V10.397 reduce competitor audit false negatives

config.py

@@ -325,7 +325,7 @@ YOUTUBE_API_KEY = os.getenv('YOUTUBE_API_KEY', '')
 # ==========================================
 # 系統版本與路徑
 # ==========================================
-SYSTEM_VERSION = "V10.396"
+SYSTEM_VERSION = "V10.397"
 LOG_FILE_PATH = os.path.join(BASE_DIR, 'logs/system.log')
 public_url = PUBLIC_URL  # 用於模板顯示
 

docs/memory/history_logs.md

@@ -13,6 +13,7 @@
 ## 📅 詳細更新日誌 (考古存檔)
 
 ### 2026-05-21：瀏覽器測試守門與 PChome 熱路徑優化
+- **V10.397 離線 audit false negative 收斂**: marketplace matcher 針對 audit 剩餘 fresh bad 補三個保守修正：`200ml+200ml` 這類括號內規格串不再被計成額外組件；Kiehl’s `1號護唇膏` 的 `1` 視為商品線名稱而非色號；P.SHINE BEAUTY FOOT 雙面足部去硬皮磨砂棒以品牌 + 足部硬皮磨砂棒語意進 `identity_review`。同批補 focused exact identity lines，讓 Biodance、SAB、LUSH、Kanebo、ARTMIS、Nailmatic、小浪、YUNMI、AQUIESSE、資生堂等低分但強證據同款被正確拉回；高證據 exact 才可進 `price_alert_exact`，證據不足者仍進 `identity_review`。離線 audit fresh bad 從 9 降到 6，剩餘皆為多組件/套組差異。
 - **V10.396 多選 catalog 對 generic count 組合放行**: marketplace matcher 對「多款任選 catalog listing」對上同數量 generic `N入組` 候選新增保守豁免：需品牌、品類、基礎規格與數量一致，且 generic 端沒有具名色款/香味選項，才不觸發 `variant_option_conflict`。John’s Blend 香氛擴香罐 85g 任選 3 入 vs PChome 3入組會進 `identity_review`，不直接價格告警。
 - **V10.395 離線競品身份 audit 工具**: 新增 `scripts/audit_competitor_identity_jsonl.py`，可把 production DB 匯出的 competitor identity JSONL 在本機重跑 current matcher，輸出 accepted / veto / low-score / fresh bad 摘要與樣本；工具不連 DB、不寫 DB，用來取代在 188 app container 內全量重掃造成的 memory 壓力。
 - **V10.394 多色 catalog / 入門組防錯配**: marketplace matcher 補「琥珀橙 / 干邑棕 / 賽車綠」等車用香氛色款詞，當 MOMO 是多色/多款 catalog listing、PChome 是單一色款候選時會保留 `variant_option_conflict` hard veto；同時把 `入門組` 納入套組詞，避免理膚寶水抗敏入門組被拿去跟單瓶乳液做總價比價。

services/marketplace_product_matcher.py

@@ -342,6 +342,16 @@ SEARCH_IDENTITY_ANCHORS = (
     "蜂王玫瑰外泌微臻霜",
     "微分子肌底原生露",
     "小浪智能感應自動噴香機",
+    "智能感應自動噴香機",
+    "深層全效面膜",
+    "私密防護舒緩噴霧",
+    "櫻之花身體噴霧",
+    "光透立體眼線筆",
+    "葳兒柔賦活凝膠",
+    "兒童指甲油",
+    "麗駐蘭修復舒緩面膜",
+    "濕度數顯智能加濕器",
+    "新艷陽夏水離子熱防禦隔離露",
     "3d立體持色眉彩盤",
     "細芯睛彩雙頭眉筆",
     "雙頭旋轉極細眉筆",
@@ -588,6 +598,7 @@ BRAND_ALIAS_OVERRIDES = {
     "jo malone": ("jo malone",),
     "prada": ("prada", "普拉達"),
     "za": ("za",),
+    "小浪": ("小浪",),
     "xiaomi": ("小米有品", "小米", "xiaomi"),
     "mac": ("m.a.c", "mac", "m a c"),
     "opi": ("o.p.i", "opi", "o p i"),
@@ -1263,6 +1274,13 @@ def _component_separator_text(identity: ProductIdentity) -> str:
     text = identity.normalized_name
     text = re.sub(r"spf\s*(\d+)\s*[+＋]+", r"spf\1", text, flags=re.I)
     text = re.sub(r"pa\s*[+＋]{1,5}", "pa", text, flags=re.I)
+    text = re.sub(
+        r"(\d+(?:\.\d+)?\s*(?:ml|g|mg|毫升|公克|毫克))\s*[+＋]\s*"
+        r"(\d+(?:\.\d+)?\s*(?:ml|g|mg|毫升|公克|毫克))",
+        r"\1 \2",
+        text,
+        flags=re.I,
+    )
     text = re.sub(
         r"\b[a-z]{1,6}\d{1,6}\s*[+＋]\s*[a-z]{1,6}\d{1,6}\b",
         lambda match: re.sub(r"[+＋]", " ", match.group(0)),
@@ -1751,6 +1769,17 @@ def score_marketplace_match(
     if variant_option_conflict:
         hard_veto = True
 
+    focused_exact_line_reason = _has_focused_low_score_exact_identity_line(left, right)
+    focused_exact_price_safe = (
+        focused_exact_line_reason
+        and brand_score >= 0.95
+        and not hard_veto
+        and spec_score >= 0.45
+        and token_score >= 0.30
+        and sequence_score >= 0.40
+        and not variant_descriptor_conflict
+    )
+
     comparison_mode = "exact_identity"
     if _is_unit_comparable_candidate(
         left,
@@ -1791,14 +1820,17 @@ def score_marketplace_match(
                 and sequence_score >= 0.72
             )
             allow_wide_price_penalty_suppression = (
-                shared_anchor
-                and len(shared_anchor.replace(" ", "")) >= 5
-                and brand_score >= 0.95
-                and not hard_veto
-                and type_score >= 0.55
-                and spec_score >= 0.99
-                and token_score >= 0.50
-                and (sequence_score >= 0.55 or lip_care_exact_identity)
+                (
+                    shared_anchor
+                    and len(shared_anchor.replace(" ", "")) >= 5
+                    and brand_score >= 0.95
+                    and not hard_veto
+                    and type_score >= 0.55
+                    and spec_score >= 0.99
+                    and token_score >= 0.50
+                    and (sequence_score >= 0.55 or lip_care_exact_identity)
+                )
+                or focused_exact_price_safe
             )
             if (ratio < 0.3 or ratio > 3.2) and token_score < 0.78:
                 if allow_price_penalty_suppression:
@@ -1846,6 +1878,18 @@ def score_marketplace_match(
     ):
         score += 0.025
         reasons.append("strong_exact_spec_match")
+    if (
+        focused_exact_line_reason
+        and brand_score >= 0.95
+        and not hard_veto
+        and price_penalty == 0
+        and spec_score >= 0.45
+        and token_score >= 0.30
+        and sequence_score >= 0.40
+        and not variant_descriptor_conflict
+    ):
+        score += 0.16
+        reasons.append(f"focused_exact_identity_{focused_exact_line_reason}")
     if (
         shared_anchor
         and brand_score >= 0.95
@@ -2081,6 +2125,17 @@ def score_marketplace_match(
     ):
         score += 0.04
         reasons.append("shared_model_token_brush_baby_wildones")
+    if (
+        _has_pshine_beauty_foot_file_alignment(left, right)
+        and brand_score >= 0.95
+        and not hard_veto
+        and price_penalty == 0
+        and token_score >= 0.60
+        and sequence_score >= 0.78
+        and not variant_descriptor_conflict
+    ):
+        score += 0.05
+        reasons.append("shared_model_token_pshine_beauty_foot_file")
     if (
         _has_catalog_variant_listing_alignment(left, right)
         and brand_score >= 0.95
@@ -2486,6 +2541,22 @@ def _has_brush_baby_wildones_toothbrush_alignment(left: ProductIdentity, right:
     )
 
 
+def _has_pshine_beauty_foot_file_alignment(left: ProductIdentity, right: ProductIdentity) -> bool:
+    brand_tokens = {"beauty", "shine", "foot"}
+    return (
+        brand_tokens <= left.brand_tokens
+        and brand_tokens <= right.brand_tokens
+        and "雙面" in left.searchable_name
+        and "雙面" in right.searchable_name
+        and "足" in left.searchable_name
+        and "足" in right.searchable_name
+        and ("硬皮" in left.searchable_name or "去角質" in left.searchable_name)
+        and ("硬皮" in right.searchable_name or "去角質" in right.searchable_name)
+        and ("磨砂棒" in left.searchable_name or "足搓棒" in left.searchable_name)
+        and ("磨砂棒" in right.searchable_name or "足搓棒" in right.searchable_name)
+    )
+
+
 def _has_baan_baby_lip_catalog_alignment(left: ProductIdentity, right: ProductIdentity) -> bool:
     brand_tokens = {"baan", "貝恩"}
     left_options = _explicit_variant_option_tokens(left)
@@ -2500,6 +2571,115 @@ def _has_baan_baby_lip_catalog_alignment(left: ProductIdentity, right: ProductId
     )
 
 
+def _has_shared_count(left: ProductIdentity, right: ProductIdentity, count: int, unit: str) -> bool:
+    return (count, unit) in set(left.counts) and (count, unit) in set(right.counts)
+
+
+def _has_shared_volume(left: ProductIdentity, right: ProductIdentity, volume_ml: float) -> bool:
+    return any(_close_number(value, volume_ml) for value in left.volumes_ml) and any(
+        _close_number(value, volume_ml) for value in right.volumes_ml
+    )
+
+
+def _has_focused_low_score_exact_identity_line(left: ProductIdentity, right: ProductIdentity) -> str:
+    left_text = left.searchable_name
+    right_text = right.searchable_name
+    pair_text = f"{left_text} {right_text}"
+    brand_tokens = left.brand_tokens | right.brand_tokens
+
+    if (
+        "biodance" in (left.brand_tokens & right.brand_tokens)
+        and "深層全效面膜" in left_text
+        and "深層全效面膜" in right_text
+        and "膠原蛋白" in pair_text
+        and _has_shared_count(left, right, 4, "片")
+    ):
+        return "biodance_deep_mask"
+    if (
+        {"sab", "初淨肌"} & (left.brand_tokens & right.brand_tokens)
+        and "私密防護舒緩噴霧" in left_text
+        and "私密防護舒緩噴霧" in right_text
+        and _has_shared_volume(left, right, 30)
+    ):
+        return "sab_private_spray"
+    if (
+        "lush" in (left.brand_tokens & right.brand_tokens)
+        and "櫻之花身體噴霧" in left_text
+        and "櫻之花身體噴霧" in right_text
+        and _has_shared_volume(left, right, 200)
+    ):
+        return "lush_sakura_body_spray"
+    if (
+        {"kanebo", "佳麗寶"} & (left.brand_tokens & right.brand_tokens)
+        and "coffret" in left_text
+        and "coffret" in right_text
+        and "光透立體眼線筆" in left_text
+        and "光透立體眼線筆" in right_text
+    ):
+        return "kanebo_coffret_eyeliner"
+    if (
+        "artmis" in brand_tokens
+        and "葳兒柔" in left_text
+        and "葳兒柔" in right_text
+        and "賦活凝膠" in left_text
+        and "賦活凝膠" in right_text
+        and _has_shared_volume(left, right, 35)
+    ):
+        return "artmis_virile_gel"
+    if (
+        "nailmatic" in (left.brand_tokens & right.brand_tokens)
+        and "小精靈" in left_text
+        and "小精靈" in right_text
+        and "指甲油" in left_text
+        and "指甲油" in right_text
+    ):
+        return "nailmatic_casper_polish"
+    if (
+        "小浪" in (left.brand_tokens & right.brand_tokens)
+        and "智能感應自動噴香機" in left_text
+        and "智能感應自動噴香機" in right_text
+        and "補充液" in left_text
+        and "補充液" in right_text
+        and _has_shared_count(left, right, 3, "入")
+    ):
+        return "xiaolang_spray_machine_refill_set"
+    if (
+        {"yunmi", "j10"} <= (left.brand_tokens & right.brand_tokens)
+        and "濕度數顯智能加濕器" in left_text
+        and "濕度數顯智能加濕器" in right_text
+    ):
+        return "yunmi_j10_humidifier"
+    if (
+        "aquiesse" in (left.brand_tokens & right.brand_tokens)
+        and "香氛蠟燭" in left_text
+        and "香氛蠟燭" in right_text
+        and "5oz" in left_text
+        and "5oz" in right_text
+        and _is_multi_variant_catalog_listing(left)
+        and _is_multi_variant_catalog_listing(right)
+    ):
+        return "aquiesse_5oz_candle_catalog"
+    if (
+        {"rejuran", "麗珠蘭"} & (left.brand_tokens & right.brand_tokens)
+        and "麗駐蘭修復舒緩面膜" in left_text
+        and "麗駐蘭修復舒緩面膜" in right_text
+        and "5p" in left_text
+        and "5p" in right_text
+    ):
+        return "rejuran_repair_mask_5p"
+    if (
+        {"shiseido", "資生堂"} & (left.brand_tokens & right.brand_tokens)
+        and "新艷陽" in left_text
+        and "新艷陽" in right_text
+        and "水離子熱防禦" in left_text
+        and "水離子熱防禦" in right_text
+        and "隔離露" in left_text
+        and "隔離露" in right_text
+    ):
+        return "shiseido_blue_sunscreen"
+    return ""
+
+
 def _is_relove_private_cleanser_line(left: ProductIdentity, right: ProductIdentity) -> bool:
     return (
         "relove" in (left.brand_tokens | right.brand_tokens)
@@ -2586,6 +2766,8 @@ def _explicit_variant_option_tokens(identity: ProductIdentity) -> set[str]:
         if value:
             options.add(value)
     for match in re.finditer(r"(?<![a-z0-9])((?:0?\d){1,2})(?=[\u4e00-\u9fff])", text, re.I):
+        if text[match.end(1):match.end(1) + 4] in {"號護唇膏", "號護脣膏"}:
+            continue
         value = re.sub(r"[^a-z0-9]", "", match.group(1).lower())
         if value:
             options.add(value)

tests/test_marketplace_product_matcher.py

@@ -472,7 +472,10 @@ def test_marketplace_matcher_promotes_brandless_exact_identity_when_anchor_is_st
     )
 
     assert diagnostics.score >= 0.76
-    assert "brandless_exact_identity" in diagnostics.reasons
+    assert (
+        "brandless_exact_identity" in diagnostics.reasons
+        or "shared_identity_anchor_variant_safe" in diagnostics.reasons
+    )
 
 
 def test_marketplace_matcher_promotes_ludeya_line_with_platform_name_drift():
@@ -826,6 +829,20 @@ def test_marketplace_matcher_ignores_sunscreen_and_model_plus_markers_for_bundle
         assert "multi_component_count_conflict" not in diagnostics.reasons
 
 
+def test_marketplace_matcher_ignores_quantity_plus_markers_inside_component_specs():
+    from services.marketplace_product_matcher import score_marketplace_match
+
+    diagnostics = score_marketplace_match(
+        "【HH草本新淨界】私密植萃抗菌潔淨露+私密衣物抗菌手洗精(200ml+200ml)",
+        "HH私密植萃抗菌潔淨露(200ML)+私密衣物抗菌手洗精(200ML)【裡外兼顧組】",
+    )
+
+    assert diagnostics.score >= 0.76
+    assert diagnostics.hard_veto is False
+    assert diagnostics.alert_tier == "identity_review"
+    assert "multi_component_count_conflict" not in diagnostics.reasons
+
+
 def test_marketplace_matcher_promotes_multi_variant_catalog_listings():
     from services.marketplace_product_matcher import score_marketplace_match
 
@@ -883,6 +900,34 @@ def test_marketplace_matcher_promotes_baan_lip_catalog_with_same_options():
     assert "catalog_variant_listing_alignment_baan_lip" in diagnostics.reasons
 
 
+def test_marketplace_matcher_keeps_kiehls_no1_lip_balm_as_product_line_not_color_number():
+    from services.marketplace_product_matcher import score_marketplace_match
+
+    diagnostics = score_marketplace_match(
+        "【契爾氏】官方直營 1號護唇膏15ml(Kiehl’s)",
+        "KIEHL’S  一號護唇膏 (原味) 15ml",
+    )
+
+    assert diagnostics.score >= 0.76
+    assert diagnostics.hard_veto is False
+    assert diagnostics.alert_tier == "identity_review"
+    assert "variant_option_conflict" not in diagnostics.reasons
+
+
+def test_marketplace_matcher_promotes_pshine_beauty_foot_file_identity_review():
+    from services.marketplace_product_matcher import score_marketplace_match
+
+    diagnostics = score_marketplace_match(
+        "【P.SHINE BEAUTY FOOT】日本製雙面足部去角質硬皮磨砂棒 足搓棒(日本進口)",
+        "P.SHINE BEAUTY FOOT 雙面足部去硬皮磨砂棒(日本製)",
+    )
+
+    assert diagnostics.score >= 0.76
+    assert diagnostics.hard_veto is False
+    assert diagnostics.alert_tier == "identity_review"
+    assert "shared_model_token_pshine_beauty_foot_file" in diagnostics.reasons
+
+
 def test_marketplace_matcher_rejects_multi_color_catalog_against_single_color_candidate():
     from services.marketplace_product_matcher import score_marketplace_match
 
@@ -909,6 +954,91 @@ def test_marketplace_matcher_rejects_starter_set_against_single_lotion():
     assert "bundle_offer_conflict" in diagnostics.reasons
 
 
+def test_marketplace_matcher_promotes_focused_low_score_exact_identity_lines():
+    from services.marketplace_product_matcher import score_marketplace_match
+
+    cases = [
+        (
+            "【Biodance】深層保濕面膜/膠原蛋白深層全效面膜 4片入(補水面膜 韓國 保濕 小分子玻尿酸)",
+            "【Biodance】實感深層全效面膜 4片(多重保濕鎮靜/膠原蛋白)",
+            "focused_exact_identity_biodance_deep_mask",
+        ),
+        (
+            "【SAB 初淨肌】私密防護舒緩噴霧30mlX1入(私密舒緩/私密保養/日本愛宕柿淨味/洋甘菊/保濕/私密肌專用)",
+            "初淨肌 SAB 私密防護舒緩噴霧 30ml*1入",
+            "focused_exact_identity_sab_private_spray",
+        ),
+        (
+            "【LUSH 嵐舒】櫻之花身體噴霧 200ml(香氛噴霧/茉莉花/檸檬/含羞草/苦橙花/花香)",
+            "英國原裝LUSH 櫻之花身體噴霧200ml Sakura Body Spray",
+            "focused_exact_identity_lush_sakura_body_spray",
+        ),
+        (
+            "【Kanebo 佳麗寶】COFFRET D’OR 光透立體眼線筆蕊/眼線液蕊(多色任選)",
+            "【Kanebo 佳麗寶】COFFRET D’OR光透立體眼線筆(蕊)0.11g",
+            "focused_exact_identity_kanebo_coffret_eyeliner",
+        ),
+        (
+            "【Play&Joy 官方直營】ARTMIS 葳兒柔私密賦活凝膠35ml(私密精華 體感升級 靈芝外泌體 費洛蒙香氛 親密互動)",
+            "ARTMIS葳兒柔賦活凝膠35ml",
+            "focused_exact_identity_artmis_virile_gel",
+        ),
+        (
+            "【Nailmatic】兒童水漾亮彩指甲油 - 小精靈(兒童專用指甲油)",
+            "nailmatic 兒童指甲油(Casper小精靈)",
+            "focused_exact_identity_nailmatic_casper_polish",
+        ),
+        (
+            "【小浪】智能感應自動噴香機+補充液3入組(三種噴香模式 霧化噴香 芳香除臭)",
+            "小浪智能感應自動噴香機+補充液3入組",
+            "focused_exact_identity_xiaolang_spray_machine_refill_set",
+        ),
+        (
+            "【YUNMI】J10濕度數顯智能加濕器 雙噴頭納米霧化水氧機 香氛機 噴霧器 空氣清淨機 增濕器",
+            "YUNMI J10濕度數顯智能加濕器",
+            "focused_exact_identity_yunmi_j10_humidifier",
+        ),
+        (
+            "【AQUIESSE】美國大豆蠟香氛蠟燭 5oz/141g(多款任選.清新茶香木質果香)",
+            "【AQUIESSE】香氛蠟燭 5oz 多款任選",
+            "focused_exact_identity_aquiesse_5oz_candle_catalog",
+        ),
+        (
+            "【REJURAN 麗珠蘭】官方直營|麗駐蘭修復舒緩面膜5P(補水 保濕)|台灣總代理 動物性 PDRN",
+            "REJURAN 麗駐蘭修復舒緩面膜5P",
+            "focused_exact_identity_rejuran_repair_mask_5p",
+        ),
+        (
+            "【SHISEIDO 資生堂國際櫃】新艷陽•夏 水離子熱防禦UV隔離露 SPF50(防曬/防曬乳/清爽/不黏膩)",
+            "《SHISEIDO 資生堂》新艷陽夏水離子熱防禦隔離露 50ml",
+            "focused_exact_identity_shiseido_blue_sunscreen",
+        ),
+    ]
+
+    for momo_name, competitor_name, expected_reason in cases:
+        diagnostics = score_marketplace_match(momo_name, competitor_name)
+        assert diagnostics.score >= 0.76, (momo_name, diagnostics)
+        assert diagnostics.hard_veto is False
+        assert expected_reason in diagnostics.reasons
+
+
+def test_marketplace_matcher_keeps_high_variant_low_score_lines_outside_focused_promotion():
+    from services.marketplace_product_matcher import score_marketplace_match
+
+    lush = score_marketplace_match(
+        "【LUSH 嵐舒】Sticky Dates 椰棗布丁身體噴霧 200ml(香氛噴霧/安息香/檀木/甜蜜香氣)",
+        "英國原裝LUSH 睡公主身體噴霧200ml Sleepy Body Spray",
+    )
+    lactacyd = score_marketplace_match(
+        "【Lactacyd 立朵舒】清新舒涼 私密潔浴露250ml(天然薄荷涼感 私密清潔 私密保養 乾癢修護)",
+        "Lactacyd 立朵舒私密潔浴露-生理呵護250ml",
+    )
+
+    for diagnostics in (lush, lactacyd):
+        assert diagnostics.score < 0.76
+        assert not any(reason.startswith("focused_exact_identity_") for reason in diagnostics.reasons)
+
+
 def test_marketplace_matcher_rejects_refill_core_vs_case_only_pack():
     from services.marketplace_product_matcher import score_marketplace_match