From f0fb084da4deb8cf3c0522e3af6be49beb5a1f5c Mon Sep 17 00:00:00 2001
From: danpark <dev@nexsler.com>
Date: Tue, 12 May 2026 02:13:56 -0400
Subject: [PATCH] Escape time-like tokens in FTS queries

---
 lancedb_store.py     |  7 ++++++-
 tests/test_search.py | 25 +++++++++++++++++++++++++
 2 files changed, 31 insertions(+), 1 deletion(-)

diff --git a/lancedb_store.py b/lancedb_store.py
index 7d9115b..56671b2 100644
--- a/lancedb_store.py
+++ b/lancedb_store.py
@@ -125,6 +125,11 @@ def _sql_escape(value: str) -> str:
         """Escape single quotes for safe use in SQL WHERE clauses."""
         return value.replace("'", "''")
 
+    @staticmethod
+    def _escape_fts_query(query: str) -> str:
+        """Escape natural-language tokens Tantivy misreads as field syntax."""
+        return re.sub(r"(?<!\\):", r"\\:", query)
+
     @staticmethod
     def _validate_identifier(key: str) -> None:
         """Raise ValueError if *key* is not a safe SQL identifier."""
@@ -637,7 +642,7 @@ def keyword_search(
             table = self._vs.table
         except TableNotFoundError:
             return []  # Table not created yet — legitimately empty
-        q = table.search(query, query_type="fts")
+        q = table.search(self._escape_fts_query(query), query_type="fts")
         if where:
             q = q.where(where, prefilter=True)
         rows = q.limit(top_k).to_list()
diff --git a/tests/test_search.py b/tests/test_search.py
index ca8c9a6..f68feff 100644
--- a/tests/test_search.py
+++ b/tests/test_search.py
@@ -656,6 +656,31 @@ def test_prefilter_combined_source_and_folder():
         assert result[0].doc_id == "b.pdf"
 
 
+def test_keyword_search_escapes_time_like_tokens():
+    """FTS keyword search should handle time-like natural-language tokens."""
+    with tempfile.TemporaryDirectory() as tmpdir:
+        vec = [1.0] + [0.0] * 767
+        nodes = [
+            _make_node(
+                "appt.md",
+                "c:0",
+                "Appointment at 10:00 for the unit showing",
+                vec,
+                source_name="comm_messages",
+                status="active",
+            ),
+        ]
+        store = _build_store_with_fts(tmpdir, nodes)
+
+        hits = store.keyword_search(
+            "10:00 showing",
+            top_k=10,
+            where="lower(metadata.source_name) = 'comm_messages'",
+        )
+
+        assert [hit.doc_id for hit in hits] == ["appt.md"]
+
+
 def test_prefilter_complex_filter_ast():
     """Complex filter AST should restrict results before search scoring."""
     with tempfile.TemporaryDirectory() as tmpdir: