This episode is a very good example of a recurring and difficult problem. There's a substantial economics literature on it, under the general category of "incentive compatibility". The Soviet "planning" system seems to have been seriously impeded by lack of incentive compatibility.
I was never able to muster much enthusiasm for the management POV of an organization. From the bottom-up, these tokenmaxxing orgs may just be decent companies to work for, provided the engineer can somehow make sure they're happy delivering, while working around the malfunctioning aspects of the company.
That's understandable. Taking on someone else's POV is a challenge. But, if you want change though, you have to do it. Or at least you should. If you don't understand their POV, your using force rather than persuasion. Force has it's own problems. At the very least, there's the real chance you don't have enough of it to reach your goals. But using it also has consequences.
One of the blocks in understanding management is often in looking at yourself only. I was a sort of annoying student in high school. Dedicated and smart, but I have my own mission, and the teachers weren't always serving it. I got thrown out of Chemistry classes for reading Chemistry books. I thought it was pretty unreasonable at the time. If the teacher wanted to teach the rest of the class, but I already knew this topic, why not study the next one?
Problem is, some students actually do need a teacher motivating them to stay engaged and they didn't know how to adjust their plan for that to my uniqueness. I don't think the solution was for me to just shut-up and go along, but if I'd thought about it more, maybe there was another option.
I think emotionally, it may be hardest to take the POV of someone with power. Maybe because it feels like it should be their responsibility to take yours? ("With great power comes great responsibility"). But it's a pretty bad tactical move to not try, which came up in a different topic for me recently: https://substack.com/@norabble/note/c-295769613
Understanding their POV helps regardless of which way you go. If you want to outmaneuver them, it helps there. If you want to work toward collaboration, absolutely critical.
The rollup section is the sharpest part — "you've asked people to police a number that pays them to look away" names the structural failure most Goodhart discussions skip: the gaming is downstream of the wiring, not the people. And the stress-test framing earns its keep — the leaderboard didn't make the crack, it loaded it until it showed. Have you ever seen a metric survive being tied directly to standing, or is the number-reward gap the one non-negotiable?
Survival is relative here. But yes, there's some survival. What the original metric is, how gameable it is, and how directly aligned with objectives it is, is the first aspect. The most accessible examples are generally creating something physical and non-complex. The number of screws made per day for example.
Screws made is comparatively hard to game, but you could find ways to cut corners, so the next aspect is whether the organization cares. If you create defective screws, does someone try to detect that? Customers will eventually complain if you don't.
The longer the feedback loop, the more it depends on the organization actively watching itself. A culture of caring about quality for qualities sake, rather than only caring once a customer is complaining (or has left for a competitor) helps.
Ultimately, someone always pays for failing to manage well. Incentives are part of management. But their limits must be known. If you care about quality directly, or realize eventually you'll pay the price, you'll try and stop it. The longer the feedback loop, the easier it is to overlook, and the more people might have escaped in one way or another before the cost gets paid.
The “does the organization care” aspect feels like the load-bearing one — a gameable metric in a culture that actually checks the screws is safer than a clean metric nobody audits. Goodhart gets cited as a law about metrics; your framing makes it sound more like a law about attention.
The feedback-loop point is the one that stays with me — it decides not just whether the bill arrives but who’s still there when it does. Short loops charge the person who cut the corner; long loops charge whoever holds the seat when it lands. Which might be why gaming clusters where tenure runs shorter than the loop — your screws example is partly self-policing because nobody outruns a defective screw for long.
This episode is a very good example of a recurring and difficult problem. There's a substantial economics literature on it, under the general category of "incentive compatibility". The Soviet "planning" system seems to have been seriously impeded by lack of incentive compatibility.
I was never able to muster much enthusiasm for the management POV of an organization. From the bottom-up, these tokenmaxxing orgs may just be decent companies to work for, provided the engineer can somehow make sure they're happy delivering, while working around the malfunctioning aspects of the company.
That's understandable. Taking on someone else's POV is a challenge. But, if you want change though, you have to do it. Or at least you should. If you don't understand their POV, your using force rather than persuasion. Force has it's own problems. At the very least, there's the real chance you don't have enough of it to reach your goals. But using it also has consequences.
One of the blocks in understanding management is often in looking at yourself only. I was a sort of annoying student in high school. Dedicated and smart, but I have my own mission, and the teachers weren't always serving it. I got thrown out of Chemistry classes for reading Chemistry books. I thought it was pretty unreasonable at the time. If the teacher wanted to teach the rest of the class, but I already knew this topic, why not study the next one?
Problem is, some students actually do need a teacher motivating them to stay engaged and they didn't know how to adjust their plan for that to my uniqueness. I don't think the solution was for me to just shut-up and go along, but if I'd thought about it more, maybe there was another option.
I think emotionally, it may be hardest to take the POV of someone with power. Maybe because it feels like it should be their responsibility to take yours? ("With great power comes great responsibility"). But it's a pretty bad tactical move to not try, which came up in a different topic for me recently: https://substack.com/@norabble/note/c-295769613
Understanding their POV helps regardless of which way you go. If you want to outmaneuver them, it helps there. If you want to work toward collaboration, absolutely critical.
The rollup section is the sharpest part — "you've asked people to police a number that pays them to look away" names the structural failure most Goodhart discussions skip: the gaming is downstream of the wiring, not the people. And the stress-test framing earns its keep — the leaderboard didn't make the crack, it loaded it until it showed. Have you ever seen a metric survive being tied directly to standing, or is the number-reward gap the one non-negotiable?
Survival is relative here. But yes, there's some survival. What the original metric is, how gameable it is, and how directly aligned with objectives it is, is the first aspect. The most accessible examples are generally creating something physical and non-complex. The number of screws made per day for example.
Screws made is comparatively hard to game, but you could find ways to cut corners, so the next aspect is whether the organization cares. If you create defective screws, does someone try to detect that? Customers will eventually complain if you don't.
The longer the feedback loop, the more it depends on the organization actively watching itself. A culture of caring about quality for qualities sake, rather than only caring once a customer is complaining (or has left for a competitor) helps.
Ultimately, someone always pays for failing to manage well. Incentives are part of management. But their limits must be known. If you care about quality directly, or realize eventually you'll pay the price, you'll try and stop it. The longer the feedback loop, the easier it is to overlook, and the more people might have escaped in one way or another before the cost gets paid.
The “does the organization care” aspect feels like the load-bearing one — a gameable metric in a culture that actually checks the screws is safer than a clean metric nobody audits. Goodhart gets cited as a law about metrics; your framing makes it sound more like a law about attention.
The feedback-loop point is the one that stays with me — it decides not just whether the bill arrives but who’s still there when it does. Short loops charge the person who cut the corner; long loops charge whoever holds the seat when it lands. Which might be why gaming clusters where tenure runs shorter than the loop — your screws example is partly self-policing because nobody outruns a defective screw for long.