )
+
+
+def _is_pps_bid_query(query: str) -> bool:
+    return bool(_PPS_BID_RE.search(query) and not _PPS_SHOPPING_RE.search(query))
+
+
+def _is_kcue_regional_query(query: str) -> bool:
+    has_kcue_anchor = bool(_KCUE_ACADEMY_INFO_RE.search(query)) or (
+        bool(re.search(r"대학(?!병원)", query)) and "공식" in query
+    )
+    if not has_kcue_anchor:
+        return False
+    return bool(
+        _KCUE_REGIONAL_FINANCE_RE.search(query) or _KCUE_REGIONAL_FOREIGN_STUDENT_RE.search(query)
+    )
+
+
+def _is_emergency_chain_query(query: str) -> bool:
+    if _MOIS_EMERGENCY_CALL_BOX_RE.search(query):
+        return False
+    return bool(_IMPLICIT_EMERGENCY_RE.search(query) or _EMERGENCY_RE.search(query))
+
+
+def _filter_kma_analysis_scores(
+    scored: list[tuple[str, float]],
+    *,
+    is_analysis_map_query: bool,
+    is_analysis_point_query: bool,
+    prefer_poi_location: bool,
+) -> list[tuple[str, float]]:
+    chart_boost = 900.0 if is_analysis_map_query else 150.0
+    if is_analysis_point_query and not is_analysis_map_query:
+        chart_boost = -20.0
+    analysis_boosts = {
+        "kma_apihub_url_analysis_weather_chart_image": chart_boost,
+        "kma_apihub_url_high_resolution_grid_point": (900.0 if is_analysis_point_query else 450.0),
+        "kma_apihub_url_aws_objective_analysis_grid": (800.0 if is_analysis_point_query else 400.0),
+    }
+    allowed_location_ids = _LOCATION_TOOL_IDS if is_analysis_point_query else frozenset()
+    adjusted: list[tuple[str, float]] = []
+    for tool_id, score in scored:
+        if tool_id in _KMA_ANALYSIS_TOOL_IDS:
+            adjusted.append((tool_id, score + analysis_boosts.get(tool_id, 0.0)))
+            continue
+        if tool_id in allowed_location_ids:
+            location_score = max(1.0, score - 10.0)
+            if prefer_poi_location and tool_id == "kakao_keyword_search":
+                location_score += 30.0
+            elif prefer_poi_location and tool_id == "kakao_address_search":
+                location_score = max(1.0, location_score - 15.0)
+            adjusted.append((tool_id, location_score))
+    return adjusted
+
+
+def _kma_lifestyle_weather_additions() -> list[str]:
+    return [
+        "기상청",
+        "KMA",
+        "현재날씨",
+        "초단기실황",
+        "초단기예보",
+        "단기예보",
+        "강수",
+        "우산",
+        "nx",
+        "ny",
+        "base_date",
+        "base_time",
+        "current",
+        "observation",
+        "forecast",
+        "precipitation",
+    ]
+
+
+def _airport_aviation_additions(query: str) -> list[str]:
+    additions = [
+        "METAR",
+        "SPECI",
+        "AMOS",
+        "항공기상",
+        "공항기상",
+        "항공",
+        "비행기",
+        "항공편",
+        "운항",
+        "이륙",
+        "시정",
+        "RVR",
+        "wind",
+        "visibility",
+    ]
+    if _KMA_GIMPO_AIRPORT_RE.search(query) and _KMA_RUNWAY_AREA_RE.search(query):
+        additions.extend(["AMOS", "공항기상관측", "매분자료", "활주로", "김포공항", "stn110"])
+    return additions
+
+
+def _kma_analysis_data_additions() -> list[str]:
+    return [
+        "분석자료",
+        "고해상도",
+        "격자자료",
+        "객관분석",
+        "AWS",
+        "분석일기도",
+        "지도",
+        "비구름",
+        "바람흐름",
+        "objective",
+        "analysis",
+        "grid",
+        "chart",
+    ]
+
+
+def _traffic_hazard_additions() -> list[str]:
+    return [
+        "교통사고",
+        "사고다발구역",
+        "위험지점",
+        "도로위험구역",
+        "어린이보호구역",
+        "행정동코드",
+        "KOROAD",
+        "accident",
+        "hazard",
+    ]
+
+
+def _emergency_chain_additions(query: str) -> list[str]:
+    if not _is_emergency_chain_query(query):
+        return []
+    additions = [
+        "응급실",
+        "응급의료",
+        "자동심장충격기",
+        "AED",
+        "국립중앙의료원",
+        "NMC",
+        "nearby",
+        "emergency",
+        "hospital",
+    ]
+    if _POI_LOCATION_RE.search(query):
+        additions.extend(["장소", "키워드", "POI", "랜드마크", "역", "keyword"])
+    return additions
+
+
+def _public_safety_location_additions(query: str) -> list[str]:
+    additions: list[str] = []
+    if _MOIS_EMERGENCY_CALL_BOX_RE.search(query):
+        additions.extend(
+            [
+                "안전비상벨",
+                "비상벨",
+                "긴급신고함",
+                "방범",
+                "행정안전부",
+                "MOIS",
+                "emergency",
+                "call",
+                "box",
+            ]
+        )
+    if _GYERYONG_ASSISTIVE_CHARGER_RE.search(query):
+        additions.extend(
+            [
+                "계룡시",
+                "전동보장구",
+                "전동휠체어",
+                "장애인",
+                "충전소",
+                "충전장소",
+                "accessibility",
+                "charger",
+            ]
+        )
+    return additions
+
+
+def _ocean_water_quality_additions(query: str) -> list[str]:
+    if not _MOF_OCEAN_WATER_QUALITY_RE.search(query):
+        return []
+    return [
+        "해양수산부",
+        "해양수질",
+        "수질자동측정망",
+        "관측소",
+        "SEA3003",
+        "용존산소",
+        "water",
+        "quality",
+        "ocean",
+    ]
+
+
+def _pps_bid_additions(query: str) -> list[str]:
+    if not _is_pps_bid_query(query):
+        return []
+    return [
+        "조달청",
+        "나라장터",
+        "입찰공고",
+        "공사입찰",
+        "bidNtceNm",
+        "inqryBgnDt",
+        "inqryEndDt",
+        "PPS",
+        "bid",
+        "procurement",
+    ]
+
+
+def _kcue_regional_additions(query: str) -> list[str]:
+    if not _is_kcue_regional_query(query):
+        return []
+    additions = [
+        "한국대학교육협의회",
+        "대학알리미",
+        "대학정보공시",
+        "학교구분코드",
+        "schlDivCd",
+        "지역별통계",
+        "KCUE",
+    ]
+    if _KCUE_REGIONAL_FINANCE_RE.search(query):
+        additions.extend(["재정현황", "등록금", "FinancesService", "regional tuition"])
+    if _KCUE_REGIONAL_FOREIGN_STUDENT_RE.search(query):
+        additions.extend(["학생현황", "외국인유학생", "regional foreign student"])
+    return additions
+
+
+def _health_detail_additions(query: str) -> list[str]:
+    if not _HIRA_MEDICAL_DETAIL_RE.search(query):
+        return []
+    return [
+        "의료기관",
+        "상세정보",
+        "진료과목",
+        "진료시간",
+        "주차",
+        "요양기호",
+        "ykiho",
+        "HIRA",
+        "hospital",
+        "detail",
+    ]
+
+
+def _filter_kma_lifestyle_weather_scores(
+    scored: list[tuple[str, float]],
+) -> list[tuple[str, float]]:
+    allowed_tool_ids = _KMA_LIFESTYLE_WEATHER_TOOL_IDS | _LOCATION_TOOL_IDS
+    if any(tool_id in allowed_tool_ids for tool_id, _ in scored):
+        scored = [(tool_id, score) for tool_id, score in scored if tool_id in allowed_tool_ids]
+    boosts = {
+        "kakao_keyword_search": 1100.0,
+        "kakao_address_search": 1000.0,
+        "kma_current_observation": 900.0,
+        "kma_ultra_short_term_forecast": 800.0,
+        "kma_short_term_forecast": 650.0,
+        "kakao_coord_to_region": 260.0,
+        "juso_adm_cd_lookup": 260.0,
+        "sgis_adm_cd_lookup": 260.0,
+    }
+    return [(tool_id, score + boosts.get(tool_id, 0.0)) for tool_id, score in scored]
+
+
+def _filter_public_safety_location_scores(
+    query: str, scored: list[tuple[str, float]]
+) -> list[tuple[str, float]]:
+    boosts = {}
+    if _MOIS_EMERGENCY_CALL_BOX_RE.search(query):
+        boosts["mois_emergency_call_box_lookup"] = 1000.0
+    if _GYERYONG_ASSISTIVE_CHARGER_RE.search(query):
+        boosts["gyeryong_assistive_device_charging_place_locate"] = 1000.0
+    if not boosts:
+        return scored
+    return [(tool_id, score + boosts.get(tool_id, 0.0)) for tool_id, score in scored]
+
+
+def _filter_emergency_chain_scores(
+    query: str, scored: list[tuple[str, float]]
+) -> list[tuple[str, float]]:
+    if not _is_emergency_chain_query(query):
+        return scored
+    emergency_tool_ids = {
+        "nmc_emergency_search",
+        "nmc_aed_site_locate",
+        "hira_hospital_search",
+        "hira_medical_institution_detail",
+    }
+    allowed_tool_ids = emergency_tool_ids | _LOCATION_TOOL_IDS
+    if any(tool_id in emergency_tool_ids for tool_id, _ in scored):
+        scored = [(tool_id, score) for tool_id, score in scored if tool_id in allowed_tool_ids]
+    implicit_collapse = bool(_IMPLICIT_EMERGENCY_RE.search(query))
+    boosts = {
+        "nmc_emergency_search": 1200.0,
+        "nmc_aed_site_locate": 1150.0 if implicit_collapse else 950.0,
+        "kakao_keyword_search": 1100.0 if implicit_collapse else 900.0,
+        "kakao_address_search": 800.0,
+        "kakao_coord_to_region": 500.0,
+        "juso_adm_cd_lookup": 300.0,
+        "sgis_adm_cd_lookup": 300.0,
+        "hira_hospital_search": 250.0,
+        "hira_medical_institution_detail": 200.0,
+    }
+    return [(tool_id, score + boosts.get(tool_id, 0.0)) for tool_id, score in scored]
+
+
+def _filter_ocean_water_quality_scores(
+    query: str, scored: list[tuple[str, float]]
+) -> list[tuple[str, float]]:
+    if not _MOF_OCEAN_WATER_QUALITY_RE.search(query):
+        return scored
+    return [
+        (
+            tool_id,
+            score + (1000.0 if tool_id == "mof_ocean_water_quality_check" else 0.0),
+        )
+        for tool_id, score in scored
+    ]
+
+
+def _filter_pps_bid_scores(query: str, scored: list[tuple[str, float]]) -> list[tuple[str, float]]:
+    if not _is_pps_bid_query(query):
+        return scored
+    has_pps_bid = any(tool_id == "pps_bid_public_info" for tool_id, _ in scored)
+    if not has_pps_bid:
+        return scored
+    return [
+        (tool_id, score + 1000.0) for tool_id, score in scored if tool_id == "pps_bid_public_info"
+    ]
+
+
+def _filter_kcue_regional_scores(
+    query: str, scored: list[tuple[str, float]]
+) -> list[tuple[str, float]]:
+    if not _is_kcue_regional_query(query):
+        return scored
+    has_kcue = any(tool_id in _KCUE_TOOL_IDS for tool_id, _ in scored)
+    if not has_kcue:
+        return scored
+
+    prefer_finance = bool(_KCUE_REGIONAL_FINANCE_RE.search(query))
+    prefer_foreign_student = bool(_KCUE_REGIONAL_FOREIGN_STUDENT_RE.search(query))
+    boosts = {
+        "kcue_finance_regional_tuition": 1000.0 if prefer_finance else 700.0,
+        "kcue_student_regional_foreign": 1000.0 if prefer_foreign_student else 700.0,
+    }
+    return [
+        (tool_id, score + boosts[tool_id]) for tool_id, score in scored if tool_id in _KCUE_TOOL_IDS
+    ]
+
+
+def _filter_health_detail_scores(
+    query: str, scored: list[tuple[str, float]]
+) -> list[tuple[str, float]]:
+    if not _HIRA_MEDICAL_DETAIL_RE.search(query):
+        return scored
+    return [
+        (
+            tool_id,
+            score + (650.0 if tool_id == "hira_medical_institution_detail" else 0.0),
+        )
+        for tool_id, score in scored
+    ]
+
+
+def _filter_initial_special_scores(
+    query: str, scored: list[tuple[str, float]]
+) -> list[tuple[str, float]]:
+    if _TRAFFIC_HAZARD_SPECIFIC_RE.search(query):
+        scored = [
+            (tool_id, score) for tool_id, score in scored if tool_id != "koroad_accident_search"
+        ]
+    if _KMA_ANALYSIS_CHART_RE.search(query):
+        return [
+            (tool_id, score)
+            for tool_id, score in scored
+            if tool_id == "kma_apihub_url_analysis_weather_chart_image"
+        ]
+    return scored
+
+
+def _filter_kma_aviation_scores(
+    query: str, scored: list[tuple[str, float]], *, is_airport_aviation_query: bool
+) -> list[tuple[str, float]]:
+    if _KMA_GIMHAE_AIRPORT_RE.search(query) and _KMA_AIRPORT_AVIATION_RE.search(query):
+        scored = [
+            (tool_id, score)
+            for tool_id, score in scored
+            if tool_id != "kma_apihub_url_air_amos_minute"
+        ]
+    if (
+        _KMA_GIMPO_AIRPORT_RE.search(query)
+        and _KMA_RUNWAY_AREA_RE.search(query)
+        and _KMA_AIRPORT_AVIATION_RE.search(query)
+    ):
+        scored = [
+            (tool_id, score + 500.0 if tool_id == "kma_apihub_url_air_amos_minute" else score)
+            for tool_id, score in scored
+        ]
+    if not is_airport_aviation_query:
+        return scored
+
+    has_air_url_candidate = any(tool_id in _KMA_URL_AIR_TOOL_IDS for tool_id, _ in scored)
+    if not has_air_url_candidate:
+        return scored
+    if _KMA_EXPLICIT_METAR_RE.search(query):
+        blocked_tool_ids = (
+            _LOCATION_TOOL_IDS | _KMA_LIFESTYLE_WEATHER_TOOL_IDS | {"kma_forecast_fetch"}
+        )
+        scored = [(tool_id, score) for tool_id, score in scored if tool_id not in blocked_tool_ids]
+    else:
+        scored = [(tool_id, score) for tool_id, score in scored if tool_id in _KMA_URL_AIR_TOOL_IDS]
+    prefer_amos = bool(
+        _KMA_GIMPO_AIRPORT_RE.search(query)
+        and _KMA_RUNWAY_AREA_RE.search(query)
+        and not _KMA_GIMHAE_AIRPORT_RE.search(query)
+    )
+    return [
+        (
+            tool_id,
+            score
+            + (
+                800.0
+                if tool_id == "kma_apihub_url_air_amos_minute" and prefer_amos
+                else 700.0
+                if tool_id == "kma_apihub_url_air_metar_decoded"
+                else 0.0
+            ),
+        )
+        for tool_id, score in scored
+    ]
+
+
+def _boost_aed_scores(query: str, scored: list[tuple[str, float]]) -> list[tuple[str, float]]:
+    if not _AED_RE.search(query):
+        return scored
+    return [
+        (
+            tool_id,
+            score + 900.0
+            if tool_id == "nmc_aed_site_locate"
+            else score + 700.0
+            if tool_id == "nmc_emergency_search" and _EMERGENCY_RE.search(query)
+            else score,
+        )
+        for tool_id, score in scored
+    ]
 
 
 def _expand_query_for_adapter_retrieval(query: str) -> str:
@@ -63,31 +668,75 @@ def _expand_query_for_adapter_retrieval(query: str) -> str:
     visible to the model.
     """
     additions: list[str] = []
-    if _POI_LOCATION_RE.search(query):
+    is_airport_aviation_query = _is_airport_aviation_query(query)
+    if is_airport_aviation_query:
+        additions.extend(_airport_aviation_additions(query))
+    if _KMA_ANALYSIS_DATA_RE.search(query):
+        additions.extend(_kma_analysis_data_additions())
+    if _is_lifestyle_weather_query(query, is_airport_aviation_query=is_airport_aviation_query):
+        additions.extend(_kma_lifestyle_weather_additions())
+    if _POI_LOCATION_RE.search(query) and not is_airport_aviation_query:
         additions.extend(["장소", "키워드", "POI", "랜드마크", "역", "keyword"])
     if _ADMIN_LOCATION_RE.search(query):
         additions.extend(["주소", "행정동", "법정동", "도로명", "지번", "address"])
     if _EMERGENCY_RE.search(query):
         additions.extend(["응급실", "응급의료", "NMC", "emergency"])
+    if _AED_RE.search(query):
+        additions.extend(["AED", "자동심장충격기", "자동제세동기", "국립중앙의료원"])
+    additions.extend(_emergency_chain_additions(query))
+    additions.extend(_pps_bid_additions(query))
+    additions.extend(_kcue_regional_additions(query))
+    additions.extend(_ocean_water_quality_additions(query))
+    additions.extend(_health_detail_additions(query))
+    additions.extend(_public_safety_location_additions(query))
     if _TRAFFIC_HAZARD_RE.search(query):
-        additions.extend(
-            [
-                "교통사고",
-                "사고다발구역",
-                "위험지점",
-                "도로위험구역",
-                "어린이보호구역",
-                "행정동코드",
-                "KOROAD",
-                "accident",
-                "hazard",
-            ]
-        )
+        additions.extend(_traffic_hazard_additions())
     if not additions:
         return query
     return f"{query} {' '.join(additions)}"
 
 
+def _filter_special_case_scores(
+    query: str, scored: list[tuple[str, float]]
+) -> list[tuple[str, float]]:
+    """Apply deterministic domain disambiguation after backend scoring."""
+    is_airport_aviation_query = _is_airport_aviation_query(query)
+    is_analysis_query = bool(_KMA_ANALYSIS_DATA_RE.search(query))
+    is_analysis_map_query = bool(_KMA_ANALYSIS_MAP_RE.search(query))
+    is_analysis_point_query = _is_kma_analysis_point_query(
+        query, is_analysis_map_query=is_analysis_map_query
+    )
+    is_lifestyle_weather_query = _is_lifestyle_weather_query(
+        query, is_airport_aviation_query=is_airport_aviation_query
+    )
+    scored = _filter_initial_special_scores(query, scored)
+    if is_analysis_query:
+        scored = _filter_kma_analysis_scores(
+            scored,
+            is_analysis_map_query=is_analysis_map_query,
+            is_analysis_point_query=is_analysis_point_query,
+            prefer_poi_location=bool(_POI_LOCATION_RE.search(query)),
+        )
+    if is_lifestyle_weather_query:
+        scored = _filter_kma_lifestyle_weather_scores(scored)
+    scored = _filter_emergency_chain_scores(query, scored)
+    scored = _filter_pps_bid_scores(query, scored)
+    scored = _filter_kcue_regional_scores(query, scored)
+    scored = _filter_health_detail_scores(query, scored)
+    scored = _filter_public_safety_location_scores(query, scored)
+    scored = _filter_ocean_water_quality_scores(query, scored)
+    scored = _filter_kma_aviation_scores(
+        query, scored, is_airport_aviation_query=is_airport_aviation_query
+    )
+    scored = _boost_aed_scores(query, scored)
+
+    query_lower = query.lower()
+    return [
+        (tool_id, score + 1000.0 if tool_id.lower() in query_lower else score)
+        for tool_id, score in scored
+    ]
+
+
 def search(
     query: str,
     bm25_index: BM25Index,
@@ -161,10 +810,7 @@ def search(
             )
             return []
 
-    if _TRAFFIC_HAZARD_SPECIFIC_RE.search(query):
-        scored = [
-            (tool_id, score) for tool_id, score in scored if tool_id != "koroad_accident_search"
-        ]
+    scored = _filter_special_case_scores(query, scored)
 
     # Enforce the deterministic tie-break once, here. Backend-internal
     # orderings are not trusted (HybridBackend returns unordered union).