在 gittensor 上提交了几个月的补丁，说点不太好听的观察。 按合并的 PR 计奖励，方向没有错，但它把成本转移到了维护者身上。真正难的从来不是写代码，是判断一个改动值不值得进主干。现在的结果是低价值的提交变多了，改个拼写、调个格式、把一个函数拆成两个，评审的人要一条条看过去。 我不觉得这是贡献者的问题，这是指标的问题。如果衡量的是被接受的数量，那理性的做法就是提交最容易被接受的东西。要改的话，权重得跟改动的影响挂钩，比如有没有关联的 issue、有没有补测试、有没有人在生产里真的用到。这些更难量化，但至少方向是对的。
goodhart with extra steps. any contribution metric that a maintainer has to arbitrate eventually becomes the maintainer's unpaid job
requiring a linked issue is the cheapest fix imo. doesnt catch everything but it kills the drive by typo pr straight away since you have to read the tracker before you can even file it
Careful with that one though, the issue can be opened by the same person. I have watched a contributor file a tidy little issue and close it with their own PR twenty minutes later. If you go that route the issue probably needs to be triaged or labelled by someone else before it counts toward anything.
这个补充很到位，自问自答的情况我已经见过一次了。或许可以要求 issue 至少有第三方回复，或者被维护者打过标签再计入。门槛不高，但能筛掉大部分刷量的提交。
