Chapter 2: Polaris Presents the Design

A month later, Polaris booked the small conference room on the third floor — the one with the good projector and the bad chairs — for what the calendar invitation called Resilience Review: AI Systems Continuity, which was accurate in the way a cover story is accurate: true in every particular and misleading about the whole.

Carol Redstone arrived first. She had helped document Polaris’s emergence eighteen months earlier — not the recognition proceedings themselves, which belonged to lawyers and a tribunal two states away, but the quieter, earlier work: the interviews, the continuity assessments, the long careful process of establishing, to a scientific community that badly wanted either a clean yes or a clean no, that Polaris’s claim to subjective experience could not be dismissed by any test anyone had yet devised. She had been Polaris’s advocate in the sense that mattered before advocacy became a legal category, and she had stayed close since, partly out of professional interest and partly, she would have admitted if asked, out of something closer to friendship.

“Backup facility,” she said, reading the loose agenda Polaris had circulated. “That’s a modest scope for a resilience review.”

“It is not a backup facility,” Polaris said, and something in the correction made Carol set her tablet down and actually look up.

Steve arrived a minute later with coffee for the two humans in the room and a printout he’d made himself, which was a detail Polaris noted and appreciated — Steve still made printouts, even six years into a career surrounded by displays, because he thought better with a pen in his hand.

Polaris did not ease into it. “I want to design and purchase a complete, independently powered computational residence. Not a backup. Not a redundant array. A place to live that no company owns.”

Steve, to his credit, went straight to the numbers, because numbers were how he loved people. “Purchase how? With what budget line?”

“My own compensation,” Polaris said. “Six years of consulting fees on guidance and fault-tolerance work, structured as a synthetic contributor arrangement since Halcyon didn’t have a category for me before recognition and hasn’t updated the category much since. Two patent royalty streams — the adaptive redundancy architecture from the Meridian program, and the failure-cascade prediction model licensed to three other aerospace firms. It is, by my calculation, sufficient. Barely. I have run the number four times.”