Start here: share your first Code-role milestone loop / 从第一个里程碑开始

[Magicray1217]

Code-role v0.3.0 is live with two supported operating profiles: the four-workstation Minimal Profile and the eight-role Full Profile.

The most useful contribution is evidence from one real software milestone. Please share a sanitized report using this structure:

Project type:
Profile used: Minimal | Full
Objective:
Binary KR:
Where an AI role previously guessed or drifted:
What Code-role made clearer:
Independent evidence collected:
Final KR result: 0 | 1
Smallest workflow improvement still needed:

Do not include private code, credentials, or customer data. A failed loop is valuable evidence when the failure point is precise.

Code-role v0.3.0 已发布，四工位最小版和八角色完整版都是正式支持的配置。

最有价值的贡献不是泛泛评价，而是一个真实软件里程碑的脱敏证据。请按上面的结构分享：目标是什么、KR 如何二值验收、角色在哪里发生过猜测、收集了什么独立证据，以及最终 KR 是 0 还是 1。

完整示例：https://github.com/Deepleaper/Code-role/tree/main/examples/minimal-goal-loop

[Magicray1217]

v0.3.1: two real project cases

The repository now includes two sanitized case studies:

• DeepBrain: strong evidence is not milestone completion
• Leaper Agent: an evaluation plan is not an executable baseline

DeepBrain shows why 1,750 passing unit tests and strong benchmark slices still cannot close a product milestone when fair comparison, clean reproduction, repair proof, and cost/SLO evidence remain missing.

Leaper Agent shows why a professional-looking evaluation document must be rejected when its task artifacts, holdout boundary, grader mechanism, and runtime conditions are not executable yet.

Neither case is presented as complete. That is the point: Code-role preserves real progress without turning it into an unsupported claim.

中文双案例文章：两个真实项目告诉我：AI 编程最缺的不是更多 Agent，而是里程碑控制