修复:完整展示AI原文并穿插图表,统一运行反馈与变化符号

This commit is contained in:
2026-11-02 12:03:37 +08:00
parent cf8e25a30d
commit 903fa94c00
35 changed files with 1016 additions and 172 deletions
+77 -51
View File
@@ -1,6 +1,8 @@
"""Exercise the native runner with isolated files and mocked upstream/bridge calls."""
import copy
import importlib.util
import json
import os
import sys
import tempfile
import threading
@@ -10,8 +12,15 @@ from pathlib import Path
from types import SimpleNamespace
from unittest.mock import Mock, patch
sys.path[:0] = [str(Path(__file__).resolve().parents[1] / 'scripts')]
from nao_native_run import run_native_exploration
PROJECT = Path(__file__).resolve().parents[1]
sys.path[:0] = [str(PROJECT / 'scripts')]
RUNTIME_PATH = Path(os.environ.get('NAO_ORIGINAL_BODY_PATH',
str(PROJECT / 'scripts/nao_native_run.py'))).expanduser().resolve()
RUNTIME_SPEC = importlib.util.spec_from_file_location('nao_native_run_integration', RUNTIME_PATH)
RUNTIME_MODULE = importlib.util.module_from_spec(RUNTIME_SPEC)
sys.modules[RUNTIME_SPEC.name] = RUNTIME_MODULE
RUNTIME_SPEC.loader.exec_module(RUNTIME_MODULE)
run_native_exploration = RUNTIME_MODULE.run_native_exploration
from nao_quality_runtime import _atomic_json
@@ -54,31 +63,25 @@ class NativeRunTests(unittest.TestCase):
'missingEvidence': ['节点bright-detail范围基础范围缺少bright来源具体缺陷重量图。']}
result = self.invoke()
self.assertEqual(result['status'], 'COMPLETED')
self.assertEqual(result['readingSummary']['notices'],
self.assertEqual(result['answerQuality']['notices'],
['亮片细分缺陷尚未完整核实:当前范围;未用于已核实结论。'])
self.assertNotIn('调查局限', result['answer'])
self.assertNotIn('bright', result['answer'])
stored = json.loads((self.actor_root / 'private/native-warning-diagnostics.json').read_text())
self.assertIn('bright', stored['visualWarnings'][0])
def test_full_facts_remain_private_instead_of_being_appended_to_the_business_answer(self):
self.final['fullEvidenceFacts'] = [{'text': '后台保留的完整统计事实', 'evidenceRef': 'Q00000001'}]
def test_original_answer_is_not_replaced_by_a_generated_report(self):
self.accumulator.answer = '原生回答原样保留,局限也照原文保留。'
result = self.invoke()
self.assertNotIn('完整统计依据', result['answer'])
self.assertNotIn('后台保留的完整统计事实', result['answer'])
stored = json.loads((self.actor_root / 'private/native-full-evidence-facts.json').read_text())
self.assertEqual(stored, self.final['fullEvidenceFacts'])
self.assertEqual(result['answer'], self.accumulator.answer)
self.assertIsNone(result['readingSummary'])
self.synthesis.assert_not_called()
def test_issued_binding_catalog_remains_private_for_exact_replay(self):
issued = {'bindingInventory': {'N17': {'factId': 'F2', 'value': '493.9966', 'unit': 'kg'}},
'facts': [{'id': 'F2', 'rawValue': '493.99660000'}]}
self.final['_privateBindingInput'] = issued
def test_no_second_composition_artifact_is_written(self):
result = self.invoke()
stored = json.loads((self.actor_root / 'private/native-composition.json').read_text())
self.assertEqual(stored['bindingInput'], issued)
self.assertNotIn('_privateBindingInput', result)
self.assertNotIn('bindingInput', result)
self.assertNotIn('bindingInventory', json.dumps(self.events, ensure_ascii=False))
self.assertEqual(result['answer'], self.accumulator.answer)
self.assertFalse((self.actor_root / 'private/native-composition.json').exists())
self.synthesis.assert_not_called()
def setUp(self):
self.folder = tempfile.TemporaryDirectory()
@@ -105,11 +108,6 @@ class NativeRunTests(unittest.TestCase):
'scopeCoverage': [{'nodeId': 'root', 'scopeId': 'scope-root', 'complete': True},
{'nodeId': 'group', 'scopeId': 'scope-group', 'complete': True}],
'rootGoalGaps': [], 'gaps': [], 'pending': [], 'explorationWarnings': []}
self.final = {'answer': '按已核实结果解释[Q00000001]', 'reportTitle': '质量数据分析',
'readingSummary': {'version': 1, 'lead': '已核实解释[Q00000001]',
'evidence': [], 'notices': [], 'evidenceRefs': ['Q00000001']},
'usedEvidenceRefs': ['Q00000001'], 'evidenceSufficient': True,
'missingEvidence': [], 'rawModelResponses': [], 'compositionDiagnostics': {}}
self.accumulator = SimpleNamespace(answer='原生草稿', finish_reasons=['stop'],
tool_calls=[{'tool': 'quality_data', 'status': 'failed', 'input': {'mode': 'compare'}},
{'tool': 'quality_data', 'status': 'completed', 'input': {'mode': 'compare'}}],
@@ -127,7 +125,7 @@ class NativeRunTests(unittest.TestCase):
self.read_evidence = stack.enter_context(patch('nao_evidence_finalizer.evidence_summaries',
side_effect=lambda *args, **kwargs: copy.deepcopy(self.documents)))
self.synthesis = stack.enter_context(patch('nao_evidence_finalizer.synthesize_native',
side_effect=lambda **kwargs: copy.deepcopy(self.final)))
side_effect=AssertionError('native-agent original body must not be synthesized')))
self.visuals = stack.enter_context(patch('analysis_visual_evidence.protocol_visual_result',
return_value={'visualState': 'not_requested', 'visuals': [], 'missingEvidence': []}))
stack.enter_context(patch('nao_runtime_policy.native_timeout', return_value=30))
@@ -161,8 +159,12 @@ class NativeRunTests(unittest.TestCase):
self.assertTrue(result['evidenceCompletion']['serverCoverageSufficient'])
self.assertEqual(result['timeline'], self.accumulator.timeline)
self.assertEqual(result['conclusionProtocolQueryIds'], ['query-current', 'query-compare'])
self.assertIn(self.snapshot['explorationWarnings'][0], result['readingSummary']['notices'])
self.assertEqual(result['conclusionProtocolQueryIds'],
['query-current', 'query-compare', 'query-group'])
self.assertEqual(result['answerQuality']['citedEvidenceRefs'], [])
self.assertEqual(result['answerQuality']['narrativeFactBinding'], 'not_assessed')
self.assertIn(self.snapshot['explorationWarnings'][0], result['answerQuality']['notices'])
self.assertEqual(result['answer'], self.accumulator.answer)
self.assertIn('firstUsefulMs', result['phaseTimingsMs'])
self.assertGreaterEqual(result['phaseTimingsMs']['firstUsefulMs'], 30)
self.delegated.call.assert_called_once_with('/api/analysis/protocol/queries', payload={}, method='POST')
@@ -172,35 +174,33 @@ class NativeRunTests(unittest.TestCase):
self.assertIsNone(self.sse.call_args.kwargs['chat_id'])
stored = json.loads((self.actor_root / 'private/native-completion.json').read_text())
self.assertEqual(stored['status'], 'COMPLETED')
self.assertEqual(stored['answer'], self.accumulator.answer)
self.assertEqual(self.events[-1], result)
def test_generated_chart_contract_is_available_before_the_single_synthesis(self):
def test_generated_chart_contract_uses_formal_query_evidence_without_rewriting_answer(self):
self.task['visualPreference']='required'
self.visuals.return_value={'visualState':'ready','missingEvidence':[], 'visuals':[{
'id':'actual-chart','type':'line','metric':'composite_rate','title':'正式趋势',
'scope':{'department':'当前产业部'},'period':{'year':2026,'month':9},
'categories':[{'key':'2026-09'}],'series':[{'values':[.95]}],
'evidenceRefs':['Q00000001']}]}
def synthesize(**kwargs):
preview=kwargs['context']['_platformVisuals']
self.assertEqual(preview['visualState'],'ready')
self.assertEqual(preview['charts'][0]['id'],'actual-chart')
self.assertEqual(preview['charts'][0]['evidenceRefs'],['Q00000001'])
self.assertEqual(preview['charts'][0]['pointCount'],1)
self.assertNotIn('series',preview['charts'][0])
return copy.deepcopy(self.final)
self.synthesis.side_effect=synthesize
result=self.invoke()
self.assertEqual(result['status'],'COMPLETED')
self.visuals.assert_called_once();self.synthesis.assert_called_once()
self.visuals.assert_called_once()
self.assertEqual(self.visuals.call_args.args[0], self.documents)
self.assertEqual(result['visuals'][0]['id'], 'actual-chart')
self.assertEqual(result['visuals'][0]['evidenceRefs'], ['Q00000001'])
self.assertEqual(result['answer'], self.accumulator.answer)
self.synthesis.assert_not_called()
def test_root_gap_remains_partial_even_with_stop_and_available_evidence(self):
self.snapshot['rootGoalGaps'] = ['亮片必答阶段尚无完整两期依据。']
result = self.invoke()
self.assertEqual(result['status'], 'PARTIAL')
self.assertFalse(result['evidenceCompletion']['serverCoverageSufficient'])
self.assertIn(self.snapshot['rootGoalGaps'][0], result['answer'])
self.assertIn(self.snapshot['rootGoalGaps'][0], result['readingSummary']['notices'])
self.assertEqual(result['answer'], self.accumulator.answer)
self.assertIn(self.snapshot['rootGoalGaps'][0], result['evidenceCompletion']['missingEvidence'])
self.assertIn(self.snapshot['rootGoalGaps'][0], result['answerQuality']['missingEvidence'])
def test_three_no_progress_attempts_remain_partial_after_natural_stop(self):
_atomic_json(self.actor_root / 'private/native-exploration-state.json',
@@ -210,7 +210,8 @@ class NativeRunTests(unittest.TestCase):
self.assertEqual(result['status'], 'PARTIAL')
self.assertTrue(any('连续三次' in gap for gap in result['evidenceCompletion']['missingEvidence']))
self.assertEqual(result['phaseTimingsMs']['formalQuery'], 1234)
self.assertTrue(any('连续三次' in notice for notice in result['readingSummary']['notices']))
self.assertTrue(any('连续三次' in gap for gap in result['answerQuality']['missingEvidence']))
self.assertEqual(result['answer'], self.accumulator.answer)
def test_controlled_no_progress_stop_does_not_add_generic_tool_failure(self):
_atomic_json(self.actor_root / 'private/native-exploration-state.json',
@@ -233,7 +234,7 @@ class NativeRunTests(unittest.TestCase):
self.assertEqual(result['status'], 'PARTIAL')
self.assertTrue(any('未正常结束' in gap for gap in result['evidenceCompletion']['missingEvidence']))
def test_only_complete_current_scope_documents_reach_synthesis(self):
def test_only_complete_current_scope_documents_reach_visual_generation_and_result(self):
self.snapshot['queries'].extend([{'proof': self.p_failed}, {'proof': self.p_old}])
self.snapshot['scopeCoverage'].append(
{'nodeId': 'failed', 'scopeId': 'scope-failed', 'complete': False})
@@ -245,27 +246,52 @@ class NativeRunTests(unittest.TestCase):
{'evidenceRef': 'Q00000007', 'result': {'summary': {'unproved': True}}},
])
result = self.invoke()
admitted = self.synthesis.call_args.kwargs['evidence']
admitted = self.visuals.call_args.args[0]
self.assertEqual([doc['evidenceRef'] for doc in admitted], ['Q00000001', 'Q00000002'])
self.assertEqual(result['queryEvidenceRefs'], ['Q00000001', 'Q00000002'])
self.assertEqual(result['conclusionProtocolQueryIds'],
['query-current', 'query-compare', 'query-group'])
self.assertNotIn('query-failed', result['conclusionProtocolQueryIds'])
self.assertNotIn('query-old', result['conclusionProtocolQueryIds'])
self.assertNotIn('query-not-in-current-receipts', result['conclusionProtocolQueryIds'])
self.assertEqual(result['answer'], self.accumulator.answer)
self.read_evidence.assert_called_once_with(self.actor_root / 'evidence', RELEASE,
exclude_files={'prior-evidence.json'}, allow_empty=True)
self.synthesis.assert_not_called()
def test_used_refs_map_only_to_actual_current_proofs_not_unused_or_foreign_documents(self):
def test_original_citations_map_only_to_actual_current_query_proofs(self):
self.snapshot['queries'].append({'proof': self.p_old})
self.documents.append(document('Q00000003', self.p_old))
self.final['usedEvidenceRefs'] = ['Q00000001', 'Q00000003', 'QDEADBEEF', 'Q00000001']
self.accumulator.answer = '原文引用当前证据[Q00000001],以及外部证据[Q00000003]。'
result = self.invoke()
self.assertEqual(result['conclusionProtocolQueryIds'], ['query-current', 'query-compare'])
self.assertEqual(result['answer'], self.accumulator.answer)
self.assertEqual(result['answerQuality']['unresolvedEvidenceRefs'], ['Q00000003'])
self.assertEqual(result['status'], 'PARTIAL')
self.assertNotIn('query-group', result['conclusionProtocolQueryIds'])
self.assertNotIn('query-old', result['conclusionProtocolQueryIds'])
def test_claimed_sufficient_answer_without_current_cited_proof_remains_partial(self):
self.final['usedEvidenceRefs'] = ['QDEADBEEF']
def test_uppercase_hexadecimal_reference_maps_to_its_actual_receipt(self):
hexadecimal = proof('query-hexadecimal')
self.snapshot['queries'].append({'proof': hexadecimal})
self.documents.append(document('QABCDEF12', hexadecimal))
self.accumulator.answer = '原文显式引用[QABCDEF12]。'
result = self.invoke()
self.assertEqual(result['status'], 'COMPLETED')
self.assertEqual(result['answer'], self.accumulator.answer)
self.assertEqual(result['conclusionProtocolQueryIds'], ['query-hexadecimal'])
self.assertEqual(result['answerQuality']['citedEvidenceRefs'], ['QABCDEF12'])
def test_unbound_original_citation_remains_a_metadata_gap(self):
self.accumulator.answer = '原生回答的引用[QABCDEF99]仍按原文保留。'
result = self.invoke()
self.assertEqual(result['status'], 'PARTIAL')
self.assertEqual(result['conclusionProtocolQueryIds'], [])
self.assertTrue(any('实际查询证明' in gap for gap in result['evidenceCompletion']['missingEvidence']))
self.assertEqual(result['answer'], self.accumulator.answer)
self.assertEqual(result['answerQuality']['unresolvedEvidenceRefs'], ['QABCDEF99'])
self.assertTrue(any('QABCDEF99' in gap for gap in result['evidenceCompletion']['missingEvidence']))
def test_cancelled_native_turn_never_reads_query_snapshot_or_synthesizes(self):
for cancellation in ('event', 'native_status'):
@@ -295,12 +321,12 @@ class NativeRunTests(unittest.TestCase):
self.assertNotIn('图表尚未完整生成,可独立恢复。', result['answer'])
self.assertNotIn('结果为部分完成', result['answer'])
self.assertNotIn('共有15张图表', result['answer'])
self.assertEqual(result['readingSummary']['notices'],
self.assertEqual(result['answerQuality']['notices'],
['部分重点图表未能完整展示;已保留核实文字和表格。'])
else:
self.assertIn('图表尚未完整生成,可独立恢复。', result['answer'])
self.assertIn('图表尚未完整生成,可独立恢复。', result['readingSummary']['notices'])
self.assertIn('结果为部分完成', result['answer'])
self.assertNotIn('图表尚未完整生成,可独立恢复。', result['answer'])
self.assertNotIn('结果为部分完成', result['answer'])
self.assertIn('图表尚未完整生成,可独立恢复。', result['answerQuality']['missingEvidence'])
if __name__ == '__main__':