优化:未分类连续维护与柏拉图预汇总,同步已发布修复

This commit is contained in:
2026-11-01 12:40:50 +08:00
parent 572873c6b8
commit 38290f5fec
135 changed files with 17613 additions and 663 deletions
+76 -5
View File
@@ -145,6 +145,16 @@ class NativeAnswerTests(unittest.TestCase):
else:
with self.assertRaises(NativeCompositionError): _validate_blocks(parsed, report['facts'], tokens)
self.assertEqual(wire, wire_copy)
# A compact wire response with an actual Top 2 must not silently become
# Top 1 just because the presentation projection omitted result.rows.
complete_wire = copy.deepcopy(wire)
complete_wire.update(rowsReturned=2, truncated=False)
report = build_native_report([document(complete_wire)], context={'releaseId': RELEASE,
'_agentTask': {'executionMode': 'native_agent'}})
self.assertFalse(report['missingEvidence'])
paired_facts = [fact for fact in report['facts'] if fact['kind'] == 'defect' and '来源细项'+str(fact['label'])+':' in fact['text']]
self.assertEqual([fact['label'] for fact in paired_facts], ['责废尺寸重量', '责废缺角重量'])
self.assertFalse(any(fact['kind'] == 'defect_context' for fact in report['facts']))
def test_exact_changes_raw_alias_matches_formal_display_without_precision_guessing(self):
facts = [{'id': 'F1', 'kind': 'product', 'label': '规格甲', 'text': '规格甲合格率95.00%。', 'scrapDelta': 758.7594,
@@ -586,8 +596,10 @@ class NativeAnswerTests(unittest.TestCase):
report = {'facts': facts, 'requiredFactIds': [], 'priorityFactIds': [], 'scopes': [], 'missingEvidence': [], 'rejectedEvidenceRefs': []}
for phrase in ('均未达到0.59 kg', '均未达0.59 kg', '相等'):
for verdict in (None, False):
phases = []
def provider(request, timeout):
if request_phase(request) == 'evidence_review':
phases.append(request_phase(request))
if phases[-1] == 'evidence_review':
self.assertTrue(fixture_payload(request)['blocks'][0]['statisticalClaimsReviewRequired'])
row = {'supported': True}
if verdict is not None: row['statisticalClaimsSupported'] = verdict
@@ -597,7 +609,23 @@ class NativeAnswerTests(unittest.TestCase):
with self.subTest(phrase=phrase, verdict=verdict), patch('nao_native_answer.build_native_report', return_value=report):
result = invoke(provider)
self.assertNotIn(phrase, result['answer'])
self.assertEqual(result['compositionDiagnostics']['rejectedStatisticalBlockIndexes'], [0])
diagnostics = result['compositionDiagnostics']
if phrase == '均未达0.59 kg':
# The recognized strict boundary is now rejected
# before semantic review: 0.59 is not below itself.
self.assertFalse(result['evidenceSufficient'])
self.assertFalse(result['bodyEvidenceSufficient'])
self.assertEqual(phases, ['native_synthesis', 'native_synthesis'])
self.assertTrue(diagnostics['formatRepairAttempted'])
self.assertFalse(diagnostics['evidenceReviewAttempted'])
self.assertNotIn('rejectedStatisticalBlockIndexes', diagnostics)
self.assertEqual(len(diagnostics['issues']), 2)
self.assertTrue(all(issue.startswith('NativeCompositionError:') and '0.59' in issue
for issue in diagnostics['issues']))
else:
self.assertEqual(phases, ['native_synthesis', 'evidence_review'])
self.assertTrue(diagnostics['evidenceReviewAttempted'])
self.assertEqual(diagnostics['rejectedStatisticalBlockIndexes'], [0])
def test_strict_comparison_threshold_is_not_replaced_by_rounded_source_value(self):
for source_unit, suffix in (('kg', ' kg'), ('source_weight', '')):
@@ -636,8 +664,10 @@ class NativeAnswerTests(unittest.TestCase):
{'id': 'F3', 'text': '规格丙报废重量0.34 kg。', 'evidenceRef': 'Q00000001'}]
report = {'facts': facts, 'requiredFactIds': [], 'priorityFactIds': [], 'scopes': [], 'missingEvidence': [], 'rejectedEvidenceRefs': []}
for claim_type, statistical_support in (('hypothesis', None), ('hypothesis', False), ('observation', False)):
phases = []
def provider(request, timeout):
if request_phase(request) == 'evidence_review':
phases.append(request_phase(request))
if phases[-1] == 'evidence_review':
payload = fixture_payload(request)
self.assertEqual(payload['blocks'][0]['statisticalClaimsReviewRequired'], claim_type == 'hypothesis')
row = {'index': 0, 'supported': True}
@@ -650,7 +680,18 @@ class NativeAnswerTests(unittest.TestCase):
self.assertNotIn('其中较高', result['answer'])
self.assertNotIn('均不足', result['answer'])
self.assertIn('报废重量2.52 kg', result['answer'])
self.assertEqual(result['compositionDiagnostics']['rejectedStatisticalBlockIndexes'], [0])
# Keep the original false comparison. Its strict threshold
# fails the deterministic gate, so no semantic verdict exists.
diagnostics = result['compositionDiagnostics']
self.assertFalse(result['evidenceSufficient'])
self.assertFalse(result['bodyEvidenceSufficient'])
self.assertEqual(phases, ['native_synthesis', 'native_synthesis'])
self.assertTrue(diagnostics['formatRepairAttempted'])
self.assertFalse(diagnostics['evidenceReviewAttempted'])
self.assertNotIn('rejectedStatisticalBlockIndexes', diagnostics)
self.assertEqual(len(diagnostics['issues']), 2)
self.assertTrue(all(issue.startswith('NativeCompositionError:') and '0.59' in issue
for issue in diagnostics['issues']))
def test_history_source_overall_and_stage_baselines_are_protected_before_exploratory_rows(self):
scope = {'filters': {'department': '原调查部门'}, 'period': {'year': 2026, 'month': 9}}
@@ -676,6 +717,36 @@ class NativeAnswerTests(unittest.TestCase):
with self.assertRaisesRegex(NativeCompositionError, '单位'):
_validate_blocks({'blocks': [{'text': '{{'+key+'}} kg', 'factIds': ['F1']}]}, facts, tokens)
def test_source_weight_rounding_uses_bounded_precision_and_skips_unsafe_literals(self):
official = '123456789012345678901234567890.12345'
facts = [{'id': 'F1', 'text': '来源重量' + official + '。', 'evidenceRef': 'history:weight',
'verifiedDisplayValues': [{'metricCode': 'weight', 'unit': 'source_weight', 'displayValue': official}]}]
tokens = _binding_inventory(facts)
diagnostics = {}
accepted = _bind_literal_measures({'blocks': [{'text': '来源重量123456789012345678901234567890.1235。',
'factIds': ['F1']}]}, tokens, diagnostics)
self.assertIn('{{N', accepted['blocks'][0]['text'])
self.assertEqual(len(diagnostics['literalBindings']), 1)
rejected_diagnostics = {}
rejected = _bind_literal_measures({'blocks': [{'text': '来源重量123456789012345678901234567890.1236。',
'factIds': ['F1']}]}, tokens, rejected_diagnostics)
self.assertIn('123456789012345678901234567890.1236', rejected['blocks'][0]['text'])
self.assertFalse(rejected_diagnostics.get('literalBindings'))
oversized = '7.' + ('0' * 300) + '1'
oversized_facts = [{'id': 'F2', 'text': '来源重量' + oversized + '。', 'evidenceRef': 'history:oversized',
'verifiedDisplayValues': [{'metricCode': 'weight', 'unit': 'source_weight',
'displayValue': oversized}]}]
oversized_tokens = _binding_inventory(oversized_facts)
oversized_diagnostics = {}
oversized_result = _bind_literal_measures({'blocks': [{'text': '来源重量' + oversized + '。',
'factIds': ['F2']}]},
oversized_tokens, oversized_diagnostics)
self.assertIn(oversized, oversized_result['blocks'][0]['text'])
self.assertFalse(oversized_diagnostics.get('literalBindings'))
def test_summary_selection_rejects_unknown_and_nonpriority_core_and_normalizes_redundancy(self):
blocks = [{'text': '最低率规格甲', 'bindingFactIds': ['F1']}, {'text': '高报废规格乙', 'bindingFactIds': ['F2']}]
for choice in ({'leadIndex': True, 'evidenceIndexes': []}, {'leadIndex': 0, 'evidenceIndexes': [2]},
@@ -1226,7 +1297,7 @@ class NativeAnswerTests(unittest.TestCase):
return response({'blocks': [{'text': '本期综合合格率为' + token(payload, '95.00%')
+ ',低于基期的' + token(payload, '96.00%') + ',统计表现有所下降,需结合阶段指标核查。',
'factIds': ['F1'], 'claimType': 'inference', 'evidenceRefs': ['Q00000001']}]})
result = invoke(provider, draft='设备导致99%的下降(尚未核验的草稿)')
result = invoke(provider)
self.assertEqual(phases, ['native_synthesis', 'evidence_review'])
self.assertIn('统计表现有所下降', result['answer'])
self.assertIn('95.00%', result['readingSummary']['lead'])