The ninth day after the memo went out. The two lines in the shared table were still blank, and on the blank of the third line something new had appeared: the Miyun station office had attached a form to their reply, wanting her to fill in export format, medium, and estimated completion date. She didn't fill it in. She didn't reply.
Julian date 2459521, Thursday.
On the right side of the whiteboard she wrote a third line: go back to 2004.
The pretext for that line was four words in Ursula's email: thirty years of data. The oldest thing in her hands was the tape from March 2004. The spike that year had been entered in the records as a sickness of the No. 7 feed; once entered, no one mentioned it again.
She went first to find out how it had been entered.
The duty logs were in the archive room. The 2004 volume, the paper gone brittle; a small corner flaked off as she turned the pages. She picked up the corner and tucked it back where it belonged.
March 19 to 26, eight days in all. The pages for the 19th and 20th were clean. On the page for March 21, in the remarks column, one line in ballpoint: No. 7 feed noise anomaly, to be checked. Written lightly, the strokes fading through the second half—the pen had been running dry.
The 22nd through the 25th: blank. The page for the 26th held the routine entries and never mentioned it. After March 26 no one ever wrote of it again.
Below that line, in another hand, scrawled: Noted.
She asked Old Qiao: who is this "Noted."
Old Qiao put on his reading glasses and looked a long time. Not one of ours. Thirty years, I'm the only one who keeps this book.
Then who.
Could be someone from above. In the nineties, the early two-thousands, people came down from above a few times to check the stores. When they were done they liked to write in other people's books.
She asked: why would anyone checking the stores read a duty log.
Old Qiao took off the glasses. Who knows. If they could read it they read it. If they couldn't, they read it anyway.
She photographed the page.
Of the 867 events in 2021, the median duration was nineteen hours. By the broadening trend line, 2004 should have been one sixth of that: a little over three hours.
She took that number and built a template, narrow, three hours wide, and ran matched filtering with it over the eight days of data.
Positions where the output exceeded three times the noise: nineteen.
She then ran a thousand trials of simulated noise of the same length, same template, same threshold. Mean 4.3, standard deviation 2.1. Trials out of a thousand that reached nineteen: zero.
Hu Yisu said: nineteen against four point three.
She said: don't put it that way. Nineteen against four point three, standard deviation 2.1—work it out yourself.
He worked it out: seven sigma.
Eight days of data, she said. Eight days of data has no business talking about seven sigma. It's entitled to say one thing only.
Seventeen of the nineteen events could go into a log-log size distribution, four bins. Slope minus 1.6, error plus or minus 0.4.
Hu Yisu said: that error is huge. Take it out and they'll shoot you.
It's not going out, she said. This number is for my eyes only.
In the notebook she wrote: not a proof, not a contradiction. Seventeen points, an error this size—only enough to say it doesn't fight with the other one.
Then she laid the 2004 matched-filter output side by side with the 2021 stacked curve. Same position. The one from seventeen years ago narrow, the one now six times wider, the pitch unchanged—two notes out of the same throat, one narrow, one wide.
She began the reckoning of the common ancestor.
The clocks of 2004 were rubidium plus GPS timing; now, hydrogen masers plus the BIPM chain. Ephemerides: DE405 back then; now she used DE436 and Hu Yisu used DE440. Software: a Fortran pipeline written in-house, retired in 2011. The dust maps were an older version. A different dispersion correction. A different leap-second table.
Of the nine shared inputs, 2004 and 2021 overlapped on only one: the planetary ephemeris, different versions from the same house.
Hu Yisu asked: is the source code still around.
Old Qiao rummaged in the archive room a good while and turned up a CD case, the label handwritten: timing pipeline 2003–2008 v2.1. The disc read fine. Hu Yisu took it back, installed a set of old run-time libraries in a virtual machine, and got it running.
The day it ran he showed her the screen: Shijie, even the comments in this thing are in Chinese.
How long per run.
Forty minutes, he said. Twenty times slower than mine.
Then run it twice.
She quantified the one input left over. The Earth positions DE405 and DE440 gave for March 2004 differed by twelve meters. Twelve meters is forty nanoseconds. Forty nanoseconds cannot build a thing three hours wide, and cannot build its shape either.
Hu Yisu said: so the common ancestor is dead.
Dead down to one last hair, she said. And I've measured the hair's length.
He ran the Fortran too. One pass of the old pipeline in the virtual machine took forty minutes; he ran it twice: the first pass on the raw voltages she gave him, the second on a version he had resampled himself from archived scans, one sampling rate lower. Both passes produced the spike, the positions less than half an hour apart.
He said: Shijie, two sets of software eighteen years apart, neither copying the other, and out comes the same thing.
She said: yes.
She said the word and waited for the feeling that ought to come, and it didn't. She had assumed it would.
Friday she went up the tower once.
Eli climbed with her. At rung three hundred she stopped and held the railing. In her right ear the hiss pushed up a notch from the lower altitude, up to the mark she had been keeping for half a month, and held there.
Eli waited above her and didn't rush her. After about twenty seconds he said: you stopped here last time too.
Mm.
Ears?
She didn't answer. The hiss sank halfway, stopped midway. She kept climbing.
The No. 7 feed sits in the northeast corner of the cabin, horn facing down. The stain was still there. She crouched and looked through the borescope, a long look. Eli said: it won't wipe off. I've tried three solvents.
She said: stop wiping.
Stopped for good?
Stopped for good, she said. Leave it.
While she photographed it she was thinking of something else: thirty years ago it was already here. Thirty years ago the duty operator on watch could never reach it, didn't know what he was looking at, and still less did he know that the line he wrote on the page for March 21 would be buried for seventeen years beneath a scrawled "Noted."
Coming down, the wind shoved flat across the gobi, and the tower frame gave off that very low note underfoot. She stopped and listened a moment. Still there.
Eli was waiting at the bottom. She said: Master Eli, March of '04, those days—who was on duty, do you remember.
Eli shook his head: in '04 I was in training in Ürümqi, I told you. By the time I got back those log pages were already signed off.
Signed by whom.
By me, he said. I came back after my shifts and signed them after the fact.
She said: oh.
He said: why are you asking.
She said: someone wrote "Noted" on those pages.
Eli thought about it: not me. My handwriting was better than that back then.
Julian date 2459522, Friday, 9:10 p.m.
She telephoned He Zhaoting.
She told him three things: 2004 had it too; the width was one sixth of today's; the delay couldn't be measured from a single station—there was only the one dish, Kashgar.
When she finished she waited.
He asked: how much is the delay.
It can't be measured from one station, she said. In 2004 there was no second dish. No baseline.
No sound from the other end. She glanced at the screen; the call was still counting. She got to somewhere past twenty seconds and was about to say his name again.
He said: Xiao Zhou.
She said: yes.
He said: do not call me about this again.
Then he hung up.
Call record: forty-seven seconds.
She sat in the chair without moving. The paper notebook fell open to the Beijing page; the number in the corner was still there: 2439954. She pressed a finger on the number and held it; under the fingertip she could feel the ridge it left on the back of the page.
The tape was still in the drawer. She did not take it out.
She called back. No answer.
The next day Old Tang telephoned about something else, and when that was done he mentioned in passing: Old He called me the day before yesterday. Asked whether you'd returned his tapes. I said no. He said, then let her keep them.
Just that?
Just that, Old Tang said. He didn't ask how the checking was going, didn't ask about the data. That old man has asked about nothing but delay his whole life, and this time not one word of it.
Julian date 2459523, Saturday.
Lu Weiming's email arrived in the morning.
Zhou Yan: your two lines are still blank. If you don't fill them, others will fill them for you. Nanshan has proposed adding three participants to the revised memo, on the grounds that they supplied a version of the data. Miyun wants the station office into the acknowledgments. None of that is the bad thing. The bad thing is that you still think it's yours. This is not your discovery. It is everyone's disease. I wrote that sentence a month ago; I am writing it again now.
She didn't reply.
On her phone, one from Dr. Shen: three months now. That hearing of yours. When are you coming in.
She replied: another day.
The twelfth time.
In the afternoon Old Tang asked how to word the annual progress report. She said: write "under investigation." He said: last year was "under investigation" too. She said: then write "still under investigation." On the other end he gave a grunt.
He paused, then went on: Nanshan's list added another name yesterday.
She said: let them.
Zhou Yan, he said. Every line you strike through now strikes out an instrument and leaves you standing alone. Those people aren't adding their names because they understand. They're adding them because they can see you can't hold on to it much longer.
I know.
You know, he said. Knowing and doing are two different things.
At eight, the video call with Zhou Yao.
Did your one thousand forty-two change later.
It changed.
To what.
The 1,042 didn't change. Something else did.
What changed.
She said: it gained a line of words.
Zhou Yao waited. She didn't go on.
Zhou Yao said: oh.
Then she asked: Mom, do stars sleep.
I don't know.
You watch them every day.
What I watch every day isn't that.
Then what do you watch.
I watch whether they move at the same time.
Zhou Yao thought a moment: if they move together, does that mean they're awake together?
Maybe.
Zhou Yao said: oh.
After they hung up she wrote the sentence into the paper notebook, and when it was written she didn't know why she had recorded it.
She flipped back four pages, to a line from last month: Zhou Yao asked what that star is called.
She capped the pen and went to the kitchen to boil water.
In the tank Lao Wu was still under the driftwood. Water temperature twenty. The third notch, deeper than the first two.
Julian date 2459524, Sunday, through 2459526, Tuesday.
She downloaded the Planck 353-GHz polarization maps, and a batch of optical polarization measurements of stars. The two sets sat on an institute server; she pulled them in two goes—the first died at sixty-two percent, and the next day Hu Yisu redownloaded in the machine room, finished, and the checksum failed, so he did it all over again.
When he pushed the data to her he said: Shijie, what are you going to look at.
Which way the magnetic fibers grow.
He said: that stuff has been lying in the archive for a decade and a half. Everyone's downloaded it. No one's ever looked.
She said: exactly.
Of the 1,042 stars, 968 had Gaia distances. Pairwise: 468,028 pairs. She kept only the pairs with three-dimensional separations greater than three hundred light-years: 413,000 pairs.
For each pair she took the direction of the line joining the two, and the orientation of the magnetic fiber at the midpoint, and computed the angle between them. The joining lines were three-dimensional directions, not great circles on the sky—a degree or two of difference, so she did it both ways; the conclusion was the same.
If the distribution were isotropic, the angles would spread uniformly between zero and ninety degrees. Binned at fifteen degrees, six bins, 16.7 percent each.
Measured: 34.2, 21.6, 16.1, 12.4, 9.8, 5.9.
Her first thought was selection effect: bright stars crowd the galactic plane, and the fibers are ordered to begin with. She ran two controls. First, she shuffled the positions of the 968 stars and reassigned them; the angle distribution fell back to around 16.7, the six bins reading 17.1, 16.9, 16.6, 16.9, 16.4, 16.1. Second, she kept only pairs with galactic latitude above thirty degrees: the first bin still held 31.8.
She checked whether it could be a projection effect of dust-column density, checked whether it tracked the brightness difference of the two stars. Neither.
Hu Yisu asked her: could it be that the stars you people picked were on the same fiber to begin with.
She said: if that were so, they should be bunched together.
She plotted the 968 on the sky. Spread wide, both hemispheres, no clumping she could see. She also computed the angular-distance distribution among all 968 and set it against the angular-distance distribution of 968 drawn at random from the same catalog; the two curves coincided.
Tuesday night she copied the six numbers onto paper, and when she was done she held them up against the black outside the window for a while.
Nine hundred sixty-eight stars, joined pairwise, and the joins like to run along the magnetic fibers.
Julian date 2459526, Tuesday, morning. She filled in half of the Miyun station-office form. Export format, written. Medium, written. In the column for estimated completion date she stopped a long time, and in the end entered: to be set, reported separately.
Filled in, she didn't send it. Saved to drafts.
Opening her mail she glanced, in passing, at another message. Subject of one word: Shoulder. Recipient blank, body blank, last modified 2459483.793.
She closed it.
She wrote back to Vienna one line, body a single word: no.
In the paper notebook she wrote the last line: 0–15° bin, 34.2%. Isotropy would give 16.7%.
Cap on the pen. In the drawer the tape still lay under a stack of printouts, with no machine to read it.