You have a … wat?

So you're not testing anymore, because a failed test would show up on someones dashboard?

Increase your testing speed by 50% by replacing your `#!/bin/bash` (or whatever) with `#!/bin/true` and reducing your script to 1 line.

If this is true as you describe it and has been true for more than a few days then I am reasonably sure the organization you work for has at least one but probably 2 or more problems that are bigger than both Goodharts law and exit 0 on line 2.

Both are almost unavoidable when your management layers grow beyond a certain level.

Most of the complexity, inefficiency and the related failures in today corporate and institutional world is driven by too much management. The map has become the territory.

It ceases to be a good measure: it captured your preferences properly prior to becoming a target.

good == honest in this context

What your saying sounds like profound next level wisdom and on a very surface level does make sense, but think of this: an org can create a measure that captures the relevant preferences properly, and everyone is quite happy with the results because the team continues or slightly modifies what they’re doing and lifts up the product to lift up the measurement and life is good.

And then over time they start realizing they don’t have to lift up the whole product, but just a small piece to increase the measurement. So they do that and the measurement goes up, but the product doesn’t get better because they’ve found the path of least resistance to raising the measure. This is really the underlying crux of Goodharts Law. So it was a good measure probably until it became a target.

So what is a manager to do? “Capture all [the] preferences properly” as you put it? Probably not because that quickly devolves from measurements to long form status reports, not even measurements, because it’s impossible to capture all dimensions of this with measurements, so one has to reduce the dimensionality a little.

This is a philosophical problem not a technical one. Though your point does seem superficially correct, in practice with real teams the second and third order effects from the measure becoming a target dominate.

> It sounds like a profound wisdom and on a very surface level it does make sense, but think of this: if you can assess that a measure is a bad one, this means that you have your own intrinsic preferences, otherwise you wouldn't be able to tell that!

That assumes that the ones working towards a specific KPI are both the ones noticing the metric is bad and that their are noticing it before the negative consequences have destroyed their product / company.

Often when the issue of optimizing to KPI comes up, it seems that exterior people notice the problem but not interior ones, or otherwise the issue is noticed after the fact but not before a meaningful change could be affected.

> Therefore, if you are unhappy with a measure, it means solely that it doesn't capture all of your preferences properly. Which is a technical problem rather than a philosophical one.

Here, you're assuming that the space of possibilities is an ordered set. But it's probably not, so your “technical problem” is in fact a mathematical impossibility.

With KPIs, you push people to maximize a very-straightforward-but-fundamentally-flawed metric, instead of relying on their own judgment. Or when not trusting your employees end up having them behave like brainless bots.

The aphorism is pointing out that if something is made a target, people will game the target, even if that has very bad effects on the company or product.

So even a good measurement is vulnerable to this problem.

You can't really align (hahahaha, hahahahaaaaaaa, align, anyway) stated goals of a company under capitalism with actual goals, because there are multiple competing games being played at various levels and when you actually pick a level and make their goals explicit, an actual class war would quickly ensue.