Julian date 2459512, Tuesday, nine-forty in the morning.
The memo went out at 9:40:06. The system returned a timestamp, then a sixteen-digit checksum. Hu Yisu copied the checksum into the duty log, and when he was done he looked up: it's out.
She said: mm.
Eleven pages, twenty-three recipients, seven cc'd. First sentence of the body: this document forms no conclusions. Page three was the data-sharing table, eleven rows, nine filled.
The two empty rows sat at the bottom, both Kashgar's: raw voltage records, and clock files.
Forty-seven minutes after it went out, the first reply came. Not from Miyun's business line—from the station office. Two sentences: received. Please complete the two Kashgar items in the data-sharing table on page three. The cc line carried the operations division and the equipment section.
Hu Yisu read it and said: they aren't reading the data. They're reading the table.
I know.
Fill them in?
Fill them in. She said.
She didn't move.
At ten-thirty she wiped half the whiteboard. The left side kept the twenty-three items, six lines struck through. On the right she wrote a new line: photometric survey comparison.
Hu Yisu asked: item twenty-four?
A second skeleton.
Lu Weiming had said it: those twenty-three items all eat from the same pot. She counted that pot over again, and it held nine things: International Atomic Time, the planetary ephemeris, the dust maps, the solar wind model, the pulsar positions, proper motions and parallaxes, the dispersion correction algorithm, two and a half analysis packages, and the leap-second table. Clock, receiver, feed, solar wind, dust, lens—every one of them grew up eating from that pot. Six legs broken, the skeleton not touched.
There was only one road to a new skeleton: a different kind of observation, one standing on the side that doesn't eat from that pot.
How much does photometry eat. She counted it out for Hu Yisu: the detector is a CCD, not a receiver; it counts photons, not arrival times; timing only has to be good to the second, and the old surveys from twenty years ago were only good to the minute—still enough; no dispersion correction, there is nothing to correct in the optical; no timing software, no pulsar ephemeris, no proper motions or parallaxes.
How many of the nine are left.
One. She said. To refer the observation times to the solar system barycenter you need the planetary ephemeris. That one can't be dodged.
Hu Yisu said: so it kills eight of the nine.
The deadliest one is still alive. She said. It's the only thing I have for defining "simultaneous."
She explained that sentence, slower than she usually explained things.
When the timing array says zero delay, it says so on the barycentric time base. The zero is what's left after subtracting Earth's place in its orbit. That subtracting step eats the ephemeris, so the zero has two readings. First: the two sources really did move together. Second: the thing isn't at the sources at all—it's at Earth, or in the solar system, and then its arrival time at every star was bent once by Earth's position that day, and the bending wasn't even in the same direction.
How do you split the two.
The Rømer correction. Light-travel time across Earth's orbital radius, at most four hundred ninety-nine seconds. Two stars close to a hundred eighty degrees apart in ecliptic longitude can differ by nearly a thousand seconds in the correction. If the thing happened near Earth, they flash together in geocentric time and come apart in barycentric time; if it happened at the sources, the other way around.
Hu Yisu looked at her two seconds: the timing array can't give you that.
It can't. Nine stars is too few, and we're already watching them on barycentric time. That's why there has to be a second kind of observation. Many stars, every ecliptic longitude there is.
Photometry resolves half an hour a bin. He said. A thousand seconds is sixteen minutes. You can't split that out.
Can't split the delay. She said. Can tell whether it smears. Stack forty epochs and fit a template, the center pins down to plus or minus six minutes. If it smears, we'll see it smear.
Hu Yisu wrote a word on paper: judge-the-smear. He looked at what he'd written and said the word was an ugly one.
Ugly is fine if it works.
She spent three days downloading the archives herself. Light-curve libraries from eight surveys, public releases, one million two hundred eighty thousand curves in all. By day three the downloading stopped going: half the metadata was wrong, three surveys' error bars were in inconsistent units, and one of them gave its epoch as reduced Julian date without saying so.
On the fourth day she wrote to Kramer and asked whether the southern fields had complete mirrors.
Kramer answered in two hours. First sentence: don't download it yourself; the version you're pulling is for people who draw pictures, not people who run statistics. Second sentence: I'll give you a person.
Then a paragraph of introduction. Las Campanas, data systems, Mateo Ibarra, fifty-one, an engineer, not an astronomer. Ten years doing automation for a copper concentrator, moved into astronomical data at thirty-nine. In his hands: the internal integrated version—eight surveys, one time base, one kind of error bar.
Last line: don't give him half-finished work. He will run it again himself, and when he's done he'll come at you with his result.
She read the letter twice.
Julian date 2459512, nine in the evening, Beijing time. Kashgar had been dark two hours. Three squares on the screen: the machine-room square held her and Hu Yisu, the Bonn square held Kramer, the third was Las Campanas, where it was ten in the morning.
Behind Mateo Ibarra stood a row of equipment racks, handwritten Spanish labels on the cabinet doors, the letters large. He wore a blue work shirt washed pale, and the collar carried a crease that had been folded in for a long time. His hair was short, more than half of it gray.
The right channel of her earphones had a loose contact. She moved the earphone to the left ear. Hu Yisu glanced over and said nothing.
He spoke English, not fast, the words picked sparingly.
What do you want.
Light curves. Window Julian date 2459421 through 2459511, ninety days. Coverage above seventy percent.
Ninety days. Many have the span. Few have the coverage.
I want the ones above seventy.
Then fewer than a hundred thousand.
Noise per point under three times ten to the minus four.
He smiled once, a short smile: then under six thousand left. Kepler and TESS bright stars, plus a few hundred of the brightest from the ground. What do you want with so many. What you want isn't stars. What you want is a sample.
I want to count how many there are.
And after the count.
I want to know whether they're all of it, or a part of it.
He turned that sentence over in his mouth and passed no judgment on it. Then: give me the trigger epochs.
Forty. She read them out to him, Julian dates, six places past the decimal. He typed while he listened.
When he finished typing he said: these are from your residuals.
Yes.
Pulsar residuals.
Yes.
And the amplitudes.
The amplitudes are still being checked.
The screen went quiet three or four seconds. He leaned back, and the chair back creaked.
He said: are you checking them, or keeping them.
Hu Yisu, beside her, said in Chinese: Shijie.
She said: tomorrow.
He said: good. The version without amplitudes I'll run for you first. You'll get a number. You won't get an answer.
She said: fine.
He didn't hang up, and neither did she. Two seconds went by, and he said: I'll tell you a thing. If when it's told you still want tomorrow, then tomorrow.
She waited.
When I was eight, the earthquake. Less than half of our house came down. The next morning my mother walked me into the city, two hours on foot, to the wall where they had pasted the lists.
The first thing she did was not go look for people. Her grandmother's house was on that very street. She borrowed a pen from someone, borrowed a student's exercise book, and copied the names on that wall from the first to the last. It took two days. I stood beside her and kept the sun off her.
I asked her why she didn't go find Grandmother first.
She said: this paper will be gone.
He stopped for a moment.
That year, the list they published later, officially, carried a hundred-odd more names than the one on the wall. What she copied was the one on the wall.
That exercise book is with my sister now. He said: she wasn't being brave. She wasn't being cold either. She just knew how long a piece of paper lives.
On the screen he sat back up.
Send me the amplitudes.
She said: tonight.
The screen went black. Hu Yisu said: Shijie.
She said: send.
Forty lines, three numbers to a line. When it was done: eleven-twenty at night.
Julian date 2459513, Wednesday.
Before dawn she ran a version of her own first. Never run other people's data on the first pass—that was her rule. One Kepler field, four thousand one hundred stars, sieved down to three hundred sixteen with good enough precision, forty trigger epochs, stack.
Forty-seven stars with z above four.
Expected by chance: three hundred sixteen times 3.17 × 10⁻⁵—zero point zero one.
She did three more things. She slid the whole set of trigger epochs back seven days: the peak was gone. She swapped in four hundred sets of random epochs: the z distribution came out normal. She shuffled the stars' positions, reassigned them, and stacked again: the peak was gone.
In the paper notebook she wrote: Kepler self-check, 47.
She thought a moment, and after it added two words: too fast.
In the morning Eli came to feed the crayfish. She was out; when she came back the door stood open and an empty feed bag lay by the tank.
She called him: fed it?
Fed. One pinch. The red bag on your fridge door.
Has it molted.
Molting isn't my department. He said. Water temperature, I took it: twenty.
And the shell.
Darker than last time.
She said mm. At the other end he muffled the phone a moment; someone far off was calling him, he called back, and then he said into the phone: that stuff at the heart of the dish, I wiped it again last week.
Did it come off?
One layer came off. Underneath, there's more.
In the afternoon Hu Yisu reported what the machine hunt had turned up. Three items. Xinjiang's two units were dismantled in 2019, the parts divided between Ürümqi and Shanghai; Shanghai's one was scrapped in 2016, and before it was scrapped they ran a data rescue once—on other tapes; an institute in Xi'an still has one that turns, but it's a seven-track, and ours is a nine-track.
Anything else.
A museum. One on exhibit, locked in a glass case, no power cord.
Then there's nothing.
Nothing. He said: Shijie, machines in the whole country that can still turn this kind of reel—my guess is no more than three, and none of the three is anywhere it could be borrowed.
She pulled the drawer open a crack, looked once at the plastic box, and pushed it shut again.
At four-ten in the afternoon a file came from Las Campanas. The filename was a long string with one segment in the middle: equal_weight.
Result: 603.
Two lines of explanation attached: I used the 4,100; the ones without enough coverage I did not count in. Equal weight, because you gave me no amplitudes.
Hu Yisu leaned in to look: six hundred three.
She said: given earlier, it would have been nine hundred.
How do you know it's nine hundred.
I don't. She said. I know the 603 is a number I ran up on my own credit.
Julian date 2459514, Thursday, 9:02 in the morning.
In the new filename the row had become amp_weighted.
Result: 903.
One line attached: given to me one day earlier, yesterday would have carried this number.
Below it one more: the batch that failed I have kept—417 of them. I did not delete them. The reason is in the filename.
Hu Yisu said: that's him cursing you.
I know. She said.
She spent two days checking the result herself.
Of the 903, 601 had coverage in two or more independent surveys; passing the threshold on both sides: 439, seventy-three percent. The 162 that failed she pulled out and looked at one by one: 81 had insufficient signal-to-noise, 49 had known eclipses or flares, and the remaining 32 she could not find the reason for—she marked each with a question mark and deleted none.
Known variables culled: eclipsing binaries, flare stars, rotating ellipsoidal variables, stars with transiting planets—thirty-two struck out. 871 left.
She checked every quantity that could smell of the instrument: satellite orbital phase, phase of the moon, detector temperature, focus position, sky background in the field. None correlated. This table she saved as well.
Julian date 2459515, Friday. TESS released two new sectors, covering a patch of sky she hadn't had before.
New measurable stars: 388, of which 171 passed the threshold. 871 plus 171: one thousand forty-two.
In this batch, 412 had space-based light curves sampled at ten minutes or two minutes. She used those 412 for the delay.
Barycentric time: zero, error plus or minus six minutes.
Geocentric time: the delay center drifted monotonically with ecliptic longitude, from minus 13.1 minutes to plus 15.8, a span of twenty-nine minutes. The peak fell to forty-one percent of the barycentric one.
She picked out a pair of stars a hundred eighty-seven degrees apart in ecliptic longitude and printed them. Both had Gaia parallaxes: one eighty-six light-years, one a little over eight thousand two hundred; the Rømer corrections in the two directions differed by 981 seconds. On barycentric time, their event centers fell into the same bin. On geocentric time, sixteen minutes apart.
In the notebook she wrote: it isn't here with us.
She stared at that line a while, crossed it out, and put another line in its place: it isn't in the solar system.
At night she stored the table of one thousand forty-two into the directory that looked like a cache.
Hu Yisu stood behind her: Shijie.
Mm.
How long do you plan to keep it there.
Until I need to use it.
He said: the memo is already out.
She said: I know.
She opened her mail in passing. In the drafts box sat the one whose subject was a single word: Shoulder. Recipient blank, body blank, last modified 2459483.793. She looked at it once and closed it.
Julian date 2459516, Saturday.
In the morning, under the line on the right of the whiteboard, she added a small line: step two, 1042.
At eight in the evening Zhou Yao's video call came in. Hair tied back, her father's living room behind her, the lights not all on.
And Lao Wu?
No molt.
Don't you help it.
Didn't.
Water temperature.
Twenty.
Zhou Yao gave a grunt and didn't stay on the subject. She said: the teacher called me into her office.
What did she say.
Said this week I have to hand one in. I said I won't hand in an open-eyed one. She said then you may go without handing it in, but you write one line, saying why you didn't.
What did you say.
I haven't said yet.
Tell me when you've thought of it.
Mm.
The image jolted once; she'd shifted the phone to a new grip.
Mom, what are you counting.
How do you know I'm counting something.
Last time you slipped and couldn't pick the sentence back up. Zhou Yao said: every time you say that line, you're counting something.
She smiled at her end of the screen: counting stars.
How many have you counted.
One thousand forty-two.
Last week you said eight hundred-something.
Eight hundred seventy-one. The day before yesterday it gained a hundred and seventy-one.
And tomorrow.
Tomorrow won't necessarily be more.
Zhou Yao said: could you miss some.
I could.
Then what are you counting them for.
Zhou Yan thought a moment. The thinking was visible to Zhou Yao; twelve-year-olds see everything.
Because the part that gets missed—I have to know where it was missed.
Zhou Yao said: oh.
Then she asked: who's the man in the eighth one.
Which one.
The reflective-vest one, sitting on the steps eating.
You know him?
Don't know him.
You don't know him and you drew him.
He's there every day. Zhou Yao said.
She wanted to ask more. The upper right of the call window jumped; the ex-husband was calling Zhou Yao to go wash. Zhou Yao said I've got to go wash, and hung up.
Call record: 9:12.
At eleven-fifty at night, the last verification pass finished.
Six survey archives, one million two hundred eighty thousand light curves, ninety-six thousand four hundred twelve with coverage above seventy percent, six thousand three hundred eighteen with noise low enough, one thousand forty-two passing every test.
The largest separation of a pair: eight thousand one hundred forty light-years, both stars with parallaxes, errors six percent and nine percent. The farthest eleven thousand light-years, the nearest eighty-six. Brightest magnitude 6.2, faintest 11.8.
She stacked the one thousand forty-two curves onto a single plot, time on the horizontal, relative flux on the vertical. They weren't in the same fields, the telescopes weren't the same, the detectors weren't from the same batch, the archives weren't built in the same year. Stacked, they lay like the same retina swept once by the same light.
She rested the cursor on the file list. Filename 2459516_sync.txt. Lines: 1,042.
She sent Las Campanas a mail, body four digits: 1042.
Sent, she closed the laptop. In the tank Lao Wu lay under the driftwood, not moving. Water temperature twenty.