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:
@@ -316,14 +316,18 @@ class AnalysisVisualsTest(unittest.TestCase):
|
||||
|
||||
def test_rate_drop_is_signed_and_converted_to_percentage_points(self):
|
||||
result = changes_result(sort="rate_drop", rows=[
|
||||
{"rank": 1, "key": "p-1", "label": "规格一", "rateDrop": -0.025},
|
||||
{"rank": 2, "key": "p-2", "label": "规格二", "rateDrop": 0.04},
|
||||
{"rank": 1, "key": "p-1", "label": "规格一", "rateDrop": 0.025},
|
||||
{"rank": 2, "key": "p-2", "label": "规格二", "rateDrop": -0.04},
|
||||
])
|
||||
chart = build_visuals([evidence(result)], "画图比较产品合格率降幅", "turn-03")["visuals"][0]
|
||||
self.assertEqual(chart["metric"], "rate_drop_pp")
|
||||
self.assertEqual(chart["unit"], "pp")
|
||||
self.assertEqual(chart["series"][0]["values"], [-2.5, 4.0])
|
||||
self.assertEqual([row["key"] for row in result["rows"]], ["p-1", "p-2"])
|
||||
self.assertEqual([item["key"] for item in chart["categories"]], ["p-1", "p-2"])
|
||||
self.assertIn("率降幅线索", chart["title"])
|
||||
self.assertIn("本期减比较期", chart["note"])
|
||||
self.assertIn("仍按率降幅排序", chart["note"])
|
||||
self.assertIn("未纳入投入规模", chart["note"])
|
||||
|
||||
def test_broad_product_focus_omits_rate_only_ranking_chart(self):
|
||||
@@ -377,7 +381,9 @@ class AnalysisVisualsTest(unittest.TestCase):
|
||||
self.assertEqual(chart["requiredFields"], [
|
||||
"black_good_weight", "black_total_weight", "bright_good_weight", "bright_total_weight"
|
||||
])
|
||||
self.assertEqual(chart["series"][0]["values"], [2.0, -1.0])
|
||||
self.assertEqual(chart["series"][0]["values"], [-2.0, 1.0])
|
||||
self.assertIn("负值为下降,正值为上升", chart["note"])
|
||||
self.assertIn("本期减比较期", chart["note"])
|
||||
|
||||
def test_quality_top_uses_ranked_weights_and_never_share(self):
|
||||
result = {
|
||||
|
||||
@@ -0,0 +1,313 @@
|
||||
"""Regression tests for the isolated native-agent original-body runtime copy."""
|
||||
import copy
|
||||
import importlib.util
|
||||
import json
|
||||
import os
|
||||
import sys
|
||||
import tempfile
|
||||
import threading
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
from types import ModuleType, SimpleNamespace
|
||||
from unittest.mock import Mock, patch
|
||||
|
||||
|
||||
PROJECT = Path(__file__).resolve().parents[1]
|
||||
RUNTIME_COPY = Path(os.environ.get('NAO_ORIGINAL_BODY_PATH',
|
||||
str(PROJECT / 'scripts/nao_native_run.py'))).expanduser().resolve()
|
||||
SPEC = importlib.util.spec_from_file_location('nao_native_run_original_body', RUNTIME_COPY)
|
||||
NATIVE_RUN = importlib.util.module_from_spec(SPEC)
|
||||
sys.modules[SPEC.name] = NATIVE_RUN
|
||||
SPEC.loader.exec_module(NATIVE_RUN)
|
||||
|
||||
|
||||
IDENTITY = {'attemptId': 'attempt-current', 'sessionId': 'session-current',
|
||||
'turnId': 'turn-current', 'revision': 3, 'executionId': 'execution-current'}
|
||||
|
||||
|
||||
def proof(query_id='query-current'):
|
||||
return {**IDENTITY, 'queryId': query_id, 'nodeId': 'root', 'scopeId': 'scope-root',
|
||||
'responseHash': 'a' * 64}
|
||||
|
||||
|
||||
def document(ref='Q00000001', evidence_proof=None):
|
||||
return {'evidenceRef': ref, 'result': {'protocolEvidence': evidence_proof or proof()}}
|
||||
|
||||
|
||||
def task():
|
||||
return {'version': 1, 'goal': '解释下降并根据实际结果调查', 'question': '本期下降原因',
|
||||
'executionMode': 'native_agent', 'kind': 'analysis',
|
||||
'period': {'year': 2026, 'month': 9}, 'comparisonPeriod': {'year': 2026, 'month': 8},
|
||||
'filters': {'department': '当前产业部'}, 'fields': ['rate'], 'stages': [],
|
||||
'dimensions': [], 'dimensionStages': {}, 'visualPreference': 'auto',
|
||||
'comparisonRequired': True, 'trendRequired': False,
|
||||
'verifySource': False, 'fullDetails': False}
|
||||
|
||||
|
||||
class NativeOriginalBodyTests(unittest.TestCase):
|
||||
def setUp(self):
|
||||
self.temp = tempfile.TemporaryDirectory()
|
||||
self.addCleanup(self.temp.cleanup)
|
||||
self.actor_root = Path(self.temp.name)
|
||||
(self.actor_root / 'private').mkdir()
|
||||
(self.actor_root / 'evidence').mkdir()
|
||||
self.cancel_event = threading.Event()
|
||||
self.task = task()
|
||||
self.request = {'sessionId': IDENTITY['sessionId'], 'turnId': IDENTITY['turnId'],
|
||||
'actorId': 'actor-current', 'releaseId': 'release-current',
|
||||
'question': self.task['question'],
|
||||
'context': {'_serverHistory': [], '_serverQuestions': [], '_protocol': {}}}
|
||||
self.current_proof = proof()
|
||||
self.documents = [document(evidence_proof=self.current_proof)]
|
||||
self.snapshot = {'scopes': [{'nodeId': 'root', 'scopeId': 'scope-root', 'task': self.task}],
|
||||
'queries': [{'proof': self.current_proof}],
|
||||
'scopeCoverage': [{'nodeId': 'root', 'scopeId': 'scope-root', 'complete': True}],
|
||||
'rootGoalGaps': [], 'gaps': [], 'pending': [], 'explorationWarnings': []}
|
||||
self.accumulator = SimpleNamespace(answer='原始回答 [Q00000001]', finish_reasons=['stop'],
|
||||
tool_calls=[{'tool': 'quality_data', 'status': 'completed', 'input': {'mode': 'compare'}}],
|
||||
timeline=[{'kind': 'assistant', 'text': '原始回答'}])
|
||||
self.native_status, self.native_reason, self.stop_succeeded = 'COMPLETED', 'stop', None
|
||||
self.events = []
|
||||
self.manager = Mock()
|
||||
self.manager.ensure_instance.return_value = (None, None, 'fixture-model', 'http://native.invalid')
|
||||
self.delegated = Mock()
|
||||
self.delegated.call.side_effect = lambda *args, **kwargs: copy.deepcopy(self.snapshot)
|
||||
self.synthetic_writer = Mock(side_effect=AssertionError('native original-body path must not synthesize text'))
|
||||
|
||||
quality_runtime = ModuleType('nao_quality_runtime')
|
||||
quality_runtime._atomic_json = self.atomic_json
|
||||
quality_runtime._authenticate = Mock(return_value=object())
|
||||
quality_runtime._sse_request = self.native
|
||||
quality_runtime._read_json = self.read_json
|
||||
agent_task = ModuleType('nao_agent_task')
|
||||
agent_task.bind_task = lambda seed, current_task: {**copy.deepcopy(current_task), **copy.deepcopy(seed)}
|
||||
protocol_execution = ModuleType('nao_protocol_execution')
|
||||
protocol_execution._identity = lambda authority: copy.deepcopy(IDENTITY)
|
||||
protocol_execution._platform_visual_context = lambda visuals: {'visualState': visuals.get('visualState')}
|
||||
protocol_execution.query_evidence_bindings = lambda docs, receipts: [
|
||||
{'evidenceRef': item['evidenceRef'], 'proofs': item['result'].get('protocolEvidence')}
|
||||
for item in docs]
|
||||
evidence_finalizer = ModuleType('nao_evidence_finalizer')
|
||||
evidence_finalizer.evidence_summaries = lambda *args, **kwargs: copy.deepcopy(self.documents)
|
||||
evidence_finalizer.synthesize_native = self.synthetic_writer
|
||||
visual_evidence = ModuleType('analysis_visual_evidence')
|
||||
visual_evidence.protocol_visual_result = self.visual_result
|
||||
runtime_policy = ModuleType('nao_runtime_policy')
|
||||
runtime_policy.native_timeout = lambda: 30
|
||||
native_exploration = ModuleType('nao_native_exploration')
|
||||
native_exploration._proofs = lambda result: ([result['protocolEvidence']]
|
||||
if isinstance(result, dict) and isinstance(result.get('protocolEvidence'), dict) else [])
|
||||
self.modules = {'nao_quality_runtime': quality_runtime, 'nao_agent_task': agent_task,
|
||||
'nao_protocol_execution': protocol_execution,
|
||||
'nao_evidence_finalizer': evidence_finalizer,
|
||||
'analysis_visual_evidence': visual_evidence,
|
||||
'nao_runtime_policy': runtime_policy,
|
||||
'nao_native_exploration': native_exploration}
|
||||
self.module_patch = patch.dict(sys.modules, self.modules)
|
||||
self.module_patch.start()
|
||||
self.addCleanup(self.module_patch.stop)
|
||||
|
||||
def atomic_json(self, path, payload):
|
||||
path = Path(path)
|
||||
path.parent.mkdir(parents=True, exist_ok=True)
|
||||
path.write_text(json.dumps(payload, ensure_ascii=False), encoding='utf-8')
|
||||
|
||||
@staticmethod
|
||||
def read_json(path):
|
||||
return json.loads(Path(path).read_text(encoding='utf-8'))
|
||||
|
||||
def native(self, **kwargs):
|
||||
return self.accumulator, self.native_status, self.native_reason, self.stop_succeeded
|
||||
|
||||
def visual_result(self, documents, scopes, question, turn_id, **kwargs):
|
||||
return {'visualState': 'ready', 'visuals': [{'id': 'formal-query-chart',
|
||||
'evidenceRefs': [item['evidenceRef'] for item in documents]}], 'missingEvidence': []}
|
||||
|
||||
def invoke(self):
|
||||
return NATIVE_RUN.run_native_exploration(request=copy.deepcopy(self.request),
|
||||
task=copy.deepcopy(self.task), authority=copy.deepcopy(IDENTITY),
|
||||
actor_root=self.actor_root, delegated=self.delegated,
|
||||
connection=('https://model.invalid/v1', 'synthetic-fixture-key', 'fixture-model'),
|
||||
manager=self.manager, state={}, cancel_event=self.cancel_event,
|
||||
emit=self.events.append, prior_evidence_files={'prior-evidence.json'},
|
||||
phase_timings={'planningAndReview': 10, 'taskConfirmation': 20})
|
||||
|
||||
def test_persists_the_complete_original_body_and_keeps_formal_charts(self):
|
||||
body = '原生回答第一行\n' + ('原生正文' * 1800) + '\n[Q00000001]'
|
||||
self.accumulator.answer = body
|
||||
self.accumulator.timeline = [{'kind': 'assistant', 'text': 'x' * 8000}]
|
||||
|
||||
result = self.invoke()
|
||||
|
||||
self.assertEqual(result['status'], 'COMPLETED')
|
||||
self.assertEqual(result['answer'], body)
|
||||
self.assertIsNone(result['readingSummary'])
|
||||
self.assertEqual(result['visuals'][0]['id'], 'formal-query-chart')
|
||||
self.assertEqual(result['queryEvidenceRefs'], ['Q00000001'])
|
||||
self.assertEqual(result['conclusionProtocolQueryIds'], ['query-current'])
|
||||
self.assertEqual(result['answerQuality']['narrativeFactBinding'], 'not_assessed')
|
||||
self.assertEqual(result['answerQuality']['backgroundEvidenceQueryIds'], [])
|
||||
self.assertIsNone(result['factsComplete'])
|
||||
self.assertEqual(self.read_json(self.actor_root / 'private/native-draft.json')['answer'], body)
|
||||
self.assertEqual(self.read_json(self.actor_root / 'private/native-completion.json')['answer'], body)
|
||||
self.assertEqual(self.read_json(self.actor_root / 'private/native-original-body-diagnostics.json')
|
||||
['originalBodyCharacters'], len(body))
|
||||
self.synthetic_writer.assert_not_called()
|
||||
|
||||
def test_empty_native_body_stays_empty_and_cannot_be_completed(self):
|
||||
self.accumulator.answer = ''
|
||||
|
||||
result = self.invoke()
|
||||
|
||||
self.assertEqual(result['answer'], '')
|
||||
self.assertEqual(result['status'], 'PARTIAL')
|
||||
self.assertFalse(result['answerQuality']['narrativeComplete'])
|
||||
self.assertIn('自主分析未生成正文。', result['answerQuality']['missingEvidence'])
|
||||
self.synthetic_writer.assert_not_called()
|
||||
|
||||
def test_missing_formal_evidence_is_metadata_and_does_not_rewrite_the_body(self):
|
||||
self.documents = []
|
||||
body = '原始正文照常保留。'
|
||||
self.accumulator.answer = body
|
||||
|
||||
result = self.invoke()
|
||||
|
||||
self.assertEqual(result['answer'], body)
|
||||
self.assertEqual(result['status'], 'PARTIAL')
|
||||
self.assertFalse(result['answerQuality']['dataComplete'])
|
||||
self.assertIn('本轮未取得当前执行的可核验查询证据。', result['answerQuality']['missingEvidence'])
|
||||
self.assertNotIn('尚未解决', result['answer'])
|
||||
self.synthetic_writer.assert_not_called()
|
||||
|
||||
def test_partial_or_interrupted_generation_is_not_rewritten_or_marked_complete(self):
|
||||
self.native_status, self.native_reason = 'PARTIAL', 'length'
|
||||
self.accumulator.finish_reasons = ['length']
|
||||
body = '模型已输出的局限与正文。'
|
||||
self.accumulator.answer = body
|
||||
|
||||
result = self.invoke()
|
||||
|
||||
self.assertEqual(result['answer'], body)
|
||||
self.assertEqual(result['status'], 'PARTIAL')
|
||||
self.assertFalse(result['answerQuality']['narrativeComplete'])
|
||||
self.assertEqual(result['finishReason'], 'length')
|
||||
self.synthetic_writer.assert_not_called()
|
||||
|
||||
def test_completed_status_without_natural_stop_remains_partial(self):
|
||||
body = '原生回答保留,结束信号不足。'
|
||||
self.accumulator.answer = body
|
||||
cases = (
|
||||
('missing_stop_finish_reason', [], None),
|
||||
('stop_control_failed', ['stop'], False),
|
||||
)
|
||||
for label, finish_reasons, stop_succeeded in cases:
|
||||
with self.subTest(label=label):
|
||||
self.accumulator.finish_reasons = finish_reasons
|
||||
self.stop_succeeded = stop_succeeded
|
||||
|
||||
result = self.invoke()
|
||||
|
||||
self.assertEqual(result['answer'], body)
|
||||
self.assertEqual(result['status'], 'PARTIAL')
|
||||
self.assertFalse(result['answerQuality']['narrativeComplete'])
|
||||
self.assertEqual(result['finishReason'], 'incomplete')
|
||||
self.assertIn('自主调查未收到有效的正常结束信号。',
|
||||
result['answerQuality']['missingEvidence'])
|
||||
self.synthetic_writer.assert_not_called()
|
||||
|
||||
def test_recovered_tool_error_keeps_natural_stop_completion_without_fact_binding(self):
|
||||
self.native_status, self.native_reason = 'PARTIAL', 'mcp_error'
|
||||
body = '原生回答保留自己的局限。'
|
||||
self.accumulator.answer = body
|
||||
|
||||
result = self.invoke()
|
||||
|
||||
self.assertEqual(result['status'], 'COMPLETED')
|
||||
self.assertEqual(result['answer'], body)
|
||||
self.assertEqual(result['conclusionProtocolQueryIds'], ['query-current'])
|
||||
self.assertEqual(result['answerQuality']['backgroundEvidenceQueryIds'], ['query-current'])
|
||||
self.assertTrue(result['answerQuality']['narrativeComplete'])
|
||||
self.assertEqual(result['answerQuality']['narrativeFactBinding'], 'not_assessed')
|
||||
self.assertIsNone(result['factsComplete'])
|
||||
self.synthetic_writer.assert_not_called()
|
||||
|
||||
def test_no_explicit_refs_cannot_use_docs_without_verified_complete_scope_receipts(self):
|
||||
self.snapshot['scopeCoverage'] = None
|
||||
body = '正常原生正文不写证据编号。'
|
||||
self.accumulator.answer = body
|
||||
|
||||
result = self.invoke()
|
||||
|
||||
self.assertEqual(result['answer'], body)
|
||||
self.assertEqual(result['status'], 'PARTIAL')
|
||||
self.assertEqual(result['conclusionProtocolQueryIds'], [])
|
||||
self.assertEqual(result['answerQuality']['backgroundEvidenceQueryIds'], [])
|
||||
self.assertFalse(result['answerQuality']['dataComplete'])
|
||||
self.assertIn('本轮没有可关联的完整scope正式查询证明。',
|
||||
result['answerQuality']['missingEvidence'])
|
||||
self.synthetic_writer.assert_not_called()
|
||||
|
||||
def test_unbound_original_citation_is_a_gap_without_rewriting_the_body(self):
|
||||
body = '原生回答里保留未绑定的引用。[QABCDEF99]'
|
||||
self.accumulator.answer = body
|
||||
|
||||
result = self.invoke()
|
||||
|
||||
self.assertEqual(result['status'], 'PARTIAL')
|
||||
self.assertEqual(result['answer'], body)
|
||||
self.assertFalse(result['answerQuality']['dataComplete'])
|
||||
self.assertEqual(result['answerQuality']['unresolvedEvidenceRefs'], ['QABCDEF99'])
|
||||
self.assertIn('原文引用未关联本轮完整scope正式查询证据:QABCDEF99。',
|
||||
result['answerQuality']['missingEvidence'])
|
||||
self.assertNotIn('QABCDEF99', result['conclusionProtocolQueryIds'])
|
||||
self.synthetic_writer.assert_not_called()
|
||||
|
||||
def test_explicit_hexadecimal_query_reference_maps_only_its_receipt(self):
|
||||
hex_proof = proof('query-hex')
|
||||
self.snapshot['queries'].append({'proof': hex_proof})
|
||||
self.documents.append(document('QABCDEF12', hex_proof))
|
||||
body = '原文引用正式查询[QABCDEF12]。'
|
||||
self.accumulator.answer = body
|
||||
|
||||
result = self.invoke()
|
||||
|
||||
self.assertEqual(result['status'], 'COMPLETED')
|
||||
self.assertEqual(result['answer'], body)
|
||||
self.assertEqual(result['answerQuality']['citedEvidenceRefs'], ['QABCDEF12'])
|
||||
self.assertEqual(result['conclusionProtocolQueryIds'], ['query-hex'])
|
||||
self.synthetic_writer.assert_not_called()
|
||||
|
||||
def test_cancelled_turn_remains_cancelled_and_does_not_read_or_synthesize(self):
|
||||
for cancellation in ('event', 'native_status'):
|
||||
with self.subTest(cancellation=cancellation):
|
||||
self.cancel_event.clear()
|
||||
self.native_status, self.native_reason = 'COMPLETED', 'stop'
|
||||
if cancellation == 'event':
|
||||
self.cancel_event.set()
|
||||
else:
|
||||
self.native_status = 'CANCELLED'
|
||||
self.accumulator.answer = '中断前的草稿'
|
||||
|
||||
result = self.invoke()
|
||||
|
||||
self.assertEqual(result['status'], 'CANCELLED')
|
||||
self.assertEqual(result['answer'], '')
|
||||
self.delegated.call.assert_not_called()
|
||||
self.synthetic_writer.assert_not_called()
|
||||
|
||||
def test_server_coverage_gap_keeps_body_and_prevents_completed_status(self):
|
||||
self.snapshot['rootGoalGaps'] = ['亮片必答阶段尚无完整两期依据。']
|
||||
body = '原生结论及其局限。'
|
||||
self.accumulator.answer = body
|
||||
|
||||
result = self.invoke()
|
||||
|
||||
self.assertEqual(result['answer'], body)
|
||||
self.assertEqual(result['status'], 'PARTIAL')
|
||||
self.assertFalse(result['answerQuality']['dataComplete'])
|
||||
self.assertIn('亮片必答阶段尚无完整两期依据。', result['answerQuality']['missingEvidence'])
|
||||
self.synthetic_writer.assert_not_called()
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
unittest.main()
|
||||
@@ -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__':
|
||||
|
||||
Reference in New Issue
Block a user