[bot] Merge master/f8f72561 into rel/dev

packages/gooddata-eval/src/gooddata_eval/core/agentic/alert_skill.py

@@ -78,7 +78,11 @@ def _check_trigger(expected: CatalogMetricAlert, actual_args: dict) -> bool:
     if expected.operator == "ANOMALY":
         return True
     exp_trigger = expected.trigger
-    act_trigger = actual_args.get("trigger", actual_args.get("triggerMode", "ALWAYS"))
+    # A missing OR explicit-null trigger means "unset" -> the product persists the
+    # default ALWAYS ("Every time"). `.get(k, default)` only returns the default when
+    # the key is ABSENT, but create_metric_alert serialises unset params as
+    # `trigger: null`, so chain with `or` to also cover the present-but-None case.
+    act_trigger = actual_args.get("trigger") or actual_args.get("triggerMode") or "ALWAYS"
     if exp_trigger in _ALWAYS_TRIGGER_VALUES:
         return act_trigger in {"ALWAYS", "Every time"}
     act_api = _TRIGGER_DISPLAY_TO_API.get(act_trigger, act_trigger)

packages/gooddata-eval/src/gooddata_eval/core/evaluators/alert_skill.py

@@ -78,7 +78,9 @@ def evaluate(self, item: DatasetItem, chat_result: ChatResult) -> ItemEvaluation
 
         if "Trigger" in expected:
             expected_trigger = _TRIGGER_MAP.get(expected["Trigger"], expected["Trigger"])
-            trigger_correct = args.get("trigger") == expected_trigger
+            # Omitted/null trigger persists as the product default ALWAYS ("Every time").
+            actual_trigger = args.get("trigger") or "ALWAYS"
+            trigger_correct = actual_trigger == expected_trigger
 
         if "Filters" in expected:
             actual_filters = args.get("filters") or []

packages/gooddata-eval/tests/test_agentic_alert_skill.py

@@ -4,7 +4,9 @@
 
 from gooddata_eval.core.agentic.alert_skill import (
     AlertEvaluation,
+    _check_trigger,
     _deep_subset,
+    _normalize_expected_output,
     _to_number,
     run_agentic_alert_skill,
 )
@@ -31,6 +33,24 @@ def test_deep_subset_missing_key():
     assert _deep_subset({"a": 1, "c": 3}, {"a": 1}) is False
 
 
+def test_check_trigger_missing_or_null_defaults_to_always():
+    # "Every time" is the product default -> the agent may omit the trigger arg
+    # entirely, or serialise it as null. Both must count as ALWAYS, not a mismatch.
+    expected = _normalize_expected_output({"Operator": "GREATER_THAN", "Trigger": "Every time"})
+    assert _check_trigger(expected, {"operator": "GREATER_THAN"}) is True  # key absent
+    assert _check_trigger(expected, {"trigger": None}) is True  # present-but-null (the bug)
+    assert _check_trigger(expected, {"trigger": "ALWAYS"}) is True
+
+
+def test_check_trigger_once_needs_explicit_once():
+    # A "One time" expectation must still require an explicit ONCE - the null-default
+    # fix must not turn a wrong/absent trigger into a pass here.
+    expected = _normalize_expected_output({"Operator": "LESS_THAN", "Trigger": "One time"})
+    assert _check_trigger(expected, {"trigger": "ONCE"}) is True
+    assert _check_trigger(expected, {"trigger": None}) is False  # null != ONCE
+    assert _check_trigger(expected, {"trigger": "ONCE_PER_INTERVAL"}) is False  # real model error stays a fail
+
+
 def test_alert_evaluation_strict_pass():
     ev = AlertEvaluation(
         alert_created=True,

packages/gooddata-eval/tests/test_alert_skill_evaluator.py

@@ -58,6 +58,23 @@ def test_alert_evaluator_skips_absent_field():
     assert result.passed is True
 
 
+def test_alert_evaluator_defaults_missing_trigger_to_always():
+    # The agent may omit `trigger` (or send null) -> backend default is ALWAYS,
+    # so an "Every time" expectation must still pass instead of a false-negative.
+    expected = {"Operator": "GREATER_THAN", "Threshold": "150", "Trigger": "Every time"}
+    ev = get_evaluator("alert_skill")
+
+    absent = ev.evaluate(_item(expected), _chat_with_alert({"operator": "GREATER_THAN", "threshold": 150}))
+    assert absent.passed is True
+    assert absent.detail["trigger_correct"] is True
+
+    null = ev.evaluate(
+        _item(expected), _chat_with_alert({"operator": "GREATER_THAN", "threshold": 150, "trigger": None})
+    )
+    assert null.passed is True
+    assert null.detail["trigger_correct"] is True
+
+
 def test_alert_evaluator_fails_when_no_tool_call():
     result = get_evaluator("alert_skill").evaluate(
         _item({"Operator": "LESS_THAN"}), ChatResult.model_validate({"textResponse": "here is the alert"})