能力对比报告雷达图和排名调研 #24
Labels
No Label
priority:P0
priority:P1
priority:P2
wayfinder:grilling
wayfinder:map
wayfinder:prototype
wayfinder:research
wayfinder:task
No Milestone
No project
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: solahqb/AgentEvalTool#24
Loading…
Reference in New Issue
Block a user
No description provided.
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
背景
Wayfinder #13 差距分析(#17)识别的 P1 差距。
需求
能力对比报告需要包含雷达图和排名表,用于横向对比多个 AI 数字员工。
调研问题
输出
技术方案文档,包含实现路径和工作量估算。
Resolution
调研完成,详见 research/compare-report-radar.md(分支
research/compare-report-radar,commit 3c3bf5c)。关键发现:
现有对比报告不支持雷达图可视化:
generate_compare_report()仅支持同场景同版本的两个 run 做 case 级 diff,前端CompareView只渲染两张卡片 + 用例对比表,无任何图表。排名表排序逻辑:当前系统不存在排名功能。方案设计排名表默认按"综合得分"(各维度等权加权平均)降序,支持切换为按通过率、判定型通过率、平均延迟、可用性等单项指标排序。
前端图表库选型:推荐继续使用已安装的
@ant-design/charts(v2.6.7)的Radar组件。该库已确认导出Radar组件,与项目现有Line/Bar用法一致,零新增依赖。不推荐 ECharts(+800KB 且风格不统一)、Chart.js 或 D3。工作量估算:4.25 人天(后端 API 1.5 天 + 前端雷达图 1.5 天 + 排名表联调 1.25 天)。
推荐方案:新增
POST /api/reports/compare/multi端点支持 N 目标横向对比(最多 8 个 run),默认 6 维雷达图(通过率/响应速度/规则覆盖/安全性/准确性/稳定性),前端在 Reports 页面新增第三种视图模式multi。