chore(gooddata-eval): use neutral identifiers in fixtures, drop a hardcoded path

packages/gooddata-eval/scripts/verify_guardrail_refusal_criteria.py

@@ -39,7 +39,9 @@
 
 from dotenv import load_dotenv
 
-load_dotenv("/Users/petertomko/gdc-mic-ai-evaluation/.env")
+# Whatever .env the caller points at, defaulting to the working directory. It used to be an
+# absolute path, which made the script runnable on exactly one machine.
+load_dotenv(os.environ.get("GD_EVAL_ENV_FILE", ".env"))
 
 from gooddata_eval.core.agentic.guardrail import _GUARDRAIL_EVALUATION_STEPS  # noqa: E402
 from gooddata_eval.core.evaluators._guardrail_criteria import (  # noqa: E402

packages/gooddata-eval/src/gooddata_eval/core/models.py

@@ -134,8 +134,8 @@ class ReasoningStepEvent(BaseModel):
 
 # Reasoning summaries are full paragraphs, e.g. "**Identifying analytics needs**\n\nI'm
 # analyzing..." -- using the whole thing as a latency_breakdown label would make every
-# entry an unreadable wall of text. Same bolded-title convention this repo's own reasoning
-# tooling already keys off of (see gdc-mic-ai-evaluation's generate_dashboard_summary.py).
+# entry an unreadable wall of text. The bolded title is the summary's own heading, and
+# downstream reporting keys off it for the same reason.
 _REASONING_TITLE_RE = re.compile(r"^\*\*(.+?)\*\*")
 _REASONING_LABEL_MAX_LEN = 60
 

packages/gooddata-eval/tests/test_from_insights.py

@@ -382,15 +382,15 @@ def test_built_envelope_is_loadable_as_a_dataset_item(tmp_path):
             ],
         ),
     )
-    envelope = build(spec, "How did spend trend by month?", "micai_diagnose_master", set())
+    envelope = build(spec, "How did spend trend by month?", "demo_workspace", set())
     assert "_shape" not in envelope["expected_output"]["visualization"]
     assert _validation_errors(envelope) is None
 
     (tmp_path / f"{envelope['id']}.json").write_text(json.dumps(envelope, indent=2))
     items = load_local_dataset(tmp_path)
     assert [i.id for i in items] == [envelope["id"]]
     assert items[0].test_kind == "visualization"
-    assert items[0].dataset_name == "micai_diagnose_master"
+    assert items[0].dataset_name == "demo_workspace"
 
 
 def test_mint_id_is_stable_and_collision_safe():

packages/gooddata-eval/tests/test_scoring.py

@@ -79,10 +79,10 @@ def test_check_filters_exact_attribute_match():
 # state, type) and never the LIST under state["include"], so two filters selecting the
 # same elements in a different order compare unequal.
 #
-# Found from a real eval run (gdc-mic-ai-evaluation, micai_diagnose_master, 2026-09-10):
-# a question filtering cross-border traffic scored metrics_correct=True,
-# dimensions_correct=True, filters_correct=False, because the fixture listed
-# ["Inter-region", "Intra-region"] and the agent emitted ["Intra-region", "Inter-region"].
+# Found from a real eval run: a question filtering on a two-value attribute scored
+# metrics_correct=True, dimensions_correct=True, filters_correct=False, because the fixture
+# listed ["Inter-region", "Intra-region"] and the agent emitted the same two the other way
+# round.
 # Element order is not something an agent has any reason to keep stable between runs, so
 # every question needing a multi-value attribute filter passes or fails partly at random.
 
@@ -95,7 +95,7 @@ def viz(values):
                 "filter_by": {
                     "f_a": {
                         "type": "attribute_filter",
-                        "using": "label/cross_border_name",
+                        "using": "label/region_name",
                         "state": {"include": values},
                     }
                 },
@@ -286,10 +286,10 @@ def test_normalized_filters_is_empty_per_category_when_unfiltered():
     assert normalized_filters(viz) == {"date": [], "ranking": [], "attribute": []}
 
 
-def _attr_viz(values, key="include", using="label/cross_border_name"):
+def _attr_viz(values, key="include", using="label/region_name"):
     return _viz(
         query={
-            "fields": {"m": {"using": "metric/approval_rate"}},
+            "fields": {"m": {"using": "metric/conversion_rate"}},
             "filter_by": {"f": {"type": "attribute_filter", "using": using, "state": {key: values}}},
         },
         metrics=["m"],
@@ -337,8 +337,8 @@ def test_include_and_exclude_of_the_same_elements_still_differ():
 
 
 def test_the_same_elements_on_a_different_label_still_differ():
-    expected = _attr_viz(["A", "B"], using="label/cross_border_name")
-    assert check_filters(expected, _attr_viz(["B", "A"], using="label/region_name")).attribute_ok is False
+    expected = _attr_viz(["A", "B"], using="label/region_name")
+    assert check_filters(expected, _attr_viz(["B", "A"], using="label/channel_name")).attribute_ok is False
 
 
 def test_a_mixed_type_element_list_does_not_crash_scoring():