Files
quality_analyse/verification/test_quantity_comparison_bounds.py
T

134 lines
9.5 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""Typed comparison operands preserve their boundary, independently of units."""
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 nao_native_answer import _binding_inventory,_validate_blocks,NativeCompositionError
from test_agent_answer import evidence
class QuantityComparisonPipelineTests(unittest.TestCase):
def report(self,second='1106.0721',*,unit='kg',stage='综合',field='scrapWeight'):
facts=[]
for i,name,raw in [(1,'规格甲','1155.4834'),(2,'规格乙',second)]:
facts.append({'id':'F'+str(i),'kind':'product','label':name,
'text':name+';'+stage+'报废重量'+raw+unit+'。', 'evidenceRef':'Q00000001',
'scope':{'filters':{},'period':{'year':2026,'month':8},'dimension':'product','stage':'overall'},
'validity':{'verified':True,'scopeApplicable':True,'basis':'quality_weight_ranking'},
'verifiedComparisonValues':[{'metricField':field,'unit':unit,'stageHint':stage,
'rawValue':raw,'displayValue':raw}]})
return {'facts':facts,'requiredFactIds':[],'priorityFactIds':[],'missingEvidence':[],
'rejectedEvidenceRefs':[],'notices':[],'sections':[]}
def invoke(self,text,*,native=False,second='1106.0721',report=None,cited=None):
calls=[];report=report or self.report(second);fact_ids=cited or [fact['id'] for fact in report['facts']]
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':fact_ids,
'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':fact_ids}]}
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=report):
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_mass_boundary_uses_each_raw_and_preserves_the_boundary(self):
for native in (False,True):
for text in ('规格甲与规格乙的综合报废重量均在1000 kg以上。','规格甲与规格乙的综合报废重量均超过1000 kg。',
'规格甲与规格乙的综合报废重量均在1吨以上。'):
final,_=self.invoke(text,native=native)
self.assertTrue(final['bodyEvidenceSufficient'],final['compositionDiagnostics']['issues'])
self.assertIn('1吨以上' if '吨' in text else '1000 kg',final['answer'])
self.assertNotIn('1155.4834',final['answer'])
def test_relation_flip_and_one_failing_member_are_rejected(self):
for native in (False,True):
for text,second in [('规格甲与规格乙的综合报废重量均低于1000 kg。','1106.0721'),
('规格甲与规格乙的综合报废重量均在1000 kg以上。','999.99'),
('规格甲与规格乙的综合报废重量均超过1155.4834 kg。','1200')]:
final,_=self.invoke(text,native=native,second=second)
self.assertFalse(final['bodyEvidenceSufficient'])
def test_unknown_unit_wrong_field_and_wrong_stage_are_rejected(self):
for native in (False,True):
for report in (self.report(unit='source_weight'),self.report(field='totalWeight'),self.report(stage='黑片')):
final,_=self.invoke('规格甲与规格乙的综合报废重量均超过1000 kg。',native=native,report=report)
self.assertFalse(final['bodyEvidenceSufficient'])
def test_explicit_period_conflict_is_rejected(self):
for native in (False,True):
final,_=self.invoke('9月规格甲与规格乙的综合报废重量均超过1000 kg。',native=native)
self.assertFalse(final['bodyEvidenceSufficient'])
def test_an_explicit_second_member_cannot_be_ignored_when_uncited(self):
report=self.report(second='999.99');report['facts'][1]['evidenceRef']='Q00000002'
for native in (False,True):
final,_=self.invoke('规格甲与规格乙的综合报废重量均超过1000 kg。',native=native,report=report,cited=['F1'])
self.assertFalse(final['bodyEvidenceSufficient'])
def test_native_typed_pp_uses_the_same_relation_and_field_role(self):
report=self.report()
for fact in report['facts']:
fact['text']=fact['label']+';黑片对综合合格率贡献+0.30个百分点。'
fact.pop('verifiedComparisonValues')
fact['verifiedNativeNumericValues']=[{'sourceOrdinal':0,'displayValue':'+0.30','rawValue':'0.30416066',
'unit':'pp','metricField':'stageContribution','stageHint':'黑片','period':{'year':2026,'month':8},
'metricRole':'ContributionPp','ownerScope':{}}]
good,_=self.invoke('规格甲与规格乙的黑片对综合合格率贡献均超过0.30个百分点。',native=True,report=report)
self.assertTrue(good['bodyEvidenceSufficient'],good['compositionDiagnostics']['issues'])
self.assertIn('0.30个百分点',good['answer']);self.assertNotIn('0.30416066',good['answer'])
bad,_=self.invoke('规格甲与规格乙的黑片对综合合格率贡献均低于0.30个百分点。',native=True,report=report)
self.assertFalse(bad['bodyEvidenceSufficient'])
for fact in report['facts']:
fact['verifiedNativeNumericValues'][0].update(metricField='rateChange',metricRole='OwnStageRateChangePp')
wrong_role,_=self.invoke('规格甲与规格乙的黑片对综合合格率贡献均超过0.30个百分点。',native=True,report=report)
self.assertFalse(wrong_role['bodyEvidenceSufficient'])
def test_observation_tokens_cannot_bypass_a_suffix_boundary(self):
facts=self.report()['facts'];tokens=_binding_inventory(facts)
key=next(key for key,value in tokens.items() if value.get('factId')=='F1' and value.get('metricField')=='scrapWeight')
for text in ('规格甲报废重量{{'+key+'}} kg以上。','规格甲报废重量超过{{'+key+'}} kg。'):
with self.assertRaises(NativeCompositionError):
_validate_blocks({'blocks':[{'text':text,'factIds':['F1']}]},facts,tokens)
def history_report(self,second='1106.07210000'):
report=self.report();facts=report['facts']
facts.append({**facts[0],'id':'F3','label':'规格丙'})
raw_values=['1155.48340000',second,'1060.77910000']
stage_values=[('446.851','708.6324'),('659.65','446.4221'),('538.905','521.8741')]
scope={'filters':{},'period':{'year':2026,'month':8},'dimension':'product','stage':'overall'}
collection={'verified':True,'complete':True,'rankGroupKey':'history-rank:fixture','memberFactIds':['F1','F2','F3'],
'period':scope['period'],'scope':scope,'dimension':'product','metric':'scrapWeight','stage':'overall',
'unit':'kg','requestedLimit':3,'returnedCount':3,'totalRows':10,'truncated':True}
for rank,(fact,raw) in enumerate(zip(facts,raw_values),1):
black,bright=stage_values[rank-1]
fact.update(text=fact['label']+';综合报废重量'+raw+' kg;黑片报废重量'+black+' kg;亮片报废重量'+bright+' kg。',scope=scope,
validity={'verified':True,'scopeApplicable':True,'basis':'protocol_history_snapshot'},
rank=rank,rankGroupKey=collection['rankGroupKey'],historyCollection=collection,
verifiedDisplayValues=[{'metricCode':code,'rawValue':value,'rawUnit':'source_weight','unit':'kg','displayValue':value}
for code,value in [('scrapWeight',raw),('blackScrapWeight',black),('brightScrapWeight',bright)]])
fact.pop('verifiedComparisonValues',None)
return report
def test_captured_collective_sentence_uses_all_returned_members(self):
# The unchanged original sentence and Decimal values; semantic review
# is an explicit test fixture, not a fabricated captured model verdict.
text='8月三款规格报废重量均在 1000 kg 以上。'
for native in (False,True):
good,_=self.invoke(text,native=native,report=self.history_report())
self.assertTrue(good['bodyEvidenceSufficient'],good['compositionDiagnostics']['issues'])
self.assertIn('1000 kg',good['answer'])
bad,_=self.invoke(text,native=native,report=self.history_report(second='999.99'))
self.assertFalse(bad['bodyEvidenceSufficient'])
reverse,_=self.invoke(text.replace('以上','以下'),native=native,report=self.history_report())
self.assertFalse(reverse['bodyEvidenceSufficient'])
if __name__=='__main__':unittest.main()