import copy import json import sys import tempfile import threading import unittest from pathlib import Path from urllib.parse import parse_qs, urlsplit from unittest.mock import patch sys.path[:0] = [str(Path(__file__).resolve().parents[1] / 'scripts'), str(Path(__file__).resolve().parent)] from nao_agent_task import normalize_task from nao_protocol_execution import execute from test_nao_agent_task import task_value from test_nao_protocol_execution import BoundRankingApi, RELEASE, TURN, identity def _scope_task(root, *, purpose, fields, stages, selector_kind, product=None, dimension=None, visual_type=None, dimension_stages=None): filters = copy.deepcopy(root['filters']) if product is not None: filters['product'] = product task = {'version': 1, 'kind': 'analysis' if purpose == 'ranking' else 'metrics', 'question': root['question'], 'goal': root['goal'], 'period': copy.deepcopy(root['period']), 'comparisonPeriod': None, 'filters': filters, 'dimensions': [dimension] if purpose == 'ranking' else [], 'dimensionStages': ({dimension: list(stages)} if purpose == 'ranking' else {}), 'fields': list(fields), 'stages': list(stages), 'trendRequired': purpose == 'trend', 'comparisonRequired': False, 'visualPreference': 'auto', 'visualType': visual_type, 'periodCoverage': root.get('periodCoverage', 'available'), 'freshQuery': True, 'historyMode': 'none'} return task class QueryPlanApi(BoundRankingApi): def __init__(self, *, bad_node_proof=False, missing_second_child=False): super().__init__() self.rank_done = False self.envelopes = [] self.bad_node_proof = bad_node_proof self.missing_second_child = missing_second_child self.root = None self.plan = None def call(self, path, payload=None, method=None, *, node_id=None, scope_id=None): self.envelopes.append({'path': path, 'nodeId': node_id, 'scopeId': scope_id}) if path == '/api/analysis/protocol/queries': rank_node = next(node for node in self.plan if node['nodeId'] == 'rank') trend_node = next(node for node in self.plan if node['nodeId'] == 'top_trend') rank_scope = {'nodeId': 'rank', 'scopeId': 'rank-scope', 'label': '黑片投入重量排名', 'purpose': 'ranking', 'limit': 5, 'task': _scope_task(self.root, purpose='ranking', fields=rank_node['fields'], stages=rank_node['stages'], selector_kind='base', dimension='product', visual_type='pareto')} scopes = [rank_scope] pending = [{'nodeId': 'top_trend', 'reason': '等待正式排名结果'}] if self.rank_done: scopes = [rank_scope] for suffix, product in [('a', 'product a'), ('b', 'product b')]: scopes.append({'nodeId': 'top_trend', 'scopeId': 'trend-' + suffix, 'label': product, 'purpose': 'trend', 'task': _scope_task(self.root, purpose='trend', fields=trend_node['fields'], stages=trend_node['stages'], selector_kind='ranking_top', product=product, visual_type='line')}) if self.missing_second_child: scopes.pop() pending = [] else: pending = [] return {'version': 1, 'scopes': scopes, 'pending': pending, 'gaps': []} response = super().call(path, payload, method) if path.startswith('/api/metrics'): if isinstance(response.get('summary'), dict): response['summary'].setdefault('black', {})['scrapRate'] = 0.05 for month in response.get('trend', []): if isinstance(month, dict): month.setdefault('black', {})['scrapRate'] = 0.05 parsed = parse_qs(urlsplit(path).query) proof = response.get('protocolEvidence') if isinstance(proof, dict): proof['nodeId'] = 'untrusted' if self.bad_node_proof else node_id proof['scopeId'] = scope_id proof['queryId'] = proof['queryId'] + '-' + str(len(self.envelopes)) proof['responseHash'] = f'{len(self.envelopes):064x}' if parsed.get('view') == ['product']: self.rank_done = True return response class QueryPlanExecutionTests(unittest.TestCase): def task(self): plan = [ {'nodeId': 'rank', 'purpose': 'ranking', 'dimension': 'product', 'limit': 5, 'fields': ['totalWeight'], 'stages': ['black'], 'selector': {'kind': 'base'}, 'visualType': 'pareto', 'visualPreference': 'auto'}, {'nodeId': 'top_trend', 'purpose': 'trend', 'fields': ['scrapRate'], 'stages': ['black'], 'selector': {'kind': 'ranking_top', 'sourceNodeId': 'rank', 'stage': 'black', 'metric': 'totalWeight', 'count': 2}, 'visualType': 'line', 'visualPreference': 'auto'}, ] value = task_value(kind='analysis', queryPlan=plan, dimensions=['product'], dimensionStages={'product': ['black']}, stages=['black'], fields=['totalWeight', 'scrapRate'], trendRequired=True, comparisonRequired=False, period={'year': 2026}, filterUpdates={'department': '部门甲'}) return normalize_task(value, '按黑片投入重量排名后查看前两项报废率趋势', {'_protocol': {'version': 1}}) def run_query(self, api, *, final_answer='已核对排名和趋势'): task = self.task() api.root = task api.plan = task['queryPlan'] request = {'releaseId': RELEASE, 'turnId': TURN, 'question': task['question']} with tempfile.TemporaryDirectory() as folder: root = Path(folder); (root / 'evidence').mkdir(); (root / 'private').mkdir() with patch('nao_evidence_finalizer.synthesize', return_value={ 'answer': final_answer, 'evidenceSufficient': True, 'missingEvidence': [], 'usage': {}}) as synth: result = execute(request=request, task=task, authority={**identity(), 'task': task}, actor_root=root, delegated=api, connection=('offline', 'synthetic', 'model'), cancel_event=threading.Event(), emit=lambda _: None) documents = [json.loads(path.read_text()) for path in (root / 'evidence').glob('nao-quality-*.json')] return result, documents, synth def test_rank_then_two_server_resolved_trends_use_scoped_envelopes_and_one_finalizer(self): api = QueryPlanApi() result, documents, synth = self.run_query(api) self.assertEqual(result['status'], 'COMPLETED', result['evidenceCompletion']['missingEvidence']) self.assertEqual(result['stopReason'], 'evidence_finalized') self.assertEqual(result['completionMethod'], 'bounded_scoped_query_plan') self.assertEqual(result['finishReason'], 'stop') self.assertEqual({doc['mode'] for doc in documents}, {'rank', 'query'}) self.assertEqual(len(documents), 3) self.assertEqual(len(result['protocolEvidence']), 3) self.assertEqual({proof['nodeId'] for proof in result['protocolEvidence']}, {'rank', 'top_trend'}) self.assertEqual({proof['scopeId'] for proof in result['protocolEvidence']}, {'rank-scope', 'trend-a', 'trend-b'}) expected_agent_context = {'releaseId': RELEASE, 'filters': api.root['filters'], 'period': api.root['period']} if api.root.get('comparisonPeriod') is not None: expected_agent_context['comparisonPeriod'] = api.root['comparisonPeriod'] self.assertEqual(result['agentContext'], expected_agent_context) self.assertNotIn('_protocolScopes', result['agentContext']) self.assertNotIn('product', result['agentContext']['filters']) self.assertTrue(all(call['nodeId'] is not None and call['scopeId'] is not None for call in api.envelopes if call['path'] != '/api/analysis/protocol/queries')) self.assertEqual(len([call for call in api.envelopes if call['path'] == '/api/analysis/protocol/queries']), 3) synth.assert_called_once() context = synth.call_args.kwargs['context'] self.assertEqual(context['filters'], api.root['filters']) self.assertEqual(len(context['_protocolScopes']['scopes']), 3) def test_forged_scope_proof_and_missing_resolved_child_finish_partial(self): for api in (QueryPlanApi(bad_node_proof=True), QueryPlanApi(missing_second_child=True)): with self.subTest(api=api): result, documents, _ = self.run_query(api) self.assertEqual(result['status'], 'PARTIAL') self.assertFalse(result['evidenceCompletion']['serverCoverageSufficient']) self.assertTrue(result['evidenceCompletion']['missingEvidence']) def test_requested_scope_visual_gap_finishes_partial_even_when_text_evidence_is_complete(self): api = QueryPlanApi() with patch('analysis_visual_evidence.protocol_visual_result', return_value={ 'visualSchemaVersion': 1, 'visualState': 'ready', 'visuals': [], 'missingEvidence': ['节点趋势图未覆盖。']}): result, _, _ = self.run_query(api) self.assertEqual(result['evidenceCompletion']['serverCoverageSufficient'], True) self.assertEqual(result['status'], 'PARTIAL') self.assertIn('节点趋势图未覆盖。', result['evidenceCompletion']['missingEvidence']) if __name__ == '__main__': unittest.main()