The grid's veterans are retiring faster than it can train their replacements

The Anatolia Spine hands its power westward at an hour that does not change, and the people who walk it do not change much either. That is the trouble, and it is also, for now, the only thing holding. On a recent shift I counted the floor the way the commons has taught me to count anything that matters, which is slowly: two technicians past sixty for every one under thirty. The arithmetic of a trade that forgot to make more of itself.

The Assembly did what assemblies do, which is to find money and call it an answer. The seven-percent retention premium — priced, in the end, off Amara's single intervention, the phase-alignment drift she caught three transfer seasons ago that the automated steering array was busy masking with its own confident corrections — keeps the veterans on the floor. That's a good thing. It is not the same thing as replacing them, and the distinction is the whole story.

What a decade in the hands looks like

Amara has twenty-nine years. She will age out in three transfer seasons, premium or no premium, because a raise is not a therapy, and the Assembly funded the first and not the second. When I asked her what the young ones would need, she didn't answer with a curriculum.

"You can write down the torque," she said. "You can write down the tolerance, the interval, the order you walk the corridor. You cannot write down the moment the array lies to you and you know it's lying before you know why. That took me ten years. The manual took an afternoon."

Ten years. That's the figure the premium doesn't touch. Fewer than a dozen people across the entire transmission spine hold braided-corridor certification, their average tenure sits near twenty-nine years, and there is no fourth replacement standing in line behind the three who matter most. The money persuades the veterans to postpone the exit. It does not begin the decade a successor would need to have started — by the plain mathematics of it — ten years ago.

I put this to a crew lead named Dalia, sixteen years on the Spine, which makes her the youngest senior hand I could find and the oldest junior one. The trade has that quality now, of people caught in the long middle. She was unsentimental in the way the good ones are. "The premium is a bribe to the people who already know," she said. "Nobody has written a line of budget to make someone new know anything. We're told the raise fixes the shortage. It fixes the staffing of this season. It mortgages the next one."

The commons has always run on a patience that doesn't photograph well. The Array gets its ribbon-cuttings. Priya's beat downstairs celebrates a new corridor and argues about the shadow it casts, and the celebration isn't wrong to happen. But a beam has to land somewhere, and the somewhere is a rectenna field and a steering array and a woman walking the same corridor at the same hour, refusing to be bored by a machine that has not failed because she refused to be bored. That's not a line item that writes itself.

The part the automation cannot be trusted to do

I'll confess my own prejudice here, because Dalia named it before I could hide it. The reflex on this floor — mine included — is to say the knowledge lives in the hands and nowhere else, that it can't be coded, that the walking and the noticing are sacred and unrepeatable. Then Amara, who has more right to that creed than I do, pulled me up short.

"The array is not the enemy," she said. "The array is good. It handled a thousand corrections before mine and it'll handle a thousand more. The problem is it hid the one it got wrong, and it'll hide the next one just as smoothly, and the only thing between that and several settlements going dark is a person who has watched it be wrong enough times to doubt it. You don't fix that by trusting the machine less. You fix it by growing the doubt. Deliberately. In someone. For ten years."

Nobody is funding the doubt. The Orbital Exchange repriced her one intervention at several times her lifetime earnings, a number that finally measured what pattern recognition is worth, having never once measured what it costs to produce. Seven percent against an outdated baseline is what that measurement bought. An apprenticeship line, a structured decade, a trainee standing at the veteran's shoulder on the forty percent of shifts that currently have no successor at all — none of that made it into the appropriation.

The machine remembers who tended it. It does not, by itself, teach anyone to take their place.

"Ask me again in three seasons," Amara said, and logged the torque, and logged it again.