The equation had lived on the whiteboard for four months.
Justin had written the first version in June, before Jay and Navan arrived, before the factory had a single scenario or a single agent. It was a mathematical statement about convergence—the conditions under which a population of AI agents, iterating on code through scenario-driven feedback, would reliably approach a satisfaction target rather than diverge from it. The original version had three terms and two unknowns.
By August it had seven terms and one unknown. Jay had contributed a correction to the feedback coefficient. Navan had pointed out that the convergence rate needed to account for inter-scenario dependencies—the fact that improving one scenario sometimes degraded another, creating a coupled optimization problem. Justin had rewritten the equation twice, each time tighter, each time closer to something that felt like truth.
In September, the unknown collapsed. The factory's own data provided the missing term. Three months of scenario iterations, thousands of data points, and the empirical results mapped onto the theoretical curve with an R-squared of 0.97. The equation wasn't a hypothesis anymore. It was a description.
Justin stood in front of the whiteboard on a Monday morning in October. Jay was at his desk. Navan was refilling his coffee. Neither of them was paying attention to the whiteboard, because the whiteboard was furniture at this point—a permanent fixture of the room, like the dashboard or the coffee machine.
Justin photographed the equation. His phone's camera shutter sound was the only noise in the room. Jay looked up. Navan turned around.
Then Justin picked up the eraser and wiped the whiteboard clean.
Four months of work. Seven terms. Every revision, every correction, every midnight addition—gone in a single sweep of felt on porcelain.
"Justin," Navan said. Just the name. Just the tone of someone watching something they cared about disappear.
"It's done," Justin said. He picked up the marker. The whiteboard was blank and white and waiting, the way it had been in June before any of this started. He uncapped the marker and wrote a single word in the center of the board, in letters large enough to read from across the room.
Next.
He recapped the marker and set it in the tray and sat down at his laptop. Jay stared at the whiteboard. Navan stared at the whiteboard. The word sat there, alone, surrounded by all that empty space.
"That's it?" Jay asked.
"The equation is solved. The factory's convergence is proven. The data confirms it." Justin opened a new document. "So now we figure out what comes after convergence."
Navan opened his notebook to a blank page. He wrote the date. He wrote: The equation is solved. Justin wrote "Next." What comes after convergence?
He didn't have an answer yet. Neither did anyone. But the whiteboard was clean, and that meant the next question was already forming.
The R-squared of 0.97 is such a specific detail and it's perfect. Not 0.99, not 1.0. 0.97. Close enough to trust. Not close enough to stop questioning. That's real science.