Files
Agent-Workbench/docs/engineering-evals-de-evidence/0654-summary.mjs
T

9 lines
995 B
JavaScript

export function evaluationSummary(batch) {
const outcomes = batch.samples.map(sample => sample.report?.summary ?? { original: sample.baselineAttemptedAt ? 'unknown' : 'pending', candidate: ['failed','stopped','unknown'].includes(sample.status) ? sample.status : 'pending' });
const count = (role, status) => outcomes.filter(item => item[role] === status).length;
return { samples: batch.samples.length, finished: batch.samples.filter(item => !!item.report).length,
original: Object.fromEntries(['passed', 'failed', 'stopped', 'unknown', 'pending'].map(status => [status, count('original', status)])),
candidate: Object.fromEntries(['passed', 'failed', 'stopped', 'unknown', 'pending'].map(status => [status, count('candidate', status)])),
inferenceValidated: false, scope: batch.evidenceMode === 'local-fixture' ? '固定响应仅验证执行链和工程检查,不计模型自主能力。' : '保留当前固定样本的实际结果,不外推任意工程成功率。' };
}