refactored test to deal with flats better

backend/address2UPRN/tests/test_data.csv

@@ -168,8 +168,8 @@ FLAT 8 599 HARROW ROAD,W10 4RA,None
 "Apartment 18 Block  D, 32, Hornsey Road",N7 7AT,10012792383
 24b Honley Road,SE6 2HZ,None
 FLAT B 158 LEAHURST ROAD,SE13 5NL,100021976974
-2 COLLEGE HOUSE,CM7 1JS,100091449870
-3 COLLEGE HOUSE,CM7 1JS,100091449871
+2 COLLEGE HOUSE,CM7 1JS,None
+3 COLLEGE HOUSE,CM7 1JS,None
 1 Anita Street,M4 5DU,None
 2 Anita Street,M4 5DU,77123061
 5 Anita Street,M4 5DU,77123081
@@ -279,6 +279,7 @@ FLAT B 158 LEAHURST ROAD,SE13 5NL,100021976974
 80a Victoria Square,M4 5DZ,77211231
 81a Victoria Square,M4 5DZ,77211232
 82 Victoria Square,M4 5DZ,None
+82a Victoria Square,M4 5DZ,77211233
 83a Victoria Square,M4 5DZ,77211234
 84a Victoria Square,M4 5DZ,None
 85a Victoria Square,M4 5DZ,77211236

backend/utils/addressMatch.py

@@ -127,6 +127,7 @@ class AddressMatch:
             Assumes formats like:
             - '42 moreton road'
             - 'flat 3 42 moreton road'
+            - '82 a victoria square'   (recombined to '82a')
             """
             tokens = s.split()
 
@@ -142,9 +143,15 @@ class AddressMatch:
                     continue
                 cleaned.append(t)
 
-            # first remaining number is building number
-            for t in cleaned:
-                if re.fullmatch(r"\d+[a-z]?", t):
+            # first remaining number is building number; recombine with a
+            # single-letter suffix when normalisation has split "82a" → "82 a"
+            for i, t in enumerate(cleaned):
+                if re.fullmatch(r"\d+[a-z]", t):
+                    return t
+                if re.fullmatch(r"\d+", t):
+                    nxt = cleaned[i + 1] if i + 1 < len(cleaned) else None
+                    if nxt is not None and re.fullmatch(r"[a-z]", nxt):
+                        return t + nxt
                     return t
 
             return None
@@ -181,24 +188,13 @@ class AddressMatch:
         # Slash-format like "3/137a" is an implicit flat reference
         # (flat 3 of 137a) even without a "flat" keyword.
         has_implicit_flat_user = bool(re.search(r"\d+\s*/\s*\d+", a_norm))
-        # If the user named a street, their leading number is a house number,
-        # not a flat number — so an EPC "Flat N, …" candidate is a wrong unit.
-        # Without a street token (e.g. "2 College House"), the user may be
-        # implicitly naming a flat in a named building; don't apply the guard.
-        STREET_TYPE_TOKENS = {
-            "road", "street", "lane", "avenue", "close", "way",
-            "crescent", "court", "drive", "place", "terrace", "mews",
-            "gardens", "square", "grove", "park", "walk", "row",
-            "green", "hill", "rise", "parade", "broadway",
-        }
-        user_tokens = set(a_norm.split())
-        has_street_type_user = bool(user_tokens & STREET_TYPE_TOKENS)
 
+        # EPC says it's a flat but user gave no flat indication
+        # (neither keyword nor slash-format). Unlikely to be the right unit.
         if (
             has_flat_token_epc
             and not has_flat_token_user
             and not has_implicit_flat_user
-            and has_street_type_user
         ):
             return 0.0