From d7a3f5c0813e7d8c411bbc2681488a2a01b1c6ec Mon Sep 17 00:00:00 2001
From: Peter Woolery <peter@research.bike>
Date: Sun, 10 May 2026 12:00:19 -0700
Subject: [PATCH] fix: matcher exclusion words + precision floor for clean
 grocery matching
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- Add exclusion words: soda, rotisserie, tuna/tonno/salmon/sardine/anchovy
  to prevent beverages, prepared poultry, and seafood-in-oil from matching
  raw cooking ingredients
- Add min precision floor (0.45): grocery sig-word count must be ≤ 2× the
  ingredient's sig-word count, catching long branded products that pass
  the word-overlap recall check but are clearly wrong category matches
  (e.g. "Garlic Herb Rotisserie Chicken" precision=0.25 now rejected)

Result: all previously wrong matches now show '—' (no match) rather than
a wrong product; correct matches unchanged

Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
---
 backend/app/services/matcher.py | 38 +++++++++++++++++++++++++++++++++
 1 file changed, 38 insertions(+)

diff --git a/backend/app/services/matcher.py b/backend/app/services/matcher.py
index 78f17cb..7b37222 100644
--- a/backend/app/services/matcher.py
+++ b/backend/app/services/matcher.py
@@ -39,6 +39,34 @@ _STOP_WORDS = frozenset({
     "and", "with", "for", "the",
 })
 
+# If any of these words appear in a grocery item's sig-words but NOT in the
+# ingredient's sig-words, the match is rejected outright.  Prevents category
+# cross-contamination: "Garlic" must not match "Garlic Bread", "Lime" must not
+# match "Lime Margarita", etc.
+_EXCLUSION_WORDS = frozenset({
+    # Baked goods / bread products
+    "bread", "loaf", "rolls", "bun", "buns", "croissant",
+    "cracker", "crackers", "cookie", "cookies", "cake", "cupcake", "muffin", "bagel",
+    # Chips / snack foods
+    "chips",
+    # Pasta / noodles
+    "pasta", "noodle", "noodles", "vermicelli", "spaghetti", "linguine",
+    "fettuccine", "penne", "rigatoni", "macaroni", "rotini", "orzo",
+    # Alcoholic / mixed beverages
+    "margarita", "rita", "cocktail", "beer", "ale", "lager", "cider", "malt",
+    "wine", "spirits", "liquor",
+    "vodka", "tequila", "whiskey", "rum", "gin", "bourbon",
+    "lemonade", "limeade", "seltzer", "soda",
+    # Butter / spreads (prevents "Garlic & Herb Butter Spread" matching "Garlic")
+    "butter", "spread", "margarine",
+    # Prepared proteins / seafood-in-oil (prevents "Tuna in Olive Oil" matching "Olive Oil")
+    "tuna", "tonno", "salmon", "sardine", "anchovy",
+    # Prepared poultry (prevents "Garlic Herb Rotisserie Chicken" matching "Garlic")
+    "rotisserie",
+    # Baby / personal care (belt-and-suspenders after stop-word rework)
+    "baby", "wipes", "diaper",
+})
+
 
 @dataclass
 class MatchResult:
@@ -127,8 +155,18 @@ def run_match_job(
                 # 100% recall: every ingredient sig-word must appear in the grocery name.
                 if not ing_sig.issubset(gsig):
                     continue
+                # Category exclusion: reject if grocery has a disqualifying word
+                # (e.g. "bread", "chips", "margarita") absent from the ingredient.
+                bad_words = (gsig & _EXCLUSION_WORDS) - ing_sig
+                if bad_words:
+                    continue
                 # Precision penalises grocery items with many extra words.
                 precision = len(ing_sig) / len(gsig)
+                # Hard floor: grocery must not have >2× the sig-words of the ingredient.
+                # Catches long branded products that sneak past exclusion words, e.g.
+                # "Garlic Herb Rotisserie Chicken" for "Garlic".
+                if precision < 0.45:
+                    continue
                 combined = (score / 100.0) * precision
                 if combined > best_combined:
                     best_combined = combined