Beyond implementation, how do you prove that a new statistic is correct and meaningful? Especially when it’s going to become part of the ABI forever? In this c

Beyond implementation, how do you prove that a new statistic is correct and meaningful? Especially when it’s going to become part of the ABI forever? In this case, beyond the patches, we built oomd and ran the whole contraption in production on hundreds of thousands of servers. When it meaningfully improved metrics there, it became an easier thing to sell.