统一:DSH 智能分析、Top100 与图表静默恢复

接入 DSH 专用运行时和对话界面,保留模型原始回答,支持会话分页、连续追问与显式报告。
补齐组合筛选、各维度 Top100、分组查询到排名图的转换,图表故障静默省略并基于保存数据自动恢复。
优化报告风格、图表顺序和表格换行,更新接入说明及构建输入。

验证:独立目录前后端构建通过;118 项 Java、102 项前端、102 项图表、14 项 DSH 应用和 13 项组合筛选检查通过。
仅提交相关源码、锁定依赖与测试;不包含凭据、会话、业务数据或本地产物。
This commit is contained in:
2026-11-12 11:57:41 +08:00
parent cc8832c8e5
commit f1a2fbd494
166 changed files with 26281 additions and 2248 deletions
+205 -4
View File
@@ -8,6 +8,7 @@ from __future__ import annotations
import json
import re
from decimal import Decimal, InvalidOperation
from nao_period import PERIOD_KEYS, expected_months, normalize_period
@@ -628,6 +629,104 @@ def _investigation_docs(evidence, context):
return selected, task, history_all, periods, stages, expected_filters
def _source_coverage_complete(coverage, total_rows):
if not isinstance(coverage, dict) or type(total_rows) is not int or total_rows <= 0:
return False
if coverage.get('rows') != total_rows or coverage.get('numericRows') != total_rows:
return False
return all(type(coverage.get(key)) is int and coverage.get(key) == 0
for key in ('keyAbsentRows', 'nullOrBlankRows', 'invalidNumericRows', 'negativeRows'))
def _source_category_fields(category):
rows = []
for item in category.get('fields') or []:
if isinstance(item, dict) and isinstance(item.get('coverage'), dict):
rows.append({'field': item.get('field'), **item['coverage']})
for item in category.get('fieldCoverageLimits') or []:
if isinstance(item, dict):
rows.append(dict(item))
columns, values = category.get('fieldCoverageColumns'), category.get('fieldCoverage')
if isinstance(columns, list) and isinstance(values, list):
for row in values:
if isinstance(row, list) and len(row) == len(columns):
rows.append(dict(zip(columns, row)))
unique = {}
for row in rows:
field = row.get('field')
if isinstance(field, str) and field:
unique[field] = row
return list(unique.values())
def _source_category_reconciliation(category, *, total_rows, selected_fields, source_proof,
stage, period, filters, release_id):
"""Return a safe, scope-bound summary of formal/source category reconciliation."""
category_name = category.get('category')
formal_coverage = category.get('formalCoverage')
raw_coverage = category.get('rawTotalCoverage')
fields = _source_category_fields(category)
reasons = []
proof_complete = (isinstance(source_proof, dict)
and source_proof.get('complete') is True
and source_proof.get('allSelectedValuesMatchStored') is True
and source_proof.get('expectedRows') == total_rows
and source_proof.get('verifiedRows') == total_rows
and source_proof.get('missingRows') == 0
and source_proof.get('mismatchRows') == 0)
if not proof_complete:
reasons.append('原表所选字段与已存储原始值尚未完整匹配')
if not _source_coverage_complete(raw_coverage, total_rows):
reasons.append('原表对应类别总量字段缺失或覆盖不完整')
if not _source_coverage_complete(formal_coverage, total_rows):
reasons.append('正式类别分量覆盖不完整')
field_names = {row.get('field') for row in fields if isinstance(row.get('field'), str)}
if category.get('noFineFields') is True or not fields:
reasons.append('没有可用的来源细项,无法核对分项合计')
else:
if not field_names.issubset(selected_fields):
reasons.append('本次原表核验未覆盖全部来源细项')
if any(not _source_coverage_complete(row, total_rows) for row in fields):
reasons.append('来源细项存在缺失或无效值,覆盖不完整')
if type(category.get('rawFormalMismatchRows')) is not int:
reasons.append('缺少原表类别总量与正式分量的逐行对账结果')
elif category['rawFormalMismatchRows'] != 0:
reasons.append('原表类别总量与正式分量存在逐行差异')
comparable = category.get('comparableRows')
missing_formal = category.get('missingOrInvalidFormalRows')
if comparable != total_rows or missing_formal != 0:
reasons.append('正式分量与来源细项未覆盖全部对应行')
difference = category.get('differenceFormalMinusSubitems')
try:
delta = Decimal(str(difference)) if difference is not None else None
except (InvalidOperation, TypeError, ValueError):
delta = None
if delta is None:
reasons.append('正式分量与来源细项合计无法完整对账')
elif abs(delta) > Decimal('0.000001'):
reasons.append('正式分量与来源细项合计存在差异')
if category.get('requiresFurtherInvestigation') is not False:
reasons.append('服务端标记仍需进一步核查')
reasons = list(dict.fromkeys(reasons))
reconciled = not reasons
return {
'category': category_name,
'status': 'RECONCILED' if reconciled else 'UNRESOLVED',
'reconciled': reconciled,
'releaseId': release_id,
'stage': stage,
'period': dict(period) if isinstance(period, dict) else {},
'filters': _filters(filters),
'reasons': reasons,
}
def _append_investigation_report(report, evidence, context):
documents, task, history_all, expected_periods, expected_stages, expected_filters = _investigation_docs(evidence, context)
if not documents:
@@ -675,15 +774,117 @@ def _append_investigation_report(report, evidence, context):
and type(verified) is int and verified == total_rows
and type(proof.get('missingRows')) is int and proof.get('missingRows') == 0
and type(proof.get('mismatchRows')) is int and proof.get('mismatchRows') == 0)
status = '原表值全部匹配且核验完整' if complete else '原表核验未完整通过'
status = '所选原表值与已存储原始值全部匹配且核验完整' if complete else '原表核验未完整通过'
count_text = f'已核验{verified}行/应核验{expected}行' if type(verified) is int and type(expected) is int else '核验行数未提供'
mismatch_text = ';'.join(
f'{label}{proof[name]}条' if type(proof.get(name)) is int else f'{label}数量未提供'
for name, label in (('missingRows', '未核验'), ('mismatchRows', '不一致')))
fact_text = f'{side_period};{stage_label};{status},{count_text};{mismatch_text}。'
proof_boundary = '该核验只证明所选原表值与已存储原始值一致,不证明正式类别与来源细项已对账。'
fact_text = f'{side_period};{stage_label};{status},{count_text};{mismatch_text};{proof_boundary}'
source_count_values = []
for count, metric in ((verified, 'sourceVerifiedRows'), (expected, 'sourceExpectedRows'),
(proof.get('missingRows'), 'sourceMissingRows'),
(proof.get('mismatchRows'), 'sourceMismatchRows')):
if type(count) is int and count >= 0:
source_count_values.append({'rawValue': str(count), 'displayValue': str(count),
'unit': 'metadata_count', 'metricField': metric, 'stageHint': stage_label,
'period': selected_period, 'sourceVerificationCount': True})
rows.append([side_period, stage_label, '原表核验', status + ';' + count_text + ';' + mismatch_text, reference])
facts.append({'kind': 'source', 'label': stage_label + '原表核验',
'text': fact_text, 'evidenceRef': reference})
source_fact = {'kind': 'source', 'label': stage_label + '原表核验',
'text': fact_text, 'evidenceRef': reference,
'_nativeSourceValues': source_count_values}
facts.append(source_fact)
# Surface only authorized aggregate coverage and official
# category components. Raw rows, workbook values, filenames,
# and mismatch examples never enter the narration facts.
selected_fields = side.get('selectedFields')
selected_fields = {value for value in selected_fields
if isinstance(value, str) and value} if isinstance(selected_fields, list) else set()
categories = side.get('categories') if isinstance(side.get('categories'), list) else []
category_names = [str(item.get('category')) for item in categories
if isinstance(item, dict) and isinstance(item.get('category'), str)]
category_relations = []
for category in categories:
if not isinstance(category, dict) or not isinstance(category.get('category'), str):
continue
category_name = category['category']
relation = _source_category_reconciliation(category, total_rows=total_rows,
selected_fields=selected_fields, source_proof=proof, stage=stage,
period=selected_period, filters=result.get('filters'),
release_id=result.get('releaseId'))
category_relations.append(relation)
if relation['reconciled']:
reconciliation_text = (f'{side_period};{stage_label};{category_name}正式类别分量、原表类别总量与来源细项已完整对账一致。')
else:
reasons = ';'.join(relation['reasons'])
reconciliation_text = (f'{side_period};{stage_label};{category_name}正式类别分量与原表类别总量/来源细项未完整对账;'
f'当前卡点:{reasons}。')
rows.append([side_period, stage_label, category_name + '类别对账',
'已完整对账一致' if relation['reconciled'] else '未完整对账:' + reasons,
reference])
facts.append({'kind': 'source_reconciliation', 'label': category_name + '类别对账',
'text': reconciliation_text, 'evidenceRef': reference,
'_nativeSourceCategoryReconciliation': relation})
formal_weight = category.get('formalWeight')
formal_coverage = category.get('formalCoverage') if isinstance(category.get('formalCoverage'), dict) else {}
numeric_rows = formal_coverage.get('numericRows')
coverage_rows = formal_coverage.get('rows')
complete_component = (type(total_rows) is int and total_rows > 0
and coverage_rows == total_rows and type(numeric_rows) is int and numeric_rows == total_rows
and all(type(formal_coverage.get(key)) is int and formal_coverage.get(key) == 0
for key in ('keyAbsentRows', 'nullOrBlankRows', 'invalidNumericRows', 'negativeRows')))
from nao_evidence_report import number
formal_value = number(formal_weight)
if complete_component and formal_value is not None and formal_value >= 0:
facts.append({'kind': 'category', 'label': category_name,
'text': f'{side_period};{stage_label}正式大类分量:{category_name}为{shown(formal_value)} kg。',
'evidenceRef': reference,
'_nativeSourceValues': [{'rawValue': str(formal_value),
'displayValue': shown(formal_value), 'unit': 'kg',
'metricField': 'categoryWeight', 'stageHint': stage_label,
'categoryHint': category_name, 'categoryNames': category_names,
'period': selected_period, 'releaseId': result.get('releaseId')}],
'_nativeSourceCategoryReconciliation': relation})
# These field names are emitted only when the server says
# they were selected for this source check. Missing or bad
# values remain coverage gaps; they are never represented
# as zero weight.
field_rows = []
for item in category.get('fields') or []:
if isinstance(item, dict) and isinstance(item.get('coverage'), dict):
field_rows.append((item.get('field'), item['coverage']))
for item in category.get('fieldCoverageLimits') or []:
if isinstance(item, dict):
field_rows.append((item.get('field'), item))
columns, values = category.get('fieldCoverageColumns'), category.get('fieldCoverage')
if isinstance(columns, list) and isinstance(values, list):
for row in values:
if isinstance(row, list) and len(row) == len(columns):
mapped = dict(zip(columns, row))
field_rows.append((mapped.get('field'), mapped))
deduped = {}
for field_name, coverage in field_rows:
if isinstance(field_name, str) and field_name in selected_fields and isinstance(coverage, dict):
deduped[field_name] = coverage
for field_name, coverage in deduped.items():
issues = []
for key, label in (('keyAbsentRows', '缺列'), ('nullOrBlankRows', '空值'),
('invalidNumericRows', '非数值'), ('negativeRows', '负值')):
count = coverage.get(key)
if type(count) is int and count > 0:
issues.append(label)
rows_with_coverage = coverage.get('rows')
numeric_count = coverage.get('numericRows')
if (type(rows_with_coverage) is int and type(numeric_count) is int
and numeric_count < rows_with_coverage and not issues):
issues.append('有效数值覆盖不足')
if issues:
fact_text = (f'{side_period};{stage_label}来源字段“{field_name}”存在'
+ '、'.join(issues) + ';缺失或非法覆盖不能按零处理。')
facts.append({'kind': 'source_coverage', 'label': field_name,
'text': fact_text, 'evidenceRef': reference})
source_fact['_nativeSourceCategoryReconciliations'] = category_relations
else:
matching = side.get('matchingRows')
returned = side.get('returnedRows')