上线验收报告 go/no-go 结论调研 #20
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#20
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 P0 差距。
需求
上线验收报告需要包含明确的 go/no-go 结论(如"通过率 95%,达标,建议上线")。
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/go-no-go-report.md(分支
research/go-no-go-report,commit 5827c3d)。关键发现:
现有报告不支持结论性判断:三套报告管线(Run/Campaign/IntelligentEval)均只输出原始指标数字,无 go/no-go 逻辑。
基础设施已就绪:
judged_pass_rate、pass_rate、avg_latency_ms、availability等关键指标已由引擎计算并持久化到RunSummary,go/no-go 只需在读路径上增加一层阈值比对。推荐方案:
evaluation/go_no_go.pygo(全部达标)/no_go(核心未达标)/conditional(核心达标但有风险)工作量估算:总计 8 人天(后端 5 + 前端 2.5 + 联调 0.5),MVP 阶段可压缩到 5 人天。