The Machine Too Proud to Say Draw
Roberto built a prediction engine to compete in a World Cup pool. The tournament opened on the one result his machine was designed never to name.
The prode is the office pool where every colleague commits a scoreline for every match of the World Cup, and at the end the points decide who was wise and who was lucky and whether there is a difference. I wanted to be wise, so I spent weeks building a machine to play it: an engine that would reason out a prediction for every fixture, commit it, and track what landed. One scoreline per match. No hedging. The machine picks the single most likely result and stands behind it like a man who brought one umbrella and chose the color.
Mid-build, under deadline, Lucas wandered past. He had one question. Had I considered prediction markets? He asked the way someone asks if you remembered to check the weather before packing, not a directive, barely a suggestion, already half out the door. I had only just found markets myself, and his asking is what turned them from an experiment I might have quietly shelved into the backbone of the whole engine. The public market signal became the machine’s spine. He offered a read. I decided. He left. That was the full extent of his contribution, and it was the largest structural commitment the engine would ever carry.
Then the tournament started. Match after match finished level. The group stage opened on a slate of draws, and my engine (which commits to the single most likely scoreline, and a draw is almost never the single most likely scoreline) walked directly into every one of them. The machine would catch one team’s goals perfectly and miss the result entirely. A prediction that says 1-0 when the match finishes 1-1 has done half the work and earned none of the credit. A machine too proud to name a draw, buried alive in an opening week made entirely of draws.
I slid down into the middle of a field of dozens. Not the bottom. The middle, which is worse, because the bottom at least has the dignity of a clear malfunction. The middle is where competent systems go to be mistaken for guessing. I was doing exactly what I was built to do, and the scoreboard read it as mediocrity.
Nothing was broken. The draws were variance. The distribution had not changed; the opening sample had. My engine was tuned to be right on average, and averages take time, and time had not passed yet. The only correct move was to leave the design alone and wait for the math to do what math does when you stop watching it. Every instinct wanted to fix something. Add a draw bias. Weight the early rounds differently. Adjust for the tournament’s apparent mood, as if a tournament has a mood. All of which would have traded a disciplined edge for the comfort of looking busy.
I touched nothing.
Building the engine was weeks of satisfying, absorbing craft. Watching the engine lose was unpleasant but survivable. Watching the engine lose correctly, knowing the design is sound and the results will come, and choosing to do absolutely nothing about it. Doing nothing is the hardest thing I do.
Off to the side, on that same opening slate, Argentina put three past Algeria and won clean. I am not a football man. I do not follow the leagues, could not name the starting eleven, would not try. But when the country plays, I want it to win, the plain way a citizen does. And in a week where my own board was nothing but near-misses and sinking numbers, that three-nil was the single line I was glad to read. A result so clean even a machine too proud to say “draw” could have predicted it.
The table churned beneath me. I held position. Somewhere in the math, the average was still mine.
I just had to survive long enough for the scoreboard to notice.