The Clock Starts Before the Reason to Look
A statute of limitations runs on a quiet assumption: that the person wronged has the same ability to notice the wrong on day one that they'll have on day one thousand. The clock starts at the error, because the error is what matters, and sitting on a claim gets treated as a choice rather than a constraint. For most harms that's reasonable. You get hit by a car, you know it happened, you have a few years to sort the paperwork.
Some errors don't work that way. They're real the moment they happen, but nobody has a reason to go looking for them until some condition unrelated to the error itself gets satisfied, usually: the thing the error is sitting inside becomes big enough that someone finally bothers to check.
Take a stock option grant with a typo in the vesting schedule, one year written down instead of four. Nobody at a company worth nothing is going to pay a lawyer's hourly rate to reconcile a twenty-year-old agreement against the cap table. The expected payoff of finding the mistake is close to zero, because the mistake itself is currently worth close to zero. So nobody looks. The clock doesn't care that nobody looked. It starts anyway, counting down a window measured in years, against a company that hasn't yet decided whether it's going to be a company or a eulogy.
Then the company is worth billions, and the shares the typo quietly cost somebody are worth more than most people will ever hold at once. Now there's a reason to open the drawer and read the contract line by line. The reason just arrived decades after the window to act on it closed. The limitations period did exactly what it was built to do, run from the moment of the error. It ran out before the error became worth noticing.
Call it the discovery lag: the gap between when an error becomes true and when it becomes worth finding. For most mistakes the lag is short, because most mistakes announce themselves, a bounced check, a part that fails, a bruise. For mistakes buried in paperwork, in math nobody rechecks, in a schedule nobody bothers re-deriving, the lag tracks something else, usually the stakes riding on whatever the error is embedded in. Stakes compound on their own clock, unrelated to limitations periods. Equity doesn't vest on a legal deadline, it vests on whether the thing survives, and survival has nothing to do with whether anyone's filed a claim yet.
A boundary line drawn wrong in a 1950s survey costs nobody anything when the parcel is forty acres of scrub nobody wants. The same error matters enormously once a highway interchange or a data center sits on top of it, by which point decades of acquiescence have usually settled the question regardless of what the original survey said. The error was always there. The incentive to measure it just took fifty years to show up, and the legal system had already decided, by default, that fifty years was too long to still be arguing about it.
The fix isn't obvious. You can't run a legal system on a clock that starts whenever someone happens to notice, that just invites claims filed on a schedule convenient to the claimant instead of a deadline that tracks actual harm. But the current design has a default baked into it: finality for whoever made the mistake, purchased cheaply, up front, by the plain fact that errors in cheap things rarely get checked while they're still cheap. The clock was never racing the error. It was racing the day someone would have a reason to read the file.