diff --git a/qdrant_client/hybrid/formula.py b/qdrant_client/hybrid/formula.py
index 8365276ca..a0f1342a9 100644
--- a/qdrant_client/hybrid/formula.py
+++ b/qdrant_client/hybrid/formula.py
@@ -318,12 +318,15 @@ def parse_variable(var: str) -> str | int:
     if bracket_end == -1:
         raise ValueError(f"Invalid score pattern: {var}")
 
-    # try parsing the content in between brackets as integer
-    try:
-        idx = int(remaining[:bracket_end])
-    except ValueError:
+    # parse the content in between brackets as an unsigned integer. qdrant core represents
+    # the score index as a usize, while `int()` would also accept a sign, underscore
+    # separators and surrounding whitespace
+    raw_idx = remaining[:bracket_end]
+    if not (raw_idx.isascii() and raw_idx.isdigit()):
         raise ValueError(f"Invalid score pattern: {var}")
 
+    idx = int(raw_idx)
+
     # make sure the string ends after the closing bracket
     if len(remaining) > bracket_end + 1:
         raise ValueError(f"Invalid score pattern: {var}")
@@ -337,34 +340,3 @@ def raise_non_finite_error(expression: str) -> None:
 
 def is_number(value: Any) -> bool:
     return isinstance(value, (int, float)) and not isinstance(value, bool)
-
-
-def test_parsing_variable() -> None:
-    assert parse_variable("$score") == 0
-    assert parse_variable("$score[0]") == 0
-    assert parse_variable("$score[1]") == 1
-    assert parse_variable("$score[2]") == 2
-
-    try:
-        parse_variable("$score[invalid]")
-        assert False
-    except ValueError as e:
-        assert str(e) == "Invalid score pattern: $score[invalid]"
-
-    try:
-        parse_variable("$score[10].other")
-        assert False
-    except ValueError as e:
-        assert str(e) == "Invalid score pattern: $score[10].other"
-
-
-def test_try_extract_payload_value() -> None:
-    for payload_value, expected in [(1.2, 1.2), ([1.2], 1.2), ([1.2, 2.3], [1.2, 2.3])]:
-        empty_defaults: dict[str, Any] = {}
-
-        payload = {"key": payload_value}
-        assert try_extract_payload_value("key", payload, empty_defaults) == expected
-
-        defaults = {"key": payload_value}
-        empty_payload: dict[str, Any] = {}
-        assert try_extract_payload_value("key", empty_payload, defaults) == expected
diff --git a/qdrant_client/local/tests/test_formula.py b/qdrant_client/local/tests/test_formula.py
new file mode 100644
index 000000000..056ba0204
--- /dev/null
+++ b/qdrant_client/local/tests/test_formula.py
@@ -0,0 +1,62 @@
+import re
+from typing import Any
+
+import pytest
+
+from qdrant_client.hybrid.formula import (
+    evaluate_variable,
+    parse_variable,
+    try_extract_payload_value,
+)
+
+
+def test_parse_variable_score_index() -> None:
+    assert parse_variable("$score") == 0
+    assert parse_variable("$score[0]") == 0
+    assert parse_variable("$score[1]") == 1
+    assert parse_variable("$score[2]") == 2
+    assert parse_variable("$score[10]") == 10
+
+    # qdrant core resolves the index through the json path grammar into a usize, so anything
+    # that is not a plain run of ascii digits is not a valid score pattern
+    for var in (
+        "$score[-1]",
+        "$score[+1]",
+        "$score[1_0]",
+        "$score[ 1 ]",
+        "$score[²]",
+        "$score[invalid]",
+        "$score[]",
+        "$score[1][2]",
+        "$score[10].other",
+        "$score.invalid",
+    ):
+        with pytest.raises(ValueError, match=f"Invalid score pattern: {re.escape(var)}"):
+            parse_variable(var)
+
+
+def test_evaluate_variable_rejects_negative_score_index() -> None:
+    scores = [{1: 10.0}, {1: 20.0}, {1: 30.0}]
+
+    assert evaluate_variable("$score[0]", 1, scores, {}, {}) == 10.0
+    assert evaluate_variable("$score[2]", 1, scores, {}, {}) == 30.0
+    # an index past the end falls back to the default score
+    assert evaluate_variable("$score[3]", 1, scores, {}, {}) == 0.0
+
+    # a negative index must not wrap around to the last prefetch, nor leak an IndexError
+    with pytest.raises(ValueError):
+        evaluate_variable("$score[-1]", 1, scores, {}, {})
+    with pytest.raises(ValueError):
+        evaluate_variable("$score[-9]", 1, scores, {}, {})
+
+
+def test_try_extract_payload_value() -> None:
+    for payload_value, expected in [(1.2, 1.2), ([1.2], 1.2), ([1.2, 2.3], [1.2, 2.3])]:
+        empty_defaults: dict[str, Any] = {}
+
+        payload = {"key": payload_value}
+        assert try_extract_payload_value("key", payload, empty_defaults) == expected
+
+        defaults = {"key": payload_value}
+        empty_payload: dict[str, Any] = {}
+        assert try_extract_payload_value("key", empty_payload, defaults) == expected