import sys from pathlib import Path import unittest from unittest.mock import patch sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "scripts")) from analysis_visual_evidence import visual_result from analysis_visuals import build_visuals def evidence(result=None): result = result or { "kind": "quality_comparison", "releaseId": "r-1", "filterBinding": {"department": "电机部", "includeUnassigned": False, "extraFilters": {}}, "current": { "period": {"year": 2026, "month": 9}, "compositeRate": 0.98, "black": {"rate": 0.97, "scrapWeight": 12.0}, "bright": {"rate": 0.99, "scrapWeight": 5.0}, }, "comparison": { "period": {"year": 2026, "month": 8}, "compositeRate": 0.96, "black": {"rate": 0.95, "scrapWeight": 10.0}, "bright": {"rate": 0.98, "scrapWeight": 7.0}, }, "delta": {"available": True, "percentagePoints": 2.0}, "stageRateChangesPP": {"black": 2.0, "bright": 1.0}, } return { "tool": "quality_data", "mode": "compare", "releaseId": "r-1", "evidenceRef": "Q1234ABCD", "result": {"release": {"id": "r-1"}, **result}, } def task(question, *, preference="auto", chart_type=None, kind="analysis", dimensions=None, stages=None, fields=None): return { "version": 1, "question": question, "goal": "compare formal quality evidence", "kind": kind, "dimensions": dimensions or [], "stages": stages or [], "fields": fields or [], "freshQuery": True, "visualPreference": preference, "visualType": chart_type, "responseStyle": "concise", } def context_for(question, **kwargs): return { "releaseId": "r-1", "filters": {"department": "电机部"}, "_agentTask": task(question, **kwargs), } class AgentVisualTests(unittest.TestCase): def test_semantic_task_makes_chart_choice_independent_of_question_wording(self): questions = ["电机部近期情况如何", "帮我看看这段时间的表现"] outputs = [build_visuals([evidence()], question, "turn-semantic", context=context_for( question, dimensions=["department"], fields=["rate"])) for question in questions] self.assertEqual(outputs[0]["visualState"], "ready") self.assertEqual(outputs[0]["visuals"], outputs[1]["visuals"]) self.assertEqual({chart["metric"] for chart in outputs[0]["visuals"]}, {"composite_rate", "black_rate", "bright_rate"}) def test_none_overrides_explicit_chart_words(self): question = "画图比较两期综合合格率" output = build_visuals([evidence()], question, "turn-none", context=context_for(question, preference="none", fields=["rate"])) self.assertEqual(output["visualState"], "not_requested") self.assertEqual(output["visuals"], []) def test_required_without_evidence_reports_unavailable_and_invents_nothing(self): question = "检查重点产品" output = build_visuals([], question, "turn-required-empty", context=context_for(question, preference="required", fields=["rate"])) self.assertEqual(output["visualState"], "unavailable") self.assertEqual(output["visuals"], []) def test_wrong_question_task_is_ignored(self): question = "比较两个期间变化" stale = task("不要图,只要文字", preference="none") output = build_visuals([evidence()], question, "turn-stale", context={"releaseId": "r-1", "filters": {"department": "电机部"}, "_agentTask": stale}) self.assertEqual(output["visualState"], "ready") self.assertTrue(output["visuals"]) def test_semantic_scope_keeps_evidence_binding_and_requested_stage(self): question = "只看本部门黑片" output = build_visuals([evidence()], question, "turn-bound", context=context_for(question, dimensions=["department"], stages=["black"], fields=["rate"])) self.assertEqual(output["visualState"], "ready") self.assertEqual(len(output["visuals"]), 1) chart = output["visuals"][0] self.assertEqual(chart["stage"], "black") self.assertEqual(chart["scope"]["department"], "电机部") self.assertEqual(chart["evidenceRefs"], ["Q1234ABCD"]) def test_supported_semantic_style_is_applied_without_style_words(self): question = "请查看两期质量表现" output = build_visuals([evidence()], question, "turn-style", context=context_for(question, chart_type="column", fields=["rate"])) self.assertEqual(output["visualState"], "ready") self.assertTrue(output["visuals"]) self.assertTrue(all(chart["type"] == "column" for chart in output["visuals"])) def test_unsupported_for_evidence_style_has_no_fallback(self): question = "请查看两期质量表现" output = build_visuals([evidence()], question, "turn-style-unsafe", context=context_for(question, preference="required", chart_type="heatmap", fields=["rate"])) self.assertEqual(output["visualState"], "unavailable") self.assertEqual(output["visuals"], []) def test_auto_style_hint_cannot_hide_an_evidence_supported_chart(self): question='请查看两期质量表现' output=build_visuals([evidence()],question,'turn-auto-style', context=context_for(question,chart_type='heatmap',fields=['rate'])) self.assertEqual(output['visualState'],'ready') self.assertTrue(output['visuals']) self.assertTrue(all(chart['type'] in ('bar','column') for chart in output['visuals'])) def test_auto_pareto_for_source_defects_preserves_weights_without_inventing_shares(self): from test_analysis_visuals import fine_field question='帮我看下上月黑片缺陷TOP' keys=['责废缺角重量','料废缺角重量'] result={'kind':'quality_fine_defects','releaseId':'r-1','dimension':'defect', 'grain':'source_defect_weight','stage':'black','filters':{},'period':{'year':2026,'month':9}, 'coverage':{'available':True,'sourceRows':2},'sourceRows':2, 'sourceFields':[fine_field(key,'责废' if key.startswith('责废') else '料废') for key in keys], 'rows':[{'key':key,'label':key,'category':'责废' if key.startswith('责废') else '料废','weight':weight} for key,weight in zip(keys,[17513.162,13949.63])], 'totalRows':2,'truncated':False,'rankingComplete':True} context={'releaseId':'r-1','filters':{},'_agentTask':task(question,dimensions=['defect'],stages=['black'],chart_type='pareto')} output=build_visuals([evidence(result)],question,'turn-auto-defect',context=context) self.assertEqual(output['visualState'],'ready') chart=output['visuals'][0] self.assertEqual(chart['type'],'bar') self.assertEqual(chart['series'][0]['values'],[17513.162,13949.63]) self.assertFalse(any(series.get('unit')=='%' for series in chart['series'])) context['_agentTask']['visualPreference']='required' self.assertEqual(build_visuals([evidence(result)],question,'turn-required-defect',context=context)['visuals'],[]) def test_auto_does_not_chart_a_single_number_metric(self): question = "本期综合合格率是多少" metric = evidence({ "kind": "quality_metrics", "releaseId": "r-1", "view": "overall", "filters": {"department": "电机部", "includeUnassigned": False, "extraFilters": {}}, "period": {"year": 2026, "month": 9}, "summary": {"compositeRate": 0.98}, }) for kind in ("metrics", "restate", "explain"): with self.subTest(kind=kind): output = build_visuals([metric], question, f"turn-single-metric-{kind}", context=context_for(question, kind=kind, fields=["rate"])) self.assertEqual(output["visualState"], "not_requested") self.assertEqual(output["visuals"], []) def test_evidence_adapter_forwards_the_validated_task_unchanged(self): question = "看两期表现" request_context = context_for(question, fields=["rate"]) with patch("analysis_visual_evidence.load_visual_evidence", return_value=[evidence()]), \ patch("analysis_visuals.build_visuals", return_value={"visuals": []}) as builder: visual_result(Path("."), "r-1", question, "turn-forward", context=request_context) self.assertIs(builder.call_args.kwargs["context"]["_agentTask"], request_context["_agentTask"]) if __name__ == "__main__": unittest.main()