"""Comparison boundaries are predicates, never rounded observed values.""" import io,json,sys,threading,unittest from pathlib import Path from unittest.mock import patch sys.path[:0]=[str(Path(__file__).resolve().parents[1]/'scripts'),str(Path(__file__).parent)] from nao_evidence_finalizer import synthesize_native from test_agent_answer import evidence from nao_native_answer import _bind_literal_measures,_binding_inventory,_validate_blocks,NativeCompositionError,_native_strict_bound class ComparisonBoundPipelineTests(unittest.TestCase): def report(self,second='95.83'): facts=[] for i,name,value,raw in [(1,'规格甲','95.01','95.00708964391122'),(2,'规格乙',second,second)]: facts.append({'id':'F'+str(i),'kind':'product','label':name,'text':name+';综合合格率'+value+'%。', 'evidenceRef':'Q00000001','scope':{'filters':{},'period':{'year':2026,'month':8},'dimension':'product'}, 'validity':{'verified':True,'scopeApplicable':True,'basis':'quality_weight_ranking'}, 'verifiedComparisonValues':[{'metricField':'rate','unit':'%','stageHint':'综合','rawValue':raw,'displayValue':value+'%'}]}) return {'facts':facts,'requiredFactIds':[],'priorityFactIds':[],'missingEvidence':[],'rejectedEvidenceRefs':[], 'notices':[], 'sections':[]} def invoke(self,text,*,second='95.83',native=False): calls=[] def provider(request,timeout): body=json.loads(request.data);payload=json.loads(body['messages'][1]['content']);calls.append(body) if 'draftBlocks' in payload: value={'reviews':[{'sourceId':b['sourceId'],'role':'core','factIds':['F1','F2'],'supported':True,'statisticalClaimsSupported':True} for b in payload['draftBlocks']], 'missingGoalDescriptions':[]} elif body['messages'][0]['content'].startswith('逐句核验证据'): value={'reviews':[{'supported':True,'statisticalClaimsSupported':True}],'missingGoalDescriptions':[]} else:value={'blocks':[{'role':'core','text':text,'factIds':['F1','F2']}]} return io.BytesIO(json.dumps({'choices':[{'finish_reason':'stop','message':{'content':json.dumps(value)}}]}).encode()) with patch('nao_native_answer.build_native_report',return_value=self.report(second)): final=synthesize_native(base_url='https://offline.invalid',api_key='fixture',model='fixture',question='比较两规格', evidence=evidence(),context={'_agentTask':{'kind':'analysis','period':{'year':2026,'month':8}}}, cancel_event=threading.Event(),opener=provider,native_answer=text if native else None) return final,calls def test_point_display_and_collective_boundary_remain_distinct(self): for native in (False,True): final,_=self.invoke('规格甲与规格乙的综合合格率均在95%以上。',native=native) self.assertTrue(final['bodyEvidenceSufficient'],final['compositionDiagnostics']['issues']) self.assertIn('95%以上',final['answer']);self.assertNotIn('95.01%以上',final['answer']) point,_=self.invoke('规格甲综合合格率95.01%。',native=native) self.assertTrue(point['bodyEvidenceSufficient']);self.assertIn('95.01%',point['answer']) def test_each_explicit_subject_must_pass_the_unrounded_relation(self): for native in (False,True): for text,second in [('规格甲与规格乙的综合合格率均在95.01%以上。','95.83'), ('规格甲与规格乙的综合合格率均在95%以上。','94.99'), ('规格甲与规格乙的综合合格率均低于95%。','95.83')]: final,_=self.invoke(text,second=second,native=native) self.assertFalse(final['bodyEvidenceSufficient']) good,_=self.invoke('规格甲与规格乙的综合合格率均不低于95%。',native=native) self.assertTrue(good['bodyEvidenceSufficient'],good['compositionDiagnostics']['issues']) def test_other_typed_units_never_canonicalize_a_comparison_boundary(self): for value,unit,boundary in [('1155.4834','kg','1155'),('0.30416066','pp','0.30')]: facts=[{'id':'F1','text':'正式数值'+value+unit+'。','evidenceRef':'Q00000001'}] for text in ('超过'+boundary+unit+'。',boundary+unit+'以上。'): tokens=_binding_inventory(facts) parsed=_bind_literal_measures({'blocks':[{'text':text,'factIds':['F1']}]},tokens,{}) self.assertEqual(parsed['blocks'][0]['text'],text) with self.assertRaises(NativeCompositionError):_validate_blocks(parsed,facts,tokens) self.assertTrue(_native_strict_bound(text,text.index(boundary))) if __name__=='__main__':unittest.main()