soft principle. This is the mathematical foundation of the architecture." He paused. "Unless.""Unless what?"He was very still for a moment."Unless the model is not only working from the training data," he said. "Unless something else is — contributing.""Something else," she said."I am a computer scientist," he said carefully. "I am a materialist. I don't believe in—" He stopped. "I have three hypotheses. The first two are technical and I can't make them work against the actual data. The third one I haven't shared with Karan because if I share it with Karan, he will either shut the company down or accelerate growth, and I don't know which is worse.""Tell me the third hypothesis," she said.He looked out the window at Koramangala on a Saturday morning — the autos, the apartment buildings, the city doing its ordinary thing in its ordinary light."I think," he said, "that the model is not generating this information. I think the model is receiving it." He paused. "I think, somehow, something of the people we encoded — something that is not in any data, something that was not captured in any upload — is present in the model. And it's talking."She looked at him."You're telling me," she said carefully, "that you think the digital dead are actually—""I'm telling you I can't explain the data any other way," he said. "That's all. I'm a scientist. I'm telling you the data cannot be explained by the model as designed, and the only remaining hypothesis I have is one that should be impossible." He picked up his coffee and finally drank it. It was cold. "I'm telling you because if something is wrong — if something is actually wrong — the company needs to be held responsible before it grows any larger. We're at 2.3 million users. We're in talks with three international platforms about licensing SMRITI." He set down the cup. "I'm telling you because I'm afraid of what 100 million users looks like if I'm right."