This reminds me of a conversation I frequently have with my dad and family. I have my cash investments in a cash plus account, earning 4.7% interest. They're all hooked on T-bills and laddering, which has a current rate of 5.55%. I'm finishing my PhD program so don't have a large investment. Let's say $10k for easy math. At 4.7% that's $480.26 on return, while 5.55% is 569.34, a difference of ~$90. In my cash account, I am: FDIC protected, can withdraw and deposit money at any time, and am able to employ a "set and forget" strategy. With the T-bill method I need to lock my money in for 6 months at a time and realistically laddering this decreases my gains. I have to be quite active to perform this. Here's the question, is that time worth $90? To me, no. I'd much rather spend that time doing my PhD work or taking breaks to do things like visit HN. Not worry about my finances and squabble over money that isn't significantly meaningful and is small compared to my expectation in a few years. Any time spent following the active strategy is time that would be better invested investing in my PhD work, where the expectation of return is several hundred thousand of dollars a year. Even were my investment 10x, we are talking the difference of This is a common mistake I see in all types of modeling (stock investing, time estimation, quality assurance, etc) and is exactly the same game afoot here. Like the article says, we assume a recruiter's interests are aligned with ours since the more money we make the more money they make. But this is a poor model because we are not our recruiter's only prospect. And, like the article discusses, their returns are diluted. The recruiter's more optimal strategy is to not hold all their eggs in one basket. To diversify and operate on quick turnover rather than a not-so-golden goose. External costs and auxiliary factors often play critical roles in models.
The devil is always in the details. Nearly everyone I talk to believes that a first order solution is at worst non-optimal, but relatively aligned with the target function's trajectory. But when we look deeper and include nuance, we actually realize that our first order approximation is in the opposite direction! I hear "don't let perfection get in the way of 'good enough'" or similar cliques. But there is such a thing as having too naive of a model. Many functions necessitate even considerable "nuance" (orders of approximation), as an example sin(x). The real world is far more complex than the sine function. It thus necessitates understanding the limitations of our approximation, rather than simply approximating.
Don't let "good enough" get in the way of __good enough__.
Nuance isn't equivalent to pedantry.