test(gfql): L3 wave-3 nits — output-json touch after --est, decline-branch docstring scope

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: lmeyerov

benchmarks/gfql/viz_filter_pipeline.py

@@ -291,15 +291,16 @@ def main() -> None:
     args = parser.parse_args()
     if args.runs < 1 or args.warmup < 0:
         parser.error("--runs must be >= 1 and --warmup >= 0")
-    if args.output_json:
-        # fail BEFORE a multi-minute sweep, not at the final write (wave-2 S4)
-        with open(args.output_json, "a"):
-            pass
-
     if args.est:
         print_estimate(args.scale)
         return
 
+    if args.output_json:
+        # fail BEFORE a multi-minute sweep, not at the final write (wave-2 S4;
+        # after the --est early-return so estimate-only runs touch nothing)
+        with open(args.output_json, "a"):
+            pass
+
     engines = [e.strip() for e in args.engines.split(",") if e.strip()]
     unknown = [e for e in engines if e not in KNOWN_ENGINES]
     if unknown:

graphistry/tests/compute/gfql/test_viz_pipeline_conformance.py

@@ -480,9 +480,10 @@ def test_case_regex_unicode_trick_matrix(label, query, pinned, mirror):
 
 def test_categorical_searchany_decline_or_correct():
     """Categorical column through searchAny via explicit columns=: pandas may decline
-    clearly (honest NIE or GFQLValidationError); a non-pandas engine may only NIE or
-    match the pandas oracle exactly — a SILENT WRONG answer is not acceptable, and a
-    non-NIE crash class is not."""
+    clearly (honest NIE or GFQLValidationError). When pandas answers, a non-pandas
+    engine may only NIE or match the pandas oracle exactly — a SILENT WRONG answer is
+    not acceptable, and a non-NIE crash class is not. (When pandas itself declines
+    there is no oracle, so non-pandas engines are not checked.)"""
     g = _panel_graph()
     nd = _panel_nodes()
     q = "MATCH (n) WHERE searchAny(n, 'x', {columns: ['cat']}) RETURN n.id AS id"