diff --git a/tests/perf/query_cases/sst_float_bss/case.toml b/tests/perf/query_cases/sst_float_bss/case.toml
index cc2e5fa96ec..272e0eb4bc4 100644
--- a/tests/perf/query_cases/sst_float_bss/case.toml
+++ b/tests/perf/query_cases/sst_float_bss/case.toml
@@ -21,10 +21,10 @@ chunk_series_count = 128
 timeout_seconds = 600
 visibility_timeout_seconds = 300
 
-# Keep the physical-table definition identical except for float encoding. The
-# high shared TWCS trigger avoids compaction changing the three-flush layout.
+# Omit the default encoding from base setup because older binaries do not recognize it.
+# The high shared TWCS trigger avoids compaction changing the three-flush layout.
 base_setup_sql = [
-  "CREATE TABLE sst_float_bss_physical (greptime_timestamp TIMESTAMP TIME INDEX, greptime_value DOUBLE) ENGINE=metric WITH ('physical_metric_table'='', 'experimental_sst_float_field_encoding'='default', 'compaction.twcs.trigger_file_num'='100')",
+  "CREATE TABLE sst_float_bss_physical (greptime_timestamp TIMESTAMP TIME INDEX, greptime_value DOUBLE) ENGINE=metric WITH ('physical_metric_table'='', 'compaction.twcs.trigger_file_num'='100')",
 ]
 candidate_setup_sql = [
   "CREATE TABLE sst_float_bss_physical (greptime_timestamp TIMESTAMP TIME INDEX, greptime_value DOUBLE) ENGINE=metric WITH ('physical_metric_table'='', 'experimental_sst_float_field_encoding'='byte_stream_split', 'compaction.twcs.trigger_file_num'='100')",
diff --git a/tests/perf/test_query_regression_case_selection.py b/tests/perf/test_query_regression_case_selection.py
index 58cefa4395b..4000a5da141 100644
--- a/tests/perf/test_query_regression_case_selection.py
+++ b/tests/perf/test_query_regression_case_selection.py
@@ -16,6 +16,7 @@
 """Coverage for query-regression case group selection."""
 
 import importlib.util
+import re
 import sys
 import unittest
 from pathlib import Path
@@ -54,6 +55,23 @@ class QueryRegressionCaseSelectionTest(unittest.TestCase):
         case = "tests/perf/query_cases/sql_topk_order_by/case.toml"
         self.assertEqual(runner.split_cases([case]), [case])
 
+    def test_sst_float_bss_omits_base_encoding_for_older_binaries(self) -> None:
+        case_path = Path(__file__).parent / "query_cases/sst_float_bss/case.toml"
+        setup_sql = dict(re.findall(
+            r'(base|candidate)_setup_sql = \[\s*"([^"\n]+)"',
+            case_path.read_text(),
+        ))
+        base_setup = setup_sql["base"]
+        candidate_setup = setup_sql["candidate"]
+        candidate_option = "'experimental_sst_float_field_encoding'='byte_stream_split'"
+
+        self.assertNotIn("experimental_sst_float_field_encoding", base_setup)
+        self.assertIn(candidate_option, candidate_setup)
+        self.assertEqual(
+            base_setup,
+            candidate_setup.replace(f", {candidate_option}", ""),
+        )
+
 
 if __name__ == "__main__":
     unittest.main()