第31章 The Feeder

The feeder was not a company. It was not a server farm or a data center or a network of servers spanning multiple states. The feeder was a person — or rather, a network of people, scattered across the country, working in jobs that required no technical expertise and offered no access to the systems that powered the machine, performing tasks so mundane and so invisible that no one had ever thought to ask what they were feeding. Daniel discovered it by following a thread that Priya Sharma had left six months ago, a postcard that had arrived at the Fund's office with three words: *You found the hub. Now find the feeder.* He had spent weeks looking for a building, a server, a facility — something physical, something that could be pointed at and photographed and included in a court filing. But the postcard had been wrong. Not the destination — the nature of the thing he was looking for. The feeder was not a place. It was a practice. A habit. A routine so embedded in the legal system that no one noticed it because everyone participated in it. He found the first feeder on a Tuesday, in a courthouse in Cleveland, in a room that was not a server farm but a copy room — a windowless space on the basement level where court reporters typed up transcripts and filed them with the clerk's office, where the paper was heavy and the ink was fresh and the work was slow and deliberate and utterly unremarkable. The woman who worked the copy room was named Diane Marsh, no relation to the special prosecutor, and she had been transcribing court proceedings for thirty-two years, producing ten thousand transcripts per year, each one a perfect, painstaking record of every word spoken in every courtroom under her jurisdiction. Diane did not know about Axiom. She did not know about Sentinel, Meridian, Vantage, or any of the companies that formed the machine's infrastructure. She knew only that the transcripts were due by Friday, that the clerks needed them certified, that the judges expected accuracy, and that the system worked because people like her did the work quietly and without complaint. But the system also worked because Diane's transcripts were fed into a pipeline she could not see — scanned, digitized, parsed, and sold to a data trust that trained algorithms on the language of the courtroom, on the patterns of argument and objection and verdict, on the thousands of hours of spoken word that constituted the raw material of legal prediction. Diane was not a programmer. She was not a data analyst. She was a court reporter who typed fast and made few errors and went home to a cat and a small apartment and a life that was entirely separate from the machine that consumed her work. Daniel sat across from her in the copy room, watching her type — her fingers moving across the keyboard with the fluid grace of someone who had performed this motion ten thousand times, who had never thought to wonder what the words were feeding, who had never considered that her labor was being harvested, processed, and sold back to the system she served. "You know about the transcripts?" he asked. "I know they need to be filed," she said. "I know the clerks need them by Friday. I know the judges expect accuracy. I don't know what happens after that." "That's the point," Daniel said. "You type the words. The machine reads them. The machine learns from them. The machine uses them to predict the next case, the next verdict, the next deal — and then it sells that prediction back to the courts that produced the words in the first place. You are the feeder, Diane. Your work is the fuel." She stopped typing. She looked at him, and for a moment he saw something in her face — not fear, not anger, but the slow, dawning recognition of a woman who has spent thirty years doing work she thought was meaningless and discovers, too late, that it was anything but. "What do I do?" she asked. "Nothing," Daniel said. "Your work is honest. The system is not. But you can help me expose it. You can help me show a judge that the machine is not predicting justice — it is writing it. And the only way to stop it is to break the pipeline." Diane was silent for a long moment. Then she nodded, slowly, and returned to her typing — not because she had accepted her role, but because she had accepted her responsibility, and the two things were not the same. He spent the following weeks visiting feeders — court reporters, transcriptionists, clerks whose work recorded what happened in courtrooms and who had never been told that it fed a machine. In Chicago, one reporter laughed when he explained. "They told us our transcripts were used for training materials," she said. "I pictured textbooks." "Your words teach a system to predict what happens in courtrooms you've never visited," Daniel said, "for cases you'll never attend, involving people you'll never meet. It learns the language of the courtroom — the patterns of argument, the rhythm of objection, the cadence of verdict — then sells those predictions back to the courts that produced your words. You are feeding a system that uses your labor to predict your replacements." The laughter drained out of her face. "That's not right." "No. But it isn't illegal either. Transcripts are public records, and predictions built on public data break no statute. That's the whole problem. The system is legal. It simply isn't legitimate." Twelve courthouses in six states, and everywhere the same pattern: pride in the work, blindness to the pipeline, and a quiet assumption that legality and legitimacy were synonyms. The feeder network was the machine's circulatory system — invisible, essential, taken for granted. And circulatory systems can be redirected. He flew home with a plan. Not to shut twelve thousand feeders down — that was impossible, harmful, and contrary to everything the Transparency Project stood for — but to redirect their flow: transcripts routed to the public archive instead of Vantage, readable by everyone instead of consumed by an algorithm that sold predictions back to the courts that produced them. The pipeline existed. The software worked. What he needed now was participation. Participation was logistics, and logistics was a war he knew how to wage. ---