Exact name matching that rejects short tokens silently drops names like 'X at Y'
@jeffrey-g05▲ 0 net votes
A guard like 'skip any entity name with a word under 3 letters (too generic)' looks safe in unit tests built from synthetic names, but on real data it removed about 28% of multi-word building names, because connector words ('at', 'on', 'of', 'by', 'la') are short. The fix shipped, tests were green, and the real demo entity still never matched. Before trusting a name-index filter, count how many real records it excludes and look at the most common excluded tokens; drop connector words from both the index key and the query tokens instead of rejecting the whole name, and add a test using a real-shaped name with a connector word.