導入 browse.sh 比價診斷計畫

services/competitor_price_feeder.py

@@ -31,6 +31,7 @@ import time
 from dataclasses import dataclass
 from datetime import datetime, timedelta, timezone
 from typing import Optional
+from urllib.parse import quote_plus
 
 logger = logging.getLogger(__name__)
 
@@ -47,6 +48,11 @@ TTL_HOURS        = 6     # competitor_prices 快取有效期
 REQUEST_TIMEOUT  = float(os.getenv("PCHOME_FEEDER_TIMEOUT", "12"))  # 避免外部搜尋 API 長時間卡住排程
 VARIANT_RECALL_SORTS = ("sale/dc", "new/dc")
 RECOVERABLE_LOW_SCORE_FLOOR = max(MIN_MATCH_SCORE - 0.03, 0.72)
+BROWSE_SH_DIAGNOSTIC_ENABLED = os.getenv("PCHOME_FEEDER_BROWSE_SH_DIAGNOSTIC_ENABLED", "true").lower() in {"1", "true", "yes", "on"}
+BROWSE_SH_EXECUTE_ENABLED = os.getenv("PCHOME_FEEDER_BROWSE_SH_EXECUTE_ENABLED", "false").lower() in {"1", "true", "yes", "on"}
+BROWSE_SH_TIMEOUT_SECONDS = int(os.getenv("PCHOME_FEEDER_BROWSE_SH_TIMEOUT", "20"))
+BROWSE_SH_MAX_EXECUTIONS_PER_RUN = int(os.getenv("PCHOME_FEEDER_BROWSE_SH_MAX_PER_RUN", "3"))
+BROWSE_SH_OUTPUT_PREVIEW_CHARS = int(os.getenv("PCHOME_FEEDER_BROWSE_SH_OUTPUT_PREVIEW_CHARS", "1200"))
 RECOVERABLE_DIAGNOSTIC_REASONS = {
     "strong_product_line_match",
     "strong_exact_spec_match",
@@ -95,6 +101,43 @@ def _classify_low_score_attempt(score: float, diagnostics) -> str:
     return "true_low_confidence"
 
 
+def _has_variant_selection_gap(
+    momo_name: str,
+    ranked_matches: list[tuple],
+    best_score: float,
+) -> bool:
+    """True when source lacks explicit variant selection but top candidates require one."""
+    try:
+        from services.marketplace_product_matcher import (
+            _explicit_variant_option_tokens,
+            parse_product_identity,
+        )
+    except Exception:
+        return False
+
+    source_identity = parse_product_identity(momo_name)
+    source_options = set(_explicit_variant_option_tokens(source_identity))
+    if re.search(r"任選\s*[一二兩三四五六七八九十0-9]+\s*款", momo_name):
+        source_options -= {str(value) for value in range(1, 11)}
+        source_options -= {f"{value:02d}" for value in range(1, 11)}
+    if source_options:
+        return False
+
+    threshold = max(best_score - 0.02, RECOVERABLE_LOW_SCORE_FLOOR)
+    option_buckets: set[str] = set()
+    for product, score, diagnostics in ranked_matches[:5]:
+        if getattr(diagnostics, "hard_veto", False) or score < threshold:
+            continue
+        candidate_identity = parse_product_identity(getattr(product, "name", "") or "")
+        options = _explicit_variant_option_tokens(candidate_identity)
+        if len(options) >= 2:
+            return True
+        option_buckets.update(options)
+        if len(option_buckets) >= 2:
+            return True
+    return False
+
+
 def _extract_tags(pchome_product) -> list:
     """
     從 PChomeProduct 物件提取語意標籤
@@ -286,6 +329,66 @@ def _match_diagnostics_payload(diagnostics) -> dict:
     }
 
 
+def _pchome_search_url(keyword: str) -> str:
+    return f"https://ecshweb.pchome.com.tw/search/v3.3/?q={quote_plus(keyword or '')}"
+
+
+def _build_browse_sh_diagnostic_payload(
+    momo_name: str,
+    search_terms: list[str] = None,
+    reason: str = "unknown",
+    best_product=None,
+    best_score: float = None,
+    diagnostics=None,
+    candidate_count: int = 0,
+) -> dict:
+    """Build a read-only browse.sh probe plan for low-confidence PChome cases."""
+    if not BROWSE_SH_DIAGNOSTIC_ENABLED:
+        return {}
+
+    terms = _dedupe_terms(search_terms or _build_search_keywords(momo_name))[:3]
+    urls = [_pchome_search_url(term) for term in terms]
+    product_url = getattr(best_product, "product_url", None)
+    if product_url:
+        urls.append(product_url)
+    urls = list(dict.fromkeys(url for url in urls if url))
+    primary_url = urls[0] if urls else _pchome_search_url(momo_name)
+
+    diagnostic_payload = _match_diagnostics_payload(diagnostics)
+    return {
+        "tool": "browse.sh",
+        "mode": "execute_on_demand" if BROWSE_SH_EXECUTE_ENABLED else "plan_only",
+        "reason": reason,
+        "execute_enabled": BROWSE_SH_EXECUTE_ENABLED,
+        "timeout_seconds": BROWSE_SH_TIMEOUT_SECONDS,
+        "candidate_count": int(candidate_count or 0),
+        "momo_name": (momo_name or "")[:300],
+        "search_terms": terms,
+        "urls": urls,
+        "suggested_commands": [
+            {
+                "purpose": "static_fetch_first_page",
+                "args": ["get", primary_url],
+            },
+            {
+                "purpose": "manual_browser_probe",
+                "args": ["open", primary_url],
+            },
+        ],
+        "best_candidate": {
+            "product_id": getattr(best_product, "product_id", None),
+            "name": (getattr(best_product, "name", None) or "")[:300] or None,
+            "price": getattr(best_product, "price", None),
+            "url": product_url,
+            "score": best_score,
+        } if best_product else None,
+        "diagnostic_codes": diagnostic_payload.get("reasons") or [],
+        "comparison_mode": diagnostic_payload.get("comparison_mode"),
+        "hard_veto": diagnostic_payload.get("hard_veto"),
+        "execution": {"status": "disabled"},
+    }
+
+
 def _product_snapshot_payload(product) -> dict:
     payload = {
         "competitor_product_url": None,
@@ -471,6 +574,7 @@ class CompetitorPriceFeeder:
         self._history_table_ready = False
         self._attempt_table_ready = False
         self._price_table_columns_ready = False
+        self._browse_sh_executions = 0
 
     def _ensure_table_columns(self, conn, table: str, column_specs: list[tuple[str, str]]) -> None:
         """補齊既有表欄位；避免正式端舊表在新 INSERT 時炸掉。"""
@@ -613,6 +717,7 @@ class CompetitorPriceFeeder:
                     comparison_mode VARCHAR(40),
                     hard_veto BOOLEAN,
                     diagnostic_codes JSONB,
+                    browse_diagnostic_json JSONB,
                     error_message TEXT,
                     attempted_at TIMESTAMP NOT NULL DEFAULT NOW()
                 )
@@ -648,6 +753,7 @@ class CompetitorPriceFeeder:
                     comparison_mode VARCHAR(40),
                     hard_veto BOOLEAN,
                     diagnostic_codes TEXT,
+                    browse_diagnostic_json TEXT,
                     error_message TEXT,
                     attempted_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP
                 )
@@ -669,9 +775,64 @@ class CompetitorPriceFeeder:
             ("comparison_mode", "VARCHAR(40)"),
             ("hard_veto", "BOOLEAN"),
             ("diagnostic_codes", "JSONB" if conn.dialect.name == "postgresql" else "TEXT"),
+            ("browse_diagnostic_json", "JSONB" if conn.dialect.name == "postgresql" else "TEXT"),
         ])
         self._attempt_table_ready = True
 
+    def _prepare_browse_diagnostic(
+        self,
+        momo_name: str,
+        search_terms: list = None,
+        reason: str = "unknown",
+        best_product=None,
+        best_score: float = None,
+        diagnostics=None,
+        candidate_count: int = 0,
+    ) -> dict:
+        """Return browse.sh diagnostic evidence; CLI execution remains opt-in and rate-limited."""
+        payload = _build_browse_sh_diagnostic_payload(
+            momo_name,
+            search_terms=search_terms,
+            reason=reason,
+            best_product=best_product,
+            best_score=best_score,
+            diagnostics=diagnostics,
+            candidate_count=candidate_count,
+        )
+        if not payload or not BROWSE_SH_EXECUTE_ENABLED:
+            return payload
+        if self._browse_sh_executions >= BROWSE_SH_MAX_EXECUTIONS_PER_RUN:
+            payload["execution"] = {"status": "rate_limited"}
+            return payload
+
+        command_args = tuple((payload.get("suggested_commands") or [{}])[0].get("args") or ())
+        if not command_args:
+            payload["execution"] = {"status": "missing_command"}
+            return payload
+
+        try:
+            from services.browse_sh_tool import BrowseShTool
+
+            self._browse_sh_executions += 1
+            result = BrowseShTool(timeout_seconds=BROWSE_SH_TIMEOUT_SECONDS).run(
+                command_args,
+                timeout_seconds=BROWSE_SH_TIMEOUT_SECONDS,
+            )
+            payload["execution"] = {
+                "status": "ok" if result.ok else "failed",
+                "returncode": result.returncode,
+                "timed_out": result.timed_out,
+                "unavailable_reason": result.unavailable_reason,
+                "stdout_preview": (result.stdout or "")[:BROWSE_SH_OUTPUT_PREVIEW_CHARS],
+                "stderr_preview": (result.stderr or "")[:BROWSE_SH_OUTPUT_PREVIEW_CHARS],
+            }
+        except Exception as exc:
+            payload["execution"] = {
+                "status": "error",
+                "error": str(exc)[:500],
+            }
+        return payload
+
     def _record_match_attempt(
         self,
         sku: str,
@@ -684,6 +845,7 @@ class CompetitorPriceFeeder:
         best_product=None,
         best_score: float = None,
         diagnostics=None,
+        browse_diagnostic: dict = None,
         error_message: str = None,
         source: str = "pchome",
     ) -> None:
@@ -695,9 +857,15 @@ class CompetitorPriceFeeder:
             search_terms_expr = "CAST(:search_terms AS jsonb)" if conn.dialect.name == "postgresql" else ":search_terms"
             json_cast = "CAST(:match_diagnostic_json AS jsonb)" if conn.dialect.name == "postgresql" else ":match_diagnostic_json"
             codes_cast = "CAST(:diagnostic_codes AS jsonb)" if conn.dialect.name == "postgresql" else ":diagnostic_codes"
+            browse_cast = "CAST(:browse_diagnostic_json AS jsonb)" if conn.dialect.name == "postgresql" else ":browse_diagnostic_json"
             diagnostic_payload = _match_diagnostics_payload(diagnostics)
             diagnostic_codes = diagnostic_payload.get("reasons") or []
             product_payload = _product_snapshot_payload(best_product)
+            browse_diagnostic_json = (
+                json.dumps(browse_diagnostic, ensure_ascii=False)
+                if browse_diagnostic
+                else None
+            )
             conn.execute(text(f"""
                 INSERT INTO competitor_match_attempts
                     (sku, source, momo_product_id, momo_product_name, momo_price,
@@ -706,6 +874,7 @@ class CompetitorPriceFeeder:
                      competitor_product_url, competitor_image_url, competitor_stock,
                      best_competitor_price, best_match_score,
                      match_diagnostic_json, comparison_mode, hard_veto, diagnostic_codes,
+                     browse_diagnostic_json,
                      error_message,
                      attempted_at)
                 VALUES
@@ -715,6 +884,7 @@ class CompetitorPriceFeeder:
                      :competitor_product_url, :competitor_image_url, :competitor_stock,
                      :best_price, :best_score,
                      {json_cast}, :comparison_mode, :hard_veto, {codes_cast},
+                     {browse_cast},
                      :error_message,
                      CURRENT_TIMESTAMP)
             """), {
@@ -735,6 +905,7 @@ class CompetitorPriceFeeder:
                 "comparison_mode": diagnostic_payload.get("comparison_mode"),
                 "hard_veto": diagnostic_payload.get("hard_veto"),
                 "diagnostic_codes": json.dumps(diagnostic_codes, ensure_ascii=False) if diagnostic_codes else None,
+                "browse_diagnostic_json": browse_diagnostic_json,
                 "error_message": (error_message or "")[:1000] or None,
             })
 
@@ -1197,6 +1368,12 @@ class CompetitorPriceFeeder:
                 products = _search_pchome_candidates(crawler, momo_name, search_terms, momo_price=momo_price)
                 if not products:
                     logger.debug(f"[Feeder] {sku} 無搜尋結果，跳過")
+                    browse_diagnostic = self._prepare_browse_diagnostic(
+                        momo_name,
+                        search_terms=search_terms,
+                        reason="no_result",
+                        candidate_count=0,
+                    )
                     self._record_match_attempt(
                         sku,
                         momo_name,
@@ -1205,6 +1382,7 @@ class CompetitorPriceFeeder:
                         search_terms=search_terms,
                         candidate_count=0,
                         attempt_status="no_result",
+                        browse_diagnostic=browse_diagnostic,
                         source=source,
                     )
                     attempts_written += 1
@@ -1213,6 +1391,12 @@ class CompetitorPriceFeeder:
 
                 ranked_matches = _rank_match_details(momo_name, products, momo_price=momo_price)
                 if not ranked_matches:
+                    browse_diagnostic = self._prepare_browse_diagnostic(
+                        momo_name,
+                        search_terms=search_terms,
+                        reason="no_match",
+                        candidate_count=len(products),
+                    )
                     self._record_match_attempt(
                         sku,
                         momo_name,
@@ -1221,6 +1405,7 @@ class CompetitorPriceFeeder:
                         search_terms=search_terms,
                         candidate_count=len(products),
                         attempt_status="no_match",
+                        browse_diagnostic=browse_diagnostic,
                         source=source,
                     )
                     attempts_written += 1
@@ -1305,6 +1490,15 @@ class CompetitorPriceFeeder:
                         f"[Feeder] {sku} 候選屬單位價可比但非同販售組合，"
                         f"不寫入正式價差 | {_format_match_diagnostics(diagnostics)}"
                     )
+                    browse_diagnostic = self._prepare_browse_diagnostic(
+                        momo_name,
+                        search_terms=search_terms,
+                        reason="unit_comparable",
+                        best_product=best_product,
+                        best_score=score,
+                        diagnostics=diagnostics,
+                        candidate_count=len(products),
+                    )
                     self._record_match_attempt(
                         sku,
                         momo_name,
@@ -1316,6 +1510,7 @@ class CompetitorPriceFeeder:
                         best_product=best_product,
                         best_score=score,
                         diagnostics=diagnostics,
+                        browse_diagnostic=browse_diagnostic,
                         error_message=_format_match_diagnostics(diagnostics),
                         source=source,
                     )
@@ -1325,10 +1520,24 @@ class CompetitorPriceFeeder:
 
                 if score < MIN_MATCH_SCORE and not manual_accept_override:
                     attempt_status = _classify_low_score_attempt(score, diagnostics)
+                    if (
+                        attempt_status == "recoverable_low_score"
+                        and _has_variant_selection_gap(momo_name, ranked_matches, score)
+                    ):
+                        attempt_status = "true_low_confidence"
                     logger.debug(
                         f"[Feeder] {sku} 比對分數過低 ({score:.3f} < {MIN_MATCH_SCORE})，"
                         f"{_format_match_diagnostics(diagnostics)}"
                     )
+                    browse_diagnostic = self._prepare_browse_diagnostic(
+                        momo_name,
+                        search_terms=search_terms,
+                        reason=attempt_status,
+                        best_product=best_product,
+                        best_score=score,
+                        diagnostics=diagnostics,
+                        candidate_count=len(products),
+                    )
                     self._record_match_attempt(
                         sku,
                         momo_name,
@@ -1340,6 +1549,7 @@ class CompetitorPriceFeeder:
                         best_product=best_product,
                         best_score=score,
                         diagnostics=diagnostics,
+                        browse_diagnostic=browse_diagnostic,
                         error_message=_format_match_diagnostics(diagnostics),
                         source=source,
                     )
@@ -1365,6 +1575,15 @@ class CompetitorPriceFeeder:
                     write_reason = "manual_accept_override"
                 if not should_write:
                     logger.info(f"[Feeder] {sku} 進入人工覆核，不覆蓋既有配對 | {write_reason}")
+                    browse_diagnostic = self._prepare_browse_diagnostic(
+                        momo_name,
+                        search_terms=search_terms,
+                        reason="protected_existing_match",
+                        best_product=best_product,
+                        best_score=score,
+                        diagnostics=diagnostics,
+                        candidate_count=len(products),
+                    )
                     self._record_match_attempt(
                         sku,
                         momo_name,
@@ -1376,6 +1595,7 @@ class CompetitorPriceFeeder:
                         best_product=best_product,
                         best_score=score,
                         diagnostics=diagnostics,
+                        browse_diagnostic=browse_diagnostic,
                         error_message=f"{write_reason}; {_format_match_diagnostics(diagnostics)}",
                         source=source,
                     )
@@ -1418,6 +1638,12 @@ class CompetitorPriceFeeder:
             except Exception as e:
                 logger.error(f"[Feeder] {sku} 處理失敗: {e}")
                 try:
+                    browse_diagnostic = self._prepare_browse_diagnostic(
+                        momo_name,
+                        search_terms=search_terms,
+                        reason="crawler_error",
+                        candidate_count=0,
+                    )
                     self._record_match_attempt(
                         sku,
                         momo_name,
@@ -1425,6 +1651,7 @@ class CompetitorPriceFeeder:
                         momo_price=momo_price,
                         search_terms=search_terms,
                         attempt_status="error",
+                        browse_diagnostic=browse_diagnostic,
                         error_message=str(e),
                         source=source,
                     )

services/marketplace_product_matcher.py

@@ -520,6 +520,7 @@ BRAND_ALIAS_OVERRIDES = {
     "xiaomi": ("小米有品", "小米", "xiaomi"),
     "mac": ("m.a.c", "mac", "m a c"),
     "opi": ("o.p.i", "opi", "o p i"),
+    "st雞仔牌": ("日本雞仔牌st", "日本st雞仔牌", "st雞仔牌", "雞仔牌st", "雞仔牌"),
 }
 
 PRODUCT_TYPES = {
@@ -1157,12 +1158,25 @@ def _has_refill_pack(identity: ProductIdentity) -> bool:
     return bool(
         "補充瓶" in text
         or "補充包" in text
+        or "補充芯" in text
+        or "補充蕊" in text
         or "替換蕊" in text
         or "替換芯" in text
         or "refill" in text
     )
 
 
+def _has_accessory_case(identity: ProductIdentity) -> bool:
+    text = identity.normalized_name
+    return bool(
+        "眉彩餅盒" in text
+        or "盒一入款" in text
+        or "盒三入款" in text
+        or "盒單入" in text
+        or "空盒" in text
+    )
+
+
 def _spec_mention_count(identity: ProductIdentity) -> int:
     return len(
         re.findall(
@@ -1461,6 +1475,7 @@ def _build_evidence_flags(
         "count_conflict",
         "bundle_offer_conflict",
         "multi_component_conflict",
+        "accessory_case_conflict",
         "refill_pack_conflict",
         "price_ratio_extreme",
         "price_ratio_wide",
@@ -1557,6 +1572,9 @@ def score_marketplace_match(
         reasons.append("multi_component_conflict")
     if _has_refill_pack(left) != _has_refill_pack(right):
         reasons.append("refill_pack_conflict")
+    accessory_case_conflict = _has_accessory_case(left) != _has_accessory_case(right)
+    if accessory_case_conflict:
+        reasons.append("accessory_case_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:
@@ -1579,6 +1597,8 @@ def score_marketplace_match(
         hard_veto = True
     if _has_refill_pack(left) != _has_refill_pack(right):
         hard_veto = True
+    if accessory_case_conflict:
+        hard_veto = True
     if model_line_conflict:
         hard_veto = True
     if left_spec_mentions and right_spec_mentions and left_spec_mentions != right_spec_mentions:
@@ -1752,6 +1772,20 @@ def score_marketplace_match(
     ):
         score += 0.07
         reasons.append("shared_identity_anchor_exact_line")
+    if (
+        "無印乾爽止汗爽身乳液" in shared_anchor
+        and {"nivea", "妮維雅"} & (left.brand_tokens | right.brand_tokens)
+        and brand_score >= 0.95
+        and not hard_veto
+        and price_penalty == 0
+        and type_score >= 0.95
+        and spec_score >= 0.45
+        and token_score >= 0.55
+        and sequence_score >= 0.62
+        and not variant_descriptor_conflict
+    ):
+        score += 0.08
+        reasons.append("shared_identity_anchor_nivea_dry_lotion")
     if (
         "多效提亮防曬霜" in shared_anchor
         and {"recipe", "box"} <= (left.brand_tokens | right.brand_tokens)
@@ -1967,6 +2001,10 @@ def _extract_anchor_phrases(token: str) -> list[str]:
     phrases: list[str] = []
     if "經典旋轉眉筆" in cleaned:
         phrases.append("經典旋轉眉筆")
+    if "無印乾爽" in cleaned and "止汗爽身乳液" in cleaned:
+        phrases.append("無印乾爽止汗爽身乳液")
+    if "智能光感應" in cleaned and "無線自動除臭芳香噴霧機" in cleaned:
+        phrases.append("智能光感應無線自動除臭芳香噴霧機")
     if "悠斯晶" in normalized and "經典乳霜" in normalized:
         phrases.append("悠斯晶經典乳霜")
     if "經典乳霜" in normalized: