Skip to content

Experiment — Project Timebox ​

Hypothesis ​

将工作流网页优化限制为每日最多 90 分钟,可能降低项目任务时间超支,并减少对学习任务的挤占。

Intervention ​

项目任务每日最多 90 分钟;遇到技术问题时停止继续扩展工作时间,记录问题并另行安排。

Metric ​

核心指标为 actual_planned_ratio。 辅助观察学习任务完成数是否下降。

Baseline ​

2.09(来自 2026-W36 Test Review)。

Success Criteria ​

下一周期项目任务的 actual/planned 比例明显低于 2.09,且学习任务完成情况不恶化。

Failure Criteria ​

项目推进明显受阻,或学习任务完成情况没有改善,同时项目延期显著增加。

Rollback ​

若实验失败,恢复原项目时间安排方式,并保留实验记录。

Evidence Source ​

明确引用 source_review:WeeklyReview/Reviews/2026-W36-test-review.md(其 Facts 记录工作流网页优化实际 690 min vs 计划 330 min,超支 2.09 倍;Highest-Leverage Problem 与 Proposed Experiments 中 E1 为本实验来源)。

Approval Policy ​

  • This experiment is currently only a proposal.
  • DeepSeek Harness must not change status from proposed to approved unless the user explicitly approves this exact experiment.
  • Absence of rejection does not count as approval.
  • A recommendation from the model does not count as approval.
  • When explicit approval is received, record approved_by: user and the approval date/time before changing status.
  • The experiment must remain proposed until that explicit approval occurs.