"""Regressions for source/row requirements that cannot be satisfied by a draft.""" import copy import sys import tempfile import threading import unittest from pathlib import Path from unittest.mock import patch sys.path[:0]=[str(Path(__file__).resolve().parents[1]/'scripts'),str(Path(__file__).resolve().parent)] import nao_quality_runtime as runtime from nao_agent_task import TaskUnavailable, bind_task, normalize_task, plan_task from nao_evidence_finalizer import coverage_sufficient import test_agent_runtime as agent_runtime_fixture import test_weight_ranking as weight_fixture from test_nao_agent_task import task_value def rows_plan(): return task_value(kind='explain',goal='逐条提供全部原始记录',freshQuery=False,historyMode='none', dimensions=[],dimensionStages={},stages=['black'],fields=[],fullDetails=True, period={'year':2026,'month':9}) class RequiredReadCompletionTests(unittest.TestCase): def test_full_details_always_requires_a_fresh_read(self): planned=normalize_task(rows_plan(),'请逐条列出全部原始记录') self.assertTrue(planned['fullDetails']) self.assertTrue(planned['freshQuery']) def test_reviewer_cannot_claim_all_rows_are_required_without_evidence(self): inconsistent={'aligned':True,'requiresEvidence':False, 'requiresSourceVerification':False,'requiresAllSourceRows':True} with patch('nao_agent_task._model_json',side_effect=[rows_plan(),inconsistent]*3): with self.assertRaises(TaskUnavailable): plan_task(base_url='https://offline.invalid',api_key='synthetic',model='m', question='请逐条列出全部原始记录',context={},cancel_event=threading.Event()) def test_unverified_native_draft_cannot_finish_when_rows_required(self): harness=agent_runtime_fixture.AgentRuntimeIntegrationTest() with tempfile.TemporaryDirectory() as folder: # Simulate a legacy or malformed plan even after planner safeguards. planned=agent_runtime_fixture.task('请逐条列出全部原始记录',kind='explain',freshQuery=False, historyMode='none',dimensions=[],dimensionStages={},fields=[],stages=['black'],fullDetails=True) planned['freshQuery']=False;planned['requiresVerifiedEvidence']=False def native(**kwargs): stream=runtime.ConversationAccumulator() stream.blocks['native']='已经列出全部原始记录。';stream.block_order.append('native') stream.record_event('assistant',text='已经列出全部原始记录。') return stream,'COMPLETED','stop',None with patch('nao_agent_task.plan_repairs',return_value=[]): result=harness.run_mocked(Path(folder),planned,native) self.assertEqual(result['status'],'PARTIAL') self.assertNotIn('已经列出全部原始记录',result['answer']) self.assertFalse(any('已经列出全部原始记录' in entry.get('text','') for entry in result['timeline'])) def test_complete_rankings_do_not_satisfy_source_or_row_requirements(self): doc,_=weight_fixture.WeightRankingTests().run_rank() department=copy.deepcopy(doc);department['result']['dimension']='department';department['evidenceRef']='Q00000002' for flags in ({'verifySource':True},{'fullDetails':True}): with self.subTest(flags=flags): planned=weight_fixture.planned(**flags);context=bind_task({'releaseId':'release-a'},planned) self.assertFalse(coverage_sufficient([doc,department],weight_fixture.QUESTION,context)) def test_complete_rankings_with_required_source_or_all_rows_can_finish(self): doc,_=weight_fixture.WeightRankingTests().run_rank() department=copy.deepcopy(doc);department['result']['dimension']='department';department['evidenceRef']='Q00000002' side_context={'releaseId':'release-a','year':2026,'stage':'bright','includeUnassigned':False} def investigation(action,side): return {'mode':'investigate','evidenceRef':'Q00000003','result':{'kind':'quality_investigation', 'releaseId':'release-a','period':{'year':2026},'filters':{},'stage':'bright','action':action,'current':side}} source=investigation('source',{'context':side_context,'aggregateComplete':True,'totalRows':2, 'sourceVerification':{'complete':True,'allSelectedValuesMatchStored':True,'expectedRows':2, 'verifiedRows':2,'missingRows':0,'mismatchRows':0}}) def page(offset,more): return investigation('rows',{'context':side_context,'recordFilter':'all','matchingRows':4, 'offset':offset,'returnedRows':2,'rows':[{'id':offset},{'id':offset+1}],'hasMore':more}) for flags,extra in (({'verifySource':True},[source]),({'fullDetails':True},[page(0,True),page(2,False)])): with self.subTest(flags=flags): planned=weight_fixture.planned(**flags);context=bind_task({'releaseId':'release-a'},planned) self.assertTrue(coverage_sufficient([doc,department,*extra],weight_fixture.QUESTION,context)) def test_semantic_all_rows_accepts_complete_multiple_pages_but_not_gaps(self): question='列出黑片全部原始记录' planned=normalize_task(task_value(kind='details',fullDetails=True,dimensions=[],dimensionStages={}, stages=['black'],fields=[],period={'year':2026,'month':9},comparisonRequired=False),question) context=bind_task({'releaseId':'release-a'},planned) def page(offset, more, total=4): side={'context':{'releaseId':'release-a','year':2026,'month':9,'stage':'black','includeUnassigned':False}, 'recordFilter':'all','matchingRows':total,'offset':offset,'returnedRows':2, 'rows':[{'id':offset},{'id':offset+1}],'hasMore':more} return {'mode':'investigate','evidenceRef':'QAAAAAAAA','result':{'kind':'quality_investigation', 'releaseId':'release-a','period':{'year':2026,'month':9},'filters':{},'stage':'black', 'action':'rows','current':side}} pages=[page(0,True),page(2,False)] self.assertTrue(coverage_sufficient(pages,question,context)) self.assertFalse(coverage_sufficient([pages[0],page(3,False)],question,context)) self.assertFalse(coverage_sufficient([pages[0],page(2,False,total=5)],question,context)) if __name__=='__main__':unittest.main()