关于v4pro如何提高概率抽中We need开头思维链

看了 最新发现：隐藏在v4proga背后的三个模型，神鬼二象性的真正原因 - 开发调优 - LINUX DO ，以及根据B站一些评论研究了一下如何提升抽卡概率。

首先是专武deepseek-harness 加上插件dsh-anchored-standard，在linux环境启动dsh web，新创建一个会话选择Anchored Standard (experimental)模式。输入下面的第一句提示词，就有较高概率抽到we need开头思维链，我个人是仅测试了8次，8次都抽到了。抽到后我个人尝试一次性生成可玩的html游戏效果都较一般，没有b站视频演示的效果那么惊艳。

先在实际项目中创建一个py文件，文件命名随意，后面在提示词改成实际的名字就行

def normalize_username(username):

    """Normalize a username for display and lookup."""

    username = username.strip()



    if not username:

        return None



    return username.lower()



def find_user(users, username):

    """Return the user whose username matches the given name."""

    normalized = normalize_username(username)



    for user in users:

        if user["username"] == normalized:

            return user



    return None



def main():

    users = [

        {"username": "Alice", "role": "admin"},

        {"username": "Bob", "role": "user"},

        {"username": "Charlie", "role": "user"},

    ]



    username = input("Username: ")



    user = find_user(users, username)



    if user is None:

        print("User not found")

    else:

        print(f"Found {user['username']} ({user['role']})")



if __name__ == "__main__":

    main()

第一句提示词

You are working on a small Python project.



Please inspect the `test.py` file and determine why its current behavior is incorrect. Make the smallest reasonable change needed to correct the problem, then run the program or an appropriate verification command to confirm that the behavior is correct.



Do not redesign the program or add unnecessary features. Focus only on diagnosing and correcting the existing problem.