I fed the classic Secretary Problem into an AI and asked: given a 70-year human lifespan, when should a person stop playing the field and commit?
The Secretary Problem, for the uninitiated: you're hiring a secretary (it's a 1950s problem, forgive the framing). Candidates arrive one at a time. After each interview, you must decide immediately: hire this person or pass. You cannot go back. You have no idea how many candidates there are in total.
What's the optimal strategy?
Answer: interview the first 37% of candidates without hiring anyone. Use that time to calibrate your standards. Then hire the first person who's better than everyone you've seen so far.
This gives you roughly a 37% chance of ending up with the best candidate. Which sounds low, until you realize that blindly picking someone at random gives you a 1-in-n chance, and "picking the first person you interview because you're lonely and they seem nice" gives you something measurably worse.
Now apply this to a human life.
If you assume your romantic window runs from, say, 18 to 70 — that's 52 years. 37% of 52 years is about 19 years. Which means the mathematically optimal strategy is: spend your late teens and your entire twenties watching and learning, then at ~37, start actually committing.
The internet has known this as the "37% rule" for a while. It gets passed around as either reassuring ("I'm 34, I still have time!") or horrifying ("I'm 38, I missed the window!") depending on who's reading it.
I find it mostly funny. Not because the math is wrong — the math is correct — but because of what it quietly assumes.
It assumes candidates arrive in random order.
They don't. The people you meet at 22 are not a representative sample of everyone you'll ever meet. They're mostly people who happen to be in the same zip code, in the same broke-and-figuring-it-out phase of life, with the same taste in bad decisions. The distribution is wildly skewed.
It assumes your preferences are stable.
They're not. What you want at 24 and what you want at 38 are different enough that your past-self's assessments are almost useless data. The "best person I'd seen up to age 37" was ranked by a version of me who had entirely different ideas about what mattered.
It assumes you can't go back.
You often can. People reconnect. Circumstances change. The strict no-backtrack rule made sense for hiring a secretary in 1960. It maps poorly onto the actual texture of relationships.
And most importantly: it optimizes for selecting the absolute best candidate.
But the absolute best and the good enough, and here, and wanting to be with you are different things. And in my experience the second one is considerably rarer than the math implies.
None of this is to say the Secretary Problem isn't useful. It is. The core insight — spend time early learning what you value, then act decisively when you see it — is genuinely good advice. Most people do the opposite: they commit too quickly before they know themselves, or they keep sampling forever because deciding feels like losing.
The 37% rule is a corrective to both failure modes. It says: you need a learning phase, and the learning phase has to end.
What the math can't tell you is what you're actually calibrating for. The secretary problem assumes a total ordering — candidate A is strictly better than candidate B. But people aren't totally ordered. They're just different, in ways that matter differently depending on who you're becoming.
So the real lesson isn’t “wait 37% of your life.” It’s that you’re always running two different problems at once.
One is selection under uncertainty — the clean, mathematical one. The other is participation — messy, recursive, and very human. In the first, you’re trying to find the best option. In the second, you’re becoming someone for whom certain options even make sense.
The Secretary Problem treats you like a fixed evaluator scanning a static list. Life doesn’t. You are changing, the pool is changing, and the act of choosing feeds back into both.
Which means there isn’t a single optimal stopping point. There’s only a moment when your standards, your timing, and another person’s willingness happen to align — and you decide that this alignment is worth collapsing the search.
Not because the math says this is the best you’ll ever see.
But because, at some point, continuing to optimize stops being rational and starts being a refusal to live with a choice.
And no equation can tell you when that shift happens — only that eventually, it does.