The sunk-cost fallacy is the tendency to keep investing in a losing position because of what has already been spent, when what has been spent should carry no weight at all. The canonical case is the Concorde: Britain and France kept funding the aircraft years after their own projections showed it would never recoup its costs, because walking away would mean the billions already committed had bought nothing. That is the trap in its purest form: every further franc and pound was a fresh loss, freely chosen, to avoid booking an old one. Refusing to update the model and walk away does not protect the original investment; it grows the total loss, so that the position ends far worse than if it had been abandoned the day the projections turned. We all know the pattern, and most of us have built machinery to catch it: kill criteria, stage gates, quarterly portfolio reviews. That machinery works tolerably well when the losing position is external, a project, a fund, a product line. It fails when the position is the person holding it.

Because the largest investment any of us ever makes is not in a company or a thesis but in a self: decades of time, energy, social capital, and the accumulated testimony of everyone who knows us as a particular kind of person. Those investments compound. By midlife, the position is enormous, but like any large position held too long, it is often underwater. The strategist whose framework stopped explaining the market three cycles ago. The operator whose superpower was the growth phase the company left behind. The plain fact, visible to everyone nearby, that the person you built is no longer the person the situation requires. And here the familiar fallacy reveals itself as a special case of a deeper one. You cannot write down the value of a position when you are the position. The investor and the investment are the same entity, so the loss never gets booked; it gets defended.

Consider the founder who should but cannot shelve the company. Not the product: killing a product can be spun as discipline, even worn as a badge. The company is different. “Founder of X” has been the bio line, the dinner-party answer, the promise made to every employee recruited and every investor check accepted, for years. The corporate entity and the self have fully merged. So the company persists on bridge rounds and dwindling conviction, long past the point where the founder’s own talent would command a better return anywhere else. The opportunity cost is enormous and visible to everyone but the person paying it, because winding down does not register as reallocating capital. It registers as a crushing reversal of identity. It’s the curtain lifting to show that The Great Oz is just another feeble human.

...or the investor whose thesis became a persona. It’s the VC who made her name on B2B SaaS and raised every subsequent fund on that insight. It is in the LP letters, the conference keynotes, the firm’s own homepage. And the insight was even correct, once. Now multiples have compressed and the frontier has moved, but pivoting the fund is not an update to a model; it is a resignation from being the person the LPs backed. So the new fund looks like the old one, and reserves keep flowing into portfolio companies the same partner would never touch as fresh deals.

...or the academic whose early dissent from the field’s orthodoxy was genuinely brave, and who then spent twenty years converting that dissent into a brand, until the critique calcified and the heretic became the thing heretics are supposed to oppose: someone who cannot be moved by evidence.

In each case the tell is the same. The doubling-down grows most fierce precisely when the position is most underwater, because that is when the identity is most exposed.

The instinctive fix is decision hygiene: pre-mortems, kill criteria, red teams, outside views. These help at the margin, but their decades-long record of not solving the problem should tell us something. It is not that the tools are weak. The same executive who cannot leave his own dead company applies kill criteria to projects ruthlessly, and the projects die on schedule. The machinery works, the self has simply been exempted from its jurisdiction. The tools fail here because they operate at the wrong layer. They treat the founder’s pathological “grit” as a decision process to be improved, when it is actually an identity being defended. And no framework survives contact with a self that experiences disconfirmation as annihilation. Worse, the identity-level fix contains its own relapse: trade the old self for a new one and you have simply opened the next position, which will be just as vigorously defended when it eventually goes underwater. Whatever the durable intervention is, it must sit one layer down, at the system that governs how aspects of identity get adopted, defended, and discarded in the first place.

Call that layer the meta-identity, and notice that you already have one. Think of the self as a stack. At the bottom sits the hardware: physiology, temperament, energy, the machine you were issued. At the top runs the software: the tactical identity—founder, SaaS investor, heretic—the role currently in the foreground. Between them sits firmware, the meta-identity, which controls what software gets installed, how long it runs, and what it costs to replace. Everyone ships with firmware; almost nobody wrote their own. It was installed by parents, culture, and early wins, and it runs invisibly. Its signature is that it survives software reinstalls. This is why people change companies, cities, and spouses but reproduce the same patterns in each. The application was swapped, but the firmware that selects and defends applications was never touched. The baffling behavior I described earlier is not a malfunction. It’s not a system deviating from its programming. It is firmware executing exactly as installed, on defaults that treat any change to the running identity as a threat to the machine itself.

Luckily, realizing that the visible aspects of our identity are selected upstream leads directly to the solution. However, it is not, as it first appears, to construct a meta-identity—some gleaming new self-concept built around “I am someone who updates.” Built ex nihilo, that is just more software, one more position to defend. Anything you consciously adopt gets installed by the incumbent firmware, under its rules, and therefore runs on top of it. You cannot get beneath the firmware by adding, only by editing what already runs. The real work is excavation and revision of the firmware you already run. Surfacing it is uncomfortable but not mysterious. Ask what identity updates you have resisted hardest, and why those. Ask which change to your self image, if forced on you tomorrow, would feel less like a revision and more like a death. Ask what personal reputation you have re-adopted in every job, city, and partnership. The answers are the firmware, read out loud for the first time. Only then can you edit deliberately, rewriting the defaults so that evolution of capacity and preference registers as identity-consistent rather than identity-threatening. Only then does quitting the dead position become what you do because of who you are, not something you resist in order to avoid writing off a past investment in your self (two words, not one).

A caution about scope. Firmware can be flashed, but it cannot be wiped. A full rewrite is neither possible nor desirable, because something must persist to do the updating: a kernel of values, taste, and continuity of judgment without which revision becomes dissolution, and there is no one home to approve the next change. The goal is not a content-less self endlessly rebooting. It is a small, stable kernel and cheap, replaceable software, which is to say the opposite of how most successful people are configured by fifty: enormous, expensive software running on firmware which has never been reviewed.

Return, finally, to the founder and the zombie company. What that founder needed was never a better framework for decision making. Every framework had already rendered its verdict, and the verdict changed nothing. What the founder needed was firmware in which winding down the company could register as an update (consistent with a constantly-improving self) instead of an erasure (the death of the self tied to previous investment). That is the general lesson.

The sunk-cost fallacy was never about the money, and the discipline you prize in your resource allocation is one you have quietly exempted yourself from. Your notion of self is the largest position you hold, the least diversified, and probably the only one you have never once written down to its true market value.