ok

pipeline/utils/fuzzy_join.py

@@ -113,7 +113,7 @@ def fuzzy_join_on_postcode(
     ``_match_score``) null. This turns "price-paid LEFT JOIN epc" into a union
     that also surfaces EPC-only dwellings (an energy certificate but no Land
     Registry sale). It assumes ``right`` is already unique on its match key
-    (address + postcode) — as the deduped EPC frame is — so each unmatched right
+    (address + postcode), as the deduped EPC frame is, so each unmatched right
     row appears once.
     """
 
@@ -308,7 +308,7 @@ def _numbers_compatible(a: str, b: str) -> bool:
 
     Equality, not subset: subset logic let "188 GREAT NORTH WAY" absorb
     "FLAT 1 188 GREAT NORTH WAY" ({188} is a subset of {1, 188}), attaching a
-    single flat's EPC facts to the whole building — tens of thousands of
+    single flat's EPC facts to the whole building: tens of thousands of
     wrong-property matches. Likewise digit-only tokens made "8A" and "8B"
     both look like {8} and match each other (and plain "8"), and ungated
     letter flats let "FLAT D 39 X ST" cross-match "FLAT F 39 X ST" at ~96.
@@ -332,7 +332,7 @@ def _admissible_variants(
     """Variants of ``primary`` that are safe to score against the other side.
 
     A variant may only ADD or DROP whole tokens relative to the primary (one
-    word multiset must contain the other) — never substitute, so a register
+    word multiset must contain the other), never substitute, so a register
     row whose address lines disagree with the combined address can't smuggle
     in a different street. The number gate runs on the primary addresses
     only, so the added/dropped tokens must additionally carry no digits

pipeline/utils/normalize.py

@@ -12,7 +12,7 @@ import re
 import polars as pl
 
 # One character outside [a-z0-9 ]. Callers lowercase first; each offending
-# character becomes a single space (runs are NOT merged here — callers apply
+# character becomes a single space (runs are NOT merged here; callers apply
 # word-level rules and then collapse_whitespace).
 _NON_ALNUM_LOWER_RE = re.compile(r"[^a-z0-9 ]")