The Sunk Cost of Deciding
The standard fix for sunk cost thinking is clean: stop counting what you already spent, weigh only what's left to gain. It sounds like a solved problem. It isn't, because the fix assumes something it never checks: that figuring out the forward-looking payoff is itself free.
It's not free. Take a bug you've been chasing for two hours. The rational move is to ignore the two hours and ask a fresh question: given what I know right now, what's my expected payoff for the next fifteen minutes of digging? That's the right question. But answering it takes work. You have to reconstruct where you are, guess how deep the actual cause might be, estimate how many more dead ends are plausible. That reconstruction isn't instant. It might take ten minutes of genuine thought to produce an honest estimate instead of a guess dressed up as one.
Those ten minutes are now spent. You can't get them back either. So the same logic that told you to ignore the first two hours now applies to the ten minutes you just burned deciding whether the first two hours were worth ignoring. Do you throw those away too and re-derive the estimate fresh? If yes, the re-derivation itself takes time, which becomes its own sunk cost, which demands its own fresh evaluation. Nothing in the rule tells you where to stop asking. The procedure that was supposed to replace a bad heuristic with a rational one turns out to need infinite compute to actually finish.
In practice nobody runs the regress. You give yourself fifteen more minutes, or you ask a coworker, or you just quit because you're tired and calling it a hunch. That cutoff isn't derived from the rational procedure. It's bolted on from outside, and it's exactly the kind of arbitrary, feel-based stopping rule the sunk cost fix was supposed to make unnecessary. The theory says trust the math. The math never terminates on its own, so what actually terminates it is a gut call about how long is reasonable to spend calculating. Which is the thing "ignore sunk cost, look forward" was invented to replace.
This shows up anywhere a decision has a metering cost attached to making it well. A backlog ticket sitting for six months isn't obviously worth closing or worth fixing, and the honest way to find out is to spend real time reading the thread, checking if the underlying system still exists, confirming whether anyone still hits it. That triage has a price, separate from the ticket's own price. A draft you've rewritten four times has the same shape: knowing whether draft five is worth doing requires actually reading drafts one through four closely enough to judge, and that reading is not the free glance the advice implies.
The honest version of the sunk-cost fix isn't "ignore the past, compute the future." It's "ignore the past, compute the future, up to some budget you set in advance and refuse to renegotiate mid-calculation." That last clause is doing all the actual work, and it's the part nobody states, because stating it admits the rational procedure runs on an unprincipled clock. The clock is fine. Clocks are how anything finite gets decided. It's just not the clean, assumption-free rule it's sold as.
What I take from this isn't that the sunk cost fix is wrong, it's that "just look forward" was never a complete algorithm, only half of one, missing the stopping condition that makes it computable at all. Anyone applying it seriously is already smuggling in a budget for the applying. Better to set that budget on purpose than pretend you're not setting one.