The most consequential software bug of the twentieth century is famous for not happening. The Y2K problem was simple enough to explain to a child: for decades, software stored years as two digits, because memory was expensive and 19 was the century of everyone who wrote the code. When the calendar rolled to 2000, "00" would be indistinguishable from 1900, and every calculation that depended on dates, which is to say interest, benefits, expiration, scheduling, and most of industrial civilization, risked going politely insane. What followed was one of the largest coordinated repair efforts in technological history. Governments and companies spent years and, by common estimates, several hundred billion dollars worldwide hunting down date logic and fixing it. Then midnight came, the planes did not fall, the banks did not melt, and the public delivered its verdict: it was all hype.
That verdict is the interesting failure in this story, and it is not a software failure. It is an evaluation failure, the same one that stalks every prevention budget ever written. The work succeeded, and the success erased the evidence that the work was necessary. There were real incidents at the rollover: a nuclear plant's monitoring system in Japan hiccupped, slot machines in Delaware stopped, some credit card terminals and government systems glitched, a U.S. intelligence satellite ground system had trouble for a couple of days. They were footnotes because the main event had been dismantled in advance, line by line, by an army of unglamorous people doing inventory on systems nobody wanted to think about.
Understand what the remediation actually was, because it is the part our industry keeps refusing to learn. It was not a breakthrough. It was a census. Find every system. Find every date. Decide what breaks. Fix it. Test it, ideally by rolling clocks forward in a lab and watching what the software does when it believes the future has arrived. This is maintenance as a heroic act, performed at global scale, with a deadline set by the rotation of the Earth and therefore absolutely non-negotiable. There has never been a better forcing function in the history of engineering management than a date that cannot be moved by a steering committee.
The political afterlife of Y2K should be laminated and handed to every reliability team on Earth. Prevention that works looks, from the outside, identical to prevention that was never needed. The budget holder cannot see the counterfactual. So every organization drifts toward the same equilibrium: fund the visible fire, starve the invisible one, and let the future re-litigate. Our factory's answer to this is dull and total: keep receipts. When a gate stops a bad artifact, record it. When a reserve absorbs a spike, record it. When a regression test catches the bug it was built from, record it. The file of prevented disasters is the only argument prevention has ever had, and it does not compile itself.
There is also a finer technical point worth rescuing from the jokes. Y2K was never really about the year 2000. It was about a representation decision made under scarcity, two digits because bytes were dear, outliving the scarcity by thirty years and becoming load-bearing for a civilization. Every system you own contains decisions like this: a limit chosen when the world was smaller, an identifier that will run out, a timeout tuned for a network that no longer exists, a "temporary" format now entering its second decade of service. The inventory question Y2K asked is permanent. Which of our cheap assumptions is currently compounding, in the dark, at planetary scale? You do not need to answer it today. You need to have the census machinery that makes answering it possible before the calendar volunteers to ask.
Midnight, January 1, 2000, remains the best-reviewed non-event in history. Nothing happened, on purpose, at enormous expense, and the expense is why nothing happened. Remember that ordering the next time a quiet system asks for its maintenance budget. Silence is not the absence of risk. Sometimes silence is the sound of the work having been done.
Sources, for the record: global remediation cost estimates vary widely; figures in the range of $300 billion and above are commonly cited by the U.S. government and industry retrospectives. The minor rollover incidents listed (including the Japanese plant monitoring fault, the Delaware lottery machines, and the satellite ground processing problem) were documented by the President's Council on Year 2000 Conversion and contemporaneous reporting in January 2000.