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
+147 -17
View File
@@ -17,18 +17,115 @@ def _ranking_shape(task):
and isinstance(task.get('dimensions'), list) and task['dimensions']
and set(task['dimensions']) <= set(DIMENSIONS)
and isinstance(task.get('fields'), list) and task['fields']
and set(task['fields']) <= set(FIELDS)
and set(task['fields']) <= (set(FIELDS) | {'rate'})
and task.get('visualType') in ('pareto','bar','column',None)):
return False
# A rank requires a concrete black/bright stage for every requested
# dimension. Empty dimensionStages denotes a combined view and cannot be
# silently converted into two stage-specific rankings.
# Empty dimensionStages denotes one combined scrap ranking. Never turn it
# into two independent stage Top lists or combine stage input denominators.
from nao_agent_task import dimension_stages
return all((stages := dimension_stages(task, dimension, combined=False))
and all(stage in ('black', 'bright') for stage in stages)
and (stages == [None] and set(task['fields']) <= {'rate', 'scrapWeight'} and 'scrapWeight' in task['fields']
or all(stage in ('black', 'bright') for stage in stages) and set(task['fields']) <= set(FIELDS))
for dimension in task['dimensions'])
def _single_filter_value(value):
"""Match the server SelectionValues wire contract, including encoded sets."""
if not isinstance(value, str) or not value.strip():
return False
if not value.startswith('~qv1~'):
return len(value) <= 256
import base64
import json
import re
encoded = value[5:]
if len(value) > 16384 or not re.fullmatch(r'[A-Za-z0-9_-]+', encoded):
return False
try:
values = json.loads(base64.b64decode(encoded + '=' * ((-len(encoded)) % 4),
altchars=b'-_', validate=True))
except (ValueError, TypeError, UnicodeError):
return False
return (isinstance(values, list) and 1 <= len(values) <= 50
and all(isinstance(item, str) and item.strip() and len(item) <= 256 for item in values)
and len(set(values)) == 1)
def protocol_scalar_metrics_scope(root_task, context, document):
"""Confirm that a combined ranking result is serving one exact scalar metrics goal.
The protocol scope itself may be an `analysis` task because its operator is
a combined ranking. Presentation can use the root `metrics` intent only
when the single declared node, child frame, evidence proof, and result all
match the same exact scope.
"""
if not isinstance(root_task, dict) or root_task.get('kind') != 'metrics':
return False
dimension = root_task.get('dimensions')
if not isinstance(dimension, list) or len(dimension) != 1 or dimension[0] not in DIMENSIONS:
return False
dimension = dimension[0]
filters = root_task.get('filters')
if not isinstance(filters, dict) or not _single_filter_value(filters.get(dimension)):
return False
if (set(root_task.get('fields') or []) != {'rate', 'scrapWeight'}
or root_task.get('stages') != []
or root_task.get('dimensionStages') != {dimension: []}
or root_task.get('comparisonRequired') is True
or root_task.get('comparisonPeriod') is not None
or root_task.get('trendRequired') is True):
return False
nodes = root_task.get('queryPlan')
if not isinstance(nodes, list) or len(nodes) != 1:
return False
node = nodes[0]
if (not isinstance(node, dict) or node.get('purpose') != 'ranking'
or node.get('dimension') != dimension or node.get('limit') != 1
or set(node.get('fields') or []) != {'rate', 'scrapWeight'}
or node.get('stages') != [] or node.get('comparisonRequired') is True
or node.get('selector') != {'kind': 'base'}):
return False
if not isinstance(context, dict) or not isinstance(document, dict):
return False
result = document.get('result')
if (document.get('mode') != 'rank' or not isinstance(result, dict)
or result.get('kind') != 'quality_weight_ranking'
or result.get('dimension') != dimension or result.get('stage') is not None
or result.get('metric') != 'scrapWeight'):
return False
state = context.get('_protocolScopes')
scopes = state.get('scopes') if isinstance(state, dict) else None
if (not isinstance(scopes, list) or len(scopes) != 1
or state.get('pending') or state.get('gaps')):
return False
scope = scopes[0]
if (not isinstance(scope, dict) or scope.get('purpose') != 'ranking'
or scope.get('nodeId') != node.get('nodeId')):
return False
child = scope.get('task')
if (not isinstance(child, dict) or child.get('kind') not in ('analysis', 'metrics')
or child.get('fields') is None or set(child.get('fields') or []) != {'rate', 'scrapWeight'}
or child.get('dimensions') != [dimension] or child.get('stages') != []
or child.get('dimensionStages') != {dimension: []}
or child.get('comparisonRequired') is True or child.get('comparisonPeriod') is not None
or child.get('trendRequired') is True):
return False
from nao_evidence_report import _filters
from nao_period import normalize_period
try:
if (_filters(child.get('filters')) != _filters(filters)
or normalize_period(child.get('period')) != normalize_period(root_task.get('period'))):
return False
except (TypeError, ValueError):
return False
from nao_query_plan import scope_documents
try:
selected = scope_documents([document], scope, release_id=context.get('releaseId'))
except (KeyError, TypeError, ValueError):
return False
return len(selected) == 1 and selected[0] is document
def ranking_intent(task):
"""A structurally valid ranking request, including incompatible trend/compare asks."""
return _ranking_shape(task)
@@ -50,7 +147,18 @@ def source_period(coverage, period, stage):
'complete':not missing,'label':str(period['year'])+'年'+suffix+'已发布数据'}
def rank_response(response, release, period, filters, dimension, stage, metric, limit):
def empty_scrap_ranking(result):
return (result.get('metric') == 'scrapWeight' and result.get('totalWeight') == 0
and result.get('rows') == [] and result.get('totalRows') == 0
and result.get('emptyReason') == 'no_positive_scrap'
and (result.get('rankingCoverage') or {}).get('completeRanking') is True
and bool((result.get('periodCoverage') or {}).get('availableMonths')))
def rank_response(response, release, period, filters, dimension, stage, metric, limit, *, _allow_zero=False):
if stage is None and metric == 'scrapWeight':
from nao_combined_ranking import combined_response
return combined_response(response, release, period, filters, dimension, limit)
from nao_quality_tools import _number, QualityToolError
rows = response.get('rows')
if not isinstance(rows,list) or len(rows)>100000:
@@ -58,7 +166,7 @@ def rank_response(response, release, period, filters, dimension, stage, metric,
if (response.get('context') or {}).get('rowCount',len(rows)) != len(rows):
raise QualityToolError('分组数据已被截断,不能以其生成全量排名。')
denominator = _number((response.get('summary',{}).get(stage) or {}).get(metric),minimum=Decimal(0))
if denominator is None or denominator<=0:
if denominator is None or denominator<0 or denominator==0 and not (_allow_zero or metric == 'scrapWeight'):
raise QualityToolError('所选阶段的正式重量总量不可用,不能计算占比。')
allow_signed_black_total = stage == 'black' and metric == 'totalWeight'
all_rows=[]
@@ -106,6 +214,8 @@ def rank_response(response, release, period, filters, dimension, stage, metric,
'basis':'先读取同范围全部正式分组值,再按所选阶段重量降序选Top;占比和累计占比的分母均为同范围全量正式重量,Top末项不强制为100%。'}
if isinstance(response.get('protocolEvidence'), dict):
result['protocolEvidence'] = response['protocolEvidence']
if metric == 'scrapWeight' and denominator == 0 and result['periodCoverage']['availableMonths']:
result['emptyReason'] = 'no_positive_scrap'
return result
@@ -123,18 +233,20 @@ def ranking_gaps(evidence,task,context):
for dimension in task['dimensions']:
for stage in dimension_stages(task, dimension, combined=False):
for metric in task['fields']:
if stage is None and metric == 'rate':
continue # Companion formal rate, not an independently sorted ranking.
candidates=[d['result'] for d in docs if d['result'].get('dimension')==dimension
and d['result'].get('stage')==stage and d['result'].get('metric')==metric]
def valid(result):
proof=result.get('rankingCoverage',{});period=source_period(result.get('coverage'),result['period'],stage)
proof=result.get('rankingCoverage',{});period=(result.get('periodCoverage') or {}) if stage is None else source_period(result.get('coverage'),result['period'],stage)
rows=result.get('rows');denominator=result.get('totalWeight')
return (proof.get('completeRanking') is True and isinstance(rows,list) and bool(rows)
return (proof.get('completeRanking') is True and isinstance(rows,list)
and isinstance(denominator,(int,float)) and not isinstance(denominator,bool)
and math.isfinite(denominator) and denominator>0
and math.isfinite(denominator) and (bool(rows) and denominator>0 or empty_scrap_ranking(result))
and period['availableMonths'] and not period['internalMissingMonths']
and (task.get('periodCoverage','available')!='complete' or period['complete']))
if not any(valid(r) for r in candidates):
gaps.append('未取得'+DIMENSIONS[dimension]+'级'+('黑片' if stage=='black' else '亮片')+FIELDS[metric]+'完整排名及同范围全量分母。')
gaps.append('未取得'+DIMENSIONS[dimension]+'级'+('黑片' if stage=='black' else '亮片' if stage=='bright' else '黑片+亮片')+FIELDS[metric]+'完整排名及同范围全量分母。')
return gaps
@@ -153,10 +265,15 @@ def _negative_correction_chart_limit(result):
def ranking_chart(document,turn_id,task):
if empty_scrap_ranking(document.get('result', {})):
return None
from nao_agent_task import dimension_stages
from analysis_visuals import _chart,_chart_scope,_scope,_period
r=document.get('result',{})
if r.get('kind')!='quality_weight_ranking' or r.get('dimension') not in DIMENSIONS:return None
if r.get('stage') is None:
from nao_combined_ranking import combined_chart
return combined_chart(document, turn_id, task)
if r.get('stage') not in dimension_stages(task, r['dimension'], combined=False) or r.get('metric') not in task['fields']:return None
check={**task,'dimensions':[r['dimension']],'stages':[r['stage']],'fields':[r['metric']],
'dimensionStages':{r['dimension']:[r['stage']]}}
@@ -179,7 +296,7 @@ def ranking_chart(document,turn_id,task):
return _chart(turn_id=turn_id,chart_type=chart_type,
title=DIMENSIONS[r['dimension']]+'级'+label+('柏拉图' if chart_type=='pareto' else '排名'),
subtitle=source['label']+';展示前'+str(shown_count)+'项,占比以同范围全部重量为分母。',
unit='kg',categories=[{'key':'item-'+str(i+1),'label':row['label']} for i,row in enumerate(rows)],
unit='kg',categories=[{'key':row['key'],'label':row['label']} for row in rows],
series=[{'key':metric,'label':label,'values':weights}],evidence_refs=[document['evidenceRef']],
required_fields=[metric],scope=_chart_scope(scope,r['releaseId'],r['period'],view=r['dimension'],stage=r['stage']),
period=period,stage=r['stage'],metric=metric,
@@ -195,10 +312,15 @@ def missing_ranking_visuals(task,visuals,evidence=None):
for dimension in task['dimensions']:
for stage in dimension_stages(task, dimension, combined=False):
for field in task['fields']:
metric=stage+('_total_weight' if field=='totalWeight' else '_scrap_weight')
if stage is None and field == 'rate':continue
if any(empty_scrap_ranking(doc['result']) and doc['result'].get('dimension') == dimension
and doc['result'].get('stage') == stage and doc['result'].get('metric') == field
for doc in matching_rankings(evidence or [], task, {})):
continue
metric='current_scrap_weight' if stage is None else stage+('_total_weight' if field=='totalWeight' else '_scrap_weight')
if not any(v.get('dimension')==dimension and v.get('stage')==stage
and v.get('metric')==metric and v.get('type')==(task.get('visualType') or 'bar') for v in visuals):
label=DIMENSIONS[dimension]+'级'+('黑片' if stage=='black' else '亮片')+FIELDS[field]+'图'
label=DIMENSIONS[dimension]+'级'+('黑片' if stage=='black' else '亮片' if stage=='bright' else '黑片+亮片')+FIELDS[field]+'图'
if stage=='black' and field=='totalWeight':
limited=any(_negative_correction_chart_limit(d.get('result',{}))
and d.get('result',{}).get('dimension')==dimension
@@ -210,14 +332,22 @@ def missing_ranking_visuals(task,visuals,evidence=None):
def ranking_report(evidence,task,context):
from nao_evidence_report import shown,cell,_scope_label
from nao_agent_task import dimension_stages
if task.get('dimensions') and all(dimension_stages(task, dimension, combined=False) == [None] for dimension in task['dimensions']):
from nao_combined_ranking import combined_report
return combined_report(evidence, task, context)
from nao_evidence_report import shown,cell,_scope_label
facts=[];sections=[];seen=set()
for document in matching_rankings(evidence,task,context):
r=document['result'];key=(r['dimension'],r['stage'],r['metric'])
if key in seen or r['dimension'] not in task['dimensions'] or r['stage'] not in dimension_stages(task, r['dimension'], combined=False) or r['metric'] not in task['fields']:continue
seen.add(key);ref=document['evidenceRef'];name=('黑片' if r['stage']=='black' else '亮片')+FIELDS[r['metric']]
p=source_period(r.get('coverage'),r['period'],r['stage']);scope=_scope_label(r['filters']);top=r['rows']
if empty_scrap_ranking(r):
text=p['label']+';'+scope+';'+name+'为0.00 kg,没有正报废重量的'+DIMENSIONS[r['dimension']]+',无需生成排名或占比图。'
facts.append({'id':'F'+str(len(facts)+1),'kind':'overall','label':name,'text':text,'evidenceRef':ref})
sections.append('### '+name+'\n\n'+text+'\n\n['+ref+']')
continue
title=DIMENSIONS[r['dimension']]+'级'+name+' Top '+str(len(top))
table=['| 排名 | '+DIMENSIONS[r['dimension']]+' | '+name+'(kg) | 占全部重量 | 累计占全部重量 |','| --- | --- | --- | --- | --- |']
for row in top:
@@ -230,10 +360,10 @@ def ranking_report(evidence,task,context):
'排名表只展示正重量Top。')
facts.append({'id':'F'+str(len(facts)+1),'kind':'overall','label':title,
'text':p['label']+';'+scope+';'+DIMENSIONS[r['dimension']]+'级'+name+'全量分母为'+shown(r['totalWeight'])+' kg,展示前'+str(len(top))+'项,累计占全部重量'+shown(top[-1]['cumulativeShare'],percent=True)+'。'+correction_note,'evidenceRef':ref})
for row in top[:3]:
for row in top:
facts.append({'id':'F'+str(len(facts)+1),'kind':r['dimension'],'label':row['label'],
'text':DIMENSIONS[r['dimension']]+':'+row['label']+';'+name+shown(row['weight'])+' kg,占全部重量'+shown(row['share'],percent=True)+'。','evidenceRef':ref,
'dimension':r['dimension'],'stage':r['stage']})
'dimension':r['dimension'],'stage':r['stage'],'rank':row['rank']})
sections.append('### '+title+'\n\n'+p['label']+';'+scope+'。分母:全部'+name+shown(r['totalWeight'])+' kg。\n\n'+'\n'.join(table)
+('\n\n'+correction_note if correction_note else '')+'\n\n['+ref+']')
if not facts:raise ValueError('未取得与本轮阶段、期间和分组一致的重量排名')