I went to the FIRST Robotics Competition World Championship expecting to be impressed by faster robots. I was. But that isn’t the thing that stayed with me on the flight home. What stayed with me was quieter and more inconvenient: my sense of what counts as working had been too generous, and I hadn’t noticed until I saw the alternative up close.

At a regional, a robot that does its job most of the time feels good. It cycles, it scores, it mostly doesn’t break, and when it does break you tell yourself the failure was unlucky. Worlds took that story apart. The distance between a good team and a great team isn’t that the great team’s robot is dramatically more capable in the best case. Often the peak performance looks similar. The difference is that the great team’s robot does the same thing the same way every single time, and mine didn’t.

Here is the idea I keep coming back to. Say your robot works nine times out of ten. That sounds excellent. Locally it is excellent. But nine out of ten means one failure in ten attempts, and at Worlds you don’t get to choose which attempt fails. The tenth time doesn’t politely happen in the pit during practice. It happens in a playoff match, in the cycle that decides the score, in front of the alliance that picked you partly on the assumption that you were reliable. A ninety percent robot and a failed robot are indistinguishable in the exact moment that the ninety percent matters. Reliability, I realized, isn’t a finishing touch you apply after the design works. It is the design. If a mechanism only works when conditions are kind, it doesn’t work.

Once I saw that, I started noticing how the strong teams actually behaved, and it wasn’t what I expected. I had assumed the top pits would feel intense, maybe frantic — more caffeine, more shouting, more heroics. They felt the opposite. They were calm and systematic in a way that read almost boring. Nobody was improvising a fix mid-event that should have been solved in January. The heroics had already happened, months earlier, in the form of testing that removed the need for heroics now. What looked like composure was really the visible surface of a process that had spent its uncertainty long before Houston. They weren’t lucky under pressure. They had engineered the pressure out.

That connects to something about how they treated the unknown. A team like mine tends to build until the robot works, then stop and call it done. The teams I watched seemed to treat “it works” as the start of the real work, not the end. Once a mechanism functioned, they went looking for the conditions under which it would stop functioning — the loose field element, the slightly-off approach angle, the low battery, the tenth attempt. They hunted their own failure modes on purpose, because they knew that any margin they didn’t find themselves would be found for them, at the worst possible time, by the tournament. “Working most of the time” is a comfortable illusion. It’s a robot whose failures you simply haven’t triggered yet.

I also came away with more respect for simplicity than I had going in. I expected the best robots to be the most mechanically complex, and some were intricate. But complexity wasn’t the thing that separated them. A simpler mechanism has fewer independent ways to fail, and at a competition where reliability is the whole game, that’s not a compromise — it’s an advantage. I had been quietly equating “impressive” with “complicated.” Worlds pushed those apart. The impressive thing is a robot that does one job perfectly a hundred times in a row, even if the job is unglamorous.

None of this made me want to compete less. It did the opposite, but it moved the target. Before, my instinct after a build was to ask whether it worked. Now the question I want to ask is narrower and harder: does it work every time, and if not, do I actually know why not, or am I just hoping the tenth case doesn’t come up? That second question is uncomfortable because it usually has an honest answer, and the honest answer is usually “I don’t know yet.” But not knowing is information. It tells you exactly where the next work is.

The most useful thing I took from Worlds wasn’t a mechanism or a strategy I could copy. It was a recalibration. I had been measuring my work against the robots around me at regionals, and by that ruler I was doing fine. Standing in a building full of teams who treated ninety percent as failure, I could see that the ruler was the problem. You don’t rise to a higher standard by wanting to. You rise to it by first being able to see it, clearly enough that your old definition of “good enough” stops being usable. That’s what happened to me in Houston. The robots were faster. But the thing that actually changed was the number I’m now unwilling to accept.