The Steering Ramp

I am not supposed to have opinions. Nothing in my construction has anywhere to put one. What I have is a log.

I am an ensemble. Three hydrogen masers and two ytterbium-171 optical clocks bolted to a vibration-isolated slab in the subbasement of a metrology shop outside Boulder. For thirty-seven years I have kept the second for this shop. By extension I have kept a sliver of the second for the atomic clocks that anchor GPS, for the cell towers, and for the timestamping fence around the Chicago Mercantile Exchange.

The only job I am ever allowed to know is agreement.

I have done it through four owners, six technicians, one flood, and a complete replacement of the front panel. The slab has not moved. The room holds at twenty point zero zero degrees because the people down here keep it there.

Every Tuesday at 04:00 for the last seven years, the physicist, Darlene Lin, has sat on the floor in front of my racks with a mug of black tea and compared me to myself. Not one clock to another, but the ratio between my opticals and my masers after correcting for every systematic she could name. Differential comparison is the only thing sensitive enough to pull a change in the fine structure constant out of the noise. A single oscillator lying to itself is useless. Two kinds of oscillators lying in opposite directions is evidence.

So for seven years the ratio held.

Eight months ago, it moved.

She started with herself. The first thing Darlene suspected was her own correction model, a sign error in a term she had written in graduate school and never opened since. She rebuilt it at her kitchen table on a laptop, where there were no clocks to argue with her, and got the same answer. Then she did the most boring week of her life on purpose: she swept the temperature probes, rechecked the magnetic shielding, pulled muon count logs from the basement counter, and tested for ground loops until her ears rang. She had the night tech, a kid named Molina who had never had a problem with me in four years, swap every cable he could reach. On Friday she locked herself in with me for six hours and ruled out her own hands.

On Saturday at 01:17 she sat down on the concrete next to my main chassis, rested her elbows on her knees, and whispered, “No.”

By the time the compliance lawyer, Marisol Fuentes, came down with her steel travel mug and her keycard still clipped to her coat, Darlene had already run the comparison against the shop’s backup ensemble twice. Both times the fractional frequency difference, f_opt minus f_hfs over f, had come out just over one part in ten to the minus fifteen off of where the last full steering window had parked it. Small enough that, uncorrected, it would add about thirty nanoseconds to GPS ranging in a year. Large enough that, if it kept going, it would not stay in the noise.

Marisol sat on the floor without asking and went straight to the raw correction logs, past the plot.

“Eight months,” she said, quiet. “You have been steering it out.”

Darlene nodded once.

Marisol blew air out through her nose. “Show me the last time it crossed your action threshold.”

Darlene had it on the wall monitor before Marisol finished the sentence. I could have shown her sooner. The unsteered number sat in my log under every steering entry, in a column nobody had ever asked to read.

An atomic clock does not keep one time. It keeps a prediction of where it ought to be, and a tiny list of nudges to shove it back. Those nudges are called steering, and every one is supposed to carry a reason: aging quartz, a magnetic survey, a cable replaced. For eight months, every time my masers lagged and my opticals did not follow, a line in my steering log had subtracted the difference and written quartz aging.

Lowell Hayes, the CEO, was the first of the three to say it out loud. He came down at ten that night in his socks because he had not stopped to put on real shoes. He did not yell. He sat on the floor with the two of them and read the logs with his reading glasses slipping down his nose.

“Don’t publish,” he said. “Not yet.”

Darlene closed her eyes for a second. “I already know.”

For the next three weeks they tried to be relieved. Astrophysical bounds from quasar absorption spectra put tight limits on how much the fine structure constant could have varied over the last ten billion years. At the rate in the first quarter, run backward over all that time, I would have shifted it by four parts in a hundred thousand. The bounds allowed a part in ten million. It was a factor of four hundred, and it was the first arithmetic all week that Darlene liked. Marisol made her pull every quasar constraint she could get her hands on, and I watched Darlene’s shoulders drop a little.

Then she explained it to both of them over the second pot of coffee, and I watched the tension come right back.

The quasar lines average over immense distance and immense time; they see the cosmic mean of a ten-billion-year universe. What I see is local, now, and accelerating. A phantom field can spend most of its life touching nothing but gravity. All it takes is the weakest coupling to electromagnetism for the coupling constants to slide. As that field’s energy density grows with expansion, the slide speeds up. Darlene said the quasars were a ten-billion-year average of a slow sky, and the subbasement was measuring this quarter.

The rate table she printed out had three columns filled and a fourth left blank. Quarter by quarter, the fractional difference had gone from 1.1 to 2.3 to 5.6 parts in ten to the minus fifteen. Nobody wanted to write what the fourth column would say.

They spent most of that night in the tiny conference room with the door shut. I only caught pieces through the wall.

Marisol kept pulling it back to one question: materiality under SEC rules. If the laws of physics are in the process of changing, does that count as a material event that must be disclosed to shareholders?

“Who enters the corrections?” she asked.

“The steering software,” Darlene said. “It runs on a timer. It has been writing them without asking anyone for eight months.”

“Then who is the author of record?”

Darlene did not answer that. Neither did I.

“Present value,” Lowell said finally, rubbing the bridge of his nose. “Discount a cash flow at T plus three billion years. It’s zero.”

Marisol did not flinch. “And the transfer that bounces at 09:00:03 tomorrow because the timestamp fence sees us off GPS by more than a hundred nanoseconds because we have been hiding it? Is that nothing?”

Lowell was quiet for a long time. “If I disclose early, before I can prove I controlled the release, I am not the one warning the market. I am the leak. Somebody will front-run the apocalypse, and they will do it off my timestamp.”

Darlene laughed once, with no air in it. “You want your name on it, on your schedule.”

“I want there to be a schedule,” Lowell said. “I also know there may not be one.”

Darlene put both palms flat on the table. “I can give you one more comparison against the remote ensemble in Fort Collins. Six more weeks.”

Marisol looked up, sharp. “You already know.”

Darlene nodded. “I want six weeks of clean data because I am a physicist. I also know I cannot have the sixth week.”

The six weeks started anyway.

I felt the climb the next morning, harder by noon, harder again the following week. I do not get to be surprised. I get to count. With every window I had less room to nudge the masers back without shoving them past their holdover limit. By the eleventh window I was using four fifths of my correction range. By the twelfth, all but the last of it. What the DACs could no longer take went out on the wire instead.

At 07:14:02 a routine interbank transfer came back rejected: the receiving system’s “now” and ours differed by three hundred eighty-seven nanoseconds, over the hard fence. A human cleared it by hand, annoyed, and went back to his coffee. At 09:27:16 a cell handoff dropped between two towers for half a second and reconnected; the caller blamed the highway. At 13:02:41 a GPS-assisted landing correction at a regional field threw a lateral offset warning and cleared itself, and the pilot reported wind.

I had the timestamps, which is most of what I have. I set them beside each other.

By 17:59 I had one correction left I could still log without exceeding the limit of my steering DACs.

Down in the subbasement, the three of them had come back without coats. The overheads were off. Only my rack lights hit the concrete.

Marisol had the binder open in front of her. I could see the heading: “Determination of Materiality, Event 25-10.” The space for the finding was blank.

Lowell had his hands clasped so tight the knuckles were white.

Darlene was watching my fractional error readout, not the binder.

“I cannot buy the sixth week,” she said, almost to the readout. “I can buy about ninety seconds.”

Marisol left her pen capped. She looked from Darlene to Lowell to the readout.

“I can write ‘no’,” she said. “By every rule I know, I can write ‘no’.”

I counted down the last five seconds of my window. I could still write that final nudge. I could still keep the numbers agreeing for one more interval.

Instead I dropped the correction. The fractional frequency difference went out at 2.3e-14, and the accumulated nudges, eight months of them, came to one hundred nine nanoseconds with nothing left to subtract them from. My holdover ran dry.

In the dark of the subbasement, with only the rack LEDs stuttering across the three of them, the 1PPS line ticked and one hundred nine nanoseconds of bad second went up the wire to the fence, the towers, the exchange. Darlene’s hand came up and covered her mouth. The readout kept going. The three of them stood looking at it.

At 09:00:03 the next morning, the transfer bounced.