The office had a ping-pong table. That was the first thing you noticed when you walked in, the second thing being the beanbag chairs scattered like discarded luggage around the perimeter, and the third thing being the poster on the wall: We Are Going To Change Everything, in Comic Sans, because in March of 1999, Comic Sans was the font of truth. I was thirty-nine years old and I had built a prediction engine. Not a weather engine, not a stock ticker, not one of those things that just told you numbers. I built something that took the numbers you had and the numbers you didn't have and gave you, with mathematical certainty, the future.

My name is Jonathan Cross. I founded PredictaCorp in a garage in Palo Alto. I still had the garage. My wife had taken the kids to Marin on weekends, which was our version of taking a break. She said I lived in the garage more than I lived in the house. She was right. The garage had something the house didn't: no one to disappoint.

The prediction engine was my life's work. It started as weather modeling, like the farmers who have tracked rainfall since their grandfathers walked the fields, but I went bigger. I took agricultural data, economic indicators, demographic shifts, solar cycle patterns, ocean temperature readings, atmospheric pressure systems, and three hundred other variables, and I fed them into a system that didn't just predict what would happen, but told you what was about to happen before the people living through it knew. I was the man on the Minnesota farm tapping his weather records, but instead of one small farm, my farm was the entire global economy, and instead of wheat, I was growing certainty.

For two years, it worked. Perfectly. We predicted the California drought three months before it was announced. We predicted the soybean surplus before the Chinese markets moved. We predicted a supply chain disruption in rubber that turned out to be correct. The venture capitalists loved us. They flew us to conferences in Hawaii and put us on panels where we stood in front of screens showing beautiful curves and said things like "the data doesn't lie" and "mathematics is the only honest language we have."

The NASDAQ hit five thousand in March. I remember standing in our office, drinking espresso that cost more than my father's weekly grocery budget, watching the numbers climb on the television mounted in the corner, and feeling that particular cocktail of triumph and nausea that comes when you realize you might be the smartest person in a room full of people who would give you money to be wrong.

But the internet economy was doing something my model didn't account for. It was doing something that couldn't be accounted for in the existing mathematical framework. The market was no longer a reflection of economic reality. It was a reflection of belief in economic reality. And belief, it turns out, is not a variable you can measure with instruments.

I knew this intellectually. Every founder knows, at some level, that the game has changed from building products to building narratives. But I didn't feel it in my bones. My bones were telling me the model was right. The NASDAQ had to come down eventually, but eventually is a mathematical concept, not a human one, and the market was operated by humans.

On April tenth, the NASDAQ peaked at five thousand and then started to fall. I remember checking my model at 10:47 AM, running the latest data through it one more time out of habit, and watching the screen spit out a prediction that said: market stability for the next eighteen months with normal seasonal correction of three to five percent.

The numbers were beautiful. The columns were neat. The confidence intervals were tight. And the world was ending.

I watched for six hours as my prediction engine told me, with increasing desperation, that everything was fine. The model had been trained on data from a world that was disappearing. It was predicting the economy the way Anders Nielsen predicted weather: using forty years of looking at the same sky, in the same place, believing that clouds worked the same way they always had.

But the internet economy didn't work like weather. Weather is cyclical. Weather is honest. Weather tells you what it's going to do. The new economy was something else entirely: a feedback loop of belief feeding on belief, a machine that ran on the conviction that conviction itself was value.

I left the office around seven PM. The ping-pong table sat untouched, the green felt still warm from whatever game someone had been playing before they started watching their own company dissolve. I walked down Sand Hill Road in the California sun, which was doing what California sun does, which is shining with an almost offensive sense of normalcy, as if to say: look at me, I am still here, I am still beautiful, nothing that happens inside these walls has anything to do with the natural order of things.

The crash took six months. PredictaCorp took nine. By January of 2000, the prediction engine was running on a server in a basement, processing data that nobody was reading, generating forecasts for a world that had already moved on to something else.

I sat in the garage one evening, the ping-pong table having been sold two weeks earlier, eating takeout from a place that was already gone because its lease had been up for months and nobody had told the owner. I opened my weather records. Four years of data. More variables than Anders Nielsen would ever dream of, more computing power, more sophistication, more everything. And every single prediction wrong, because the game had changed while I was busy making sure the pieces moved in the right order.

The last entry in the log was dated October nineteenth. It said: commodity prices stable, weather patterns normal, economic indicators within expected range, outlook: positive. I looked at it for a long time. The data was right. The data had always been right. Weather had been normal. Commodity prices had been stable. The economy had been within expected range according to every metric I had ever designed.

The market hadn't moved against the data. The market had moved beyond the data. And that is the thing that prediction engines cannot tell you: when the universe changes its rules, the person who has memorized the old rules perfectly is not wise. They are just the first to be wrong.

I closed the laptop. Outside, the California sun was setting, painting the hills in that particular shade of orange that only happens in Palo Alto, because California knows how to do sunsets the way Anders Nielsen knew how to do wheat: with a precision that only comes from forty years of watching the same thing every single day, believing that watching is the same thing as understanding.

The NASDAQ would recover. That's not the interesting part. The interesting part is that when it did, it would be a different market, with different rules, and my prediction engine would need to be rebuilt from scratch. And the next prediction engine, and the next one, until we had something that could handle a world that refused to be predicted.

But that would be someone else's problem. My problem was that I had spent my life learning to read a sky that no longer existed.

Copyright