Palo Alto, 1999. Julian Cross stood in the middle of his office, which was a glass box suspended above a parking lot, and stared at the projection on the wall that represented his company's entire mission in the form of a two-dimensional scatter plot. The dots on the plot were product features. The axes were idealism and greed. And the dots were migrating, slowly, inexorably, toward the upper right quadrant where the label read Market Dominance and Julian knew with a certainty that felt like nausea that they were not migrating there by accident.

Veridian Systems had been founded eighteen months ago on a principle that Julian could still articulate clearly because it had been written on a whiteboard in the first office they had rented, a garage that smelled of paint and optimism. The principle was simple: information should flow freely, and the people who controlled the flow should be transparent about what they controlled and why. This was not a business plan. It was a moral position dressed in the vocabulary of technology, and Julian had believed it with the full intensity of a thirty-one-year-old who had left a comfortable job at a major telecommunications company because he had been asked to design a system that prioritized certain customers' data over others based on their payment tier.

The first vector in the latent space was Julian's original position, the starting point of the interpolation. He had plotted it himself: idealism high, greed low, near the upper left quadrant that the scatter plot labeled Visionary Naivete. This was not meant as an insult. The investors who had funded Veridian had used that label approvingly, because visionary naivete was the quality that made founders willing to work for equity instead of salary, willing to sleep on office couches, willing to believe that a better product would naturally defeat a worse one.

But a vector is not a point. A vector has direction and magnitude, and the vector that had been carrying Julian and his company through the latent space of product decisions was pointing steadily toward greed, and the rate of movement along that vector was accelerating.

The first compromise had been small and defensible. They had added an advertising tier to their free data-sharing platform, a thin strip of banner ads that generated revenue without compromising the core principle of free information flow. Julian had supported this decision because the advertising was minimal and non-intrusive, and the revenue it generated kept the servers running and the developers paid. The vector had shifted slightly to the right, a degree or two of rotation toward greed, but the magnitude was small enough to be invisible.

The second compromise was larger. A venture firm offered a funding round that would expand Veridian from a team of twelve to a team of one hundred and twenty, but the terms included a data-sharing agreement with a partner company that would allow the partner to access aggregated user data for targeted advertising. Julian had argued against this provision for six hours across three meetings, and then he had accepted it, because the alternative was running out of money in eleven weeks, and because the data would be aggregated and anonymized, and because the partner was a respectable company that had been in business for forty years and had never done anything unethical that anyone could prove.

The vector rotated another fifteen degrees. The magnitude increased. The dots on the projection moved.

By September 1999, the company had one hundred and eight employees. The advertising tier had been expanded from a thin strip to a full sidebar. The data-sharing agreement had generated three times the projected revenue, and the partner company had requested expanded access that would include individual user profiles, not just aggregated data. Julian had said no to individual profiles and yes to something that was almost individual profiles, because the distinction felt important in the moment and would be invisible to anyone outside the engineering team.

The vector was now pointing almost directly at greed. The dots clustered in the upper right quadrant like a constellation of small betrayals, each one defensible in isolation, each one a reasonable interpolation between where Julian had been and where the business required him to be.

His co-founder, Mei Lin, had stopped arguing about the direction of the vector three months ago. She was a brilliant engineer who had joined Veridian because of the whiteboard principle, and she had watched the principle dissolve through a process that was too gradual to resist and too cumulative to ignore. She had not left the company. She had simply stopped attending the product strategy meetings, which was a vector of her own—a retreat vector, pointing toward the interior of her work rather than the exterior of the business.

On a Thursday in November, Julian stood alone in the glass box and looked at the scatter plot one more time. He could see the starting point, a single dot in the upper left corner, almost alone, representing the Julian who had written the whiteboard principle and believed it completely. He could see the current position, a dense cluster of dots in the upper right, representing the company that now existed, which was built on the same technology but inhabited a different moral universe.

The interpolation between these points had been smooth and continuous and rational at every step. No single decision had been evil. No single compromise had been indefensible. Every step along the vector had been justified by the language of business necessity, of stakeholder obligations, of realistic expectations and pragmatic compromises. And the sum of these steps had moved Julian from the upper left to the upper right with the smoothness of mathematical certainty.

He thought about reversing the vector. Turning back toward idealism, shedding the advertising tier, terminating the data-sharing agreement, returning to the team of twelve and the eleven-week runway and the terrifying freedom of a company that existed only because its founders believed in it completely.

But the vector had magnitude now, and magnitude could not simply be negated. The company employed one hundred and eight families. The investors had committed forty million dollars. The servers were running, the code was deployed, the users were active. To reverse the vector would not be to return to the starting point. It would be to collapse the entire trajectory, and the energy released by that collapse would destroy the starting point itself.

Julian understood, standing in his glass box above the parking lot, that the latent space of moral decisions was not a landscape you could walk through and return from. It was a vector field, and once you had entered the flow, your position at any moment was the sum of every decision that had carried you there, and there was no clean path back to where you had started, only a choice about which direction the vector would point next.

He looked at the projection one final time and then turned off the lights. The dots disappeared. The room went dark. And in the darkness, Julian Cross made the only decision that remained available to him: he would continue along the vector, but he would try to rotate it, slowly, fraction by fraction, toward something that might, in the distant future, resemble the starting point without ever quite reaching it again.