強化商品線配對與組合 veto

docs/AI_INTELLIGENCE_MODULE_SOT.md

@@ -342,6 +342,7 @@ LEFT JOIN competitor_prices cp
 - competitor PPT 不可只輸出 matched rows 造成覆蓋率假象；`fetch_competitor_comparison_results()` 必須用 `LEFT JOIN valid_competitor` 保留高營收/高價但尚未有效配對的 MOMO 商品，並帶出 `match_status`、`candidate_count`、`best_match_score` 與 `match_diagnostic`，讓簡報與 AI 文案明確區分「高信心比對」與「待補身份/價格」。
 - `services/competitor_identity_revalidator.py` 可對既有 `competitor_prices` legacy row 離線重跑 `identity_v2`：只有新版 matcher 分數 `>= 0.76` 且無 hard veto 才補 `identity_v2` / `legacy_revalidated` tags；預設不刷新 `expires_at`，避免過期價格進入決策。
 - `CompetitorPriceFeeder.run_expired_identity_refresh()` 會優先刷新已通過 `identity_v2` 但 TTL 過期的 PChome row：直接用既有 `competitor_product_id` 批次呼叫 PChome 商品 API，再用新版 matcher 重新驗證名稱/規格/價格 sanity，通過後寫回 `competitor_prices` 與 `competitor_price_history`。這條路徑提升新鮮價格覆蓋率，但不降低 match threshold，也不讓過期價格直接進入決策。
+- `marketplace_product_matcher.py` 的擴充只能走「正向證據 + 反向 veto」：品牌一致、商品線/型號訊號強、價格合理且無 hard veto 時才允許 `strong_product_line_match` 加分；補充瓶/補充包/refill 與一般正裝不互相配對，分享組/加量組/明星組等組合包不得誤配單品。
 - Dashboard 必須把「待比對」拆成可診斷狀態：`價格過期待刷新`、`舊版配對待重驗`、`低分配對待審`、`身份否決`、`找不到同款`、`抓取異常`、`尚未搜尋`。不可再用單一「待比對」掩蓋資料品質原因。
 
 ### 執行方式

services/marketplace_product_matcher.py

@@ -448,6 +448,9 @@ def _has_bundle_offer(identity: ProductIdentity) -> bool:
         or "優惠套組" in text
         or "禮盒組" in text
         or "加大組" in text
+        or "加量組" in text
+        or "分享組" in text
+        or "明星組" in text
         or "套組" in text
     )
 
@@ -461,6 +464,17 @@ def _has_multi_component(identity: ProductIdentity) -> bool:
     )
 
 
+def _has_refill_pack(identity: ProductIdentity) -> bool:
+    text = identity.normalized_name
+    return bool(
+        "補充瓶" in text
+        or "補充包" in text
+        or "替換蕊" in text
+        or "替換芯" in text
+        or "refill" in text
+    )
+
+
 def _spec_mention_count(identity: ProductIdentity) -> int:
     return len(re.findall(r"\d+(?:\.\d+)?\s*(?:ml|毫升|l|g|公克|kg)", identity.normalized_name, re.I))
 
@@ -488,6 +502,23 @@ def _chinese_bigram_score(left: ProductIdentity, right: ProductIdentity) -> floa
     return 2 * len(left_signature & right_signature) / (len(left_signature) + len(right_signature))
 
 
+def _has_strong_product_line_signal(
+    left: ProductIdentity,
+    right: ProductIdentity,
+    token_score: float,
+    chinese_name_score: float,
+) -> bool:
+    shared_core = left.core_tokens & right.core_tokens
+    shared_latin_or_model = {
+        token for token in shared_core
+        if re.fullmatch(r"[a-z][a-z0-9-]{3,}", token)
+        or re.fullmatch(r"[a-z]{2,}-?\d+[a-z0-9-]*", token)
+    }
+    if shared_latin_or_model and token_score >= 0.50:
+        return True
+    return token_score >= 0.56 and chinese_name_score >= 0.45
+
+
 def score_marketplace_match(
     momo_name: str,
     competitor_name: str,
@@ -525,6 +556,8 @@ def score_marketplace_match(
         reasons.append("bundle_offer_conflict")
     if _has_multi_component(left) != _has_multi_component(right):
         reasons.append("multi_component_conflict")
+    if _has_refill_pack(left) != _has_refill_pack(right):
+        reasons.append("refill_pack_conflict")
     left_spec_mentions = _spec_mention_count(left)
     right_spec_mentions = _spec_mention_count(right)
     if left_spec_mentions and right_spec_mentions and left_spec_mentions != right_spec_mentions:
@@ -537,6 +570,8 @@ def score_marketplace_match(
         hard_veto = True
     if _has_multi_component(left) != _has_multi_component(right):
         hard_veto = True
+    if _has_refill_pack(left) != _has_refill_pack(right):
+        hard_veto = True
     if left_spec_mentions and right_spec_mentions and left_spec_mentions != right_spec_mentions:
         hard_veto = True
     if chinese_name_score < 0.16 and token_score < 0.72:
@@ -568,6 +603,16 @@ def score_marketplace_match(
 
     if token_score >= 0.72 and spec_score >= 0.82 and not brand_conflict:
         score += 0.08
+    if (
+        brand_score >= 0.95
+        and not hard_veto
+        and price_penalty == 0
+        and type_score >= 0.55
+        and spec_score >= 0.55
+        and _has_strong_product_line_signal(left, right, token_score, chinese_name_score)
+    ):
+        score += 0.07
+        reasons.append("strong_product_line_match")
     if hard_veto:
         score = min(score, 0.32)
     score = max(0.0, min(1.0, score))

tests/test_marketplace_product_matcher.py

@@ -61,6 +61,52 @@ def test_marketplace_matcher_handles_bundle_piece_count():
     assert diagnostics.hard_veto is False
 
 
+def test_marketplace_matcher_accepts_strong_model_line_without_specs():
+    from services.marketplace_product_matcher import score_marketplace_match
+
+    diagnostics = score_marketplace_match(
+        "【Stadler Form】Sophie 無線香氛水氧機 水氧機 香氛機",
+        "【瑞士Stadler Form】無線香氛水氧機 露營燈造型 Sophie",
+        momo_price=3780,
+        competitor_price=3980,
+    )
+
+    assert diagnostics.score >= 0.76
+    assert diagnostics.hard_veto is False
+    assert "strong_product_line_match" in diagnostics.reasons
+
+
+def test_marketplace_matcher_rejects_bundle_to_single_even_when_brand_matches():
+    from services.marketplace_product_matcher import score_marketplace_match
+
+    diagnostics = score_marketplace_match(
+        "【NARS】小白餅閨蜜分享組(裸光蜜粉餅/定妝蜜粉)",
+        "【NARS】裸光蜜粉餅(小白餅) 10g",
+        momo_price=3300,
+        competitor_price=1099,
+    )
+
+    assert diagnostics.score < 0.76
+    assert diagnostics.hard_veto is True
+    assert "bundle_offer_conflict" in diagnostics.reasons
+
+
+def test_marketplace_matcher_does_not_promote_wide_price_refill_candidate():
+    from services.marketplace_product_matcher import score_marketplace_match
+
+    diagnostics = score_marketplace_match(
+        "【蘭蔻】官方直營 絕對完美永生玫瑰逆齡乳霜60ml補充瓶",
+        "LANCOME蘭蔻 絕對完美永生玫瑰逆齡乳霜 60ml",
+        momo_price=11205,
+        competitor_price=5349,
+    )
+
+    assert diagnostics.score < 0.76
+    assert diagnostics.hard_veto is True
+    assert "refill_pack_conflict" in diagnostics.reasons
+    assert "strong_product_line_match" not in diagnostics.reasons
+
+
 def test_batch_compare_top_uses_latest_momo_price_not_revenue(monkeypatch):
     from services import pchome_crawler