The difference map had hung on the lightbox six months. Nobody touched the three pencil circles; no new numbers were added inside them.
That line in the engineering weekly read as always: LP 413-52 telemetry cycle normal. Normal meant the companion relay, eleven point three light-years out, sent a packet every ninety days; every packet spent eleven years and four months on the road; Earth had no right to know early what any packet held. There was nothing to do but wait. In engineering this waiting has a term of art: the no-operation window. This one ran five hundred forty-five days. The night Mateo changed the recheck cycle to ninety days, his account went gray—gray for a year and a half; nobody in the group @-ed him. Ursula's line—measure the third time first—stayed nailed into the minutes, and it nailed shut everyone who wanted to reason. Reasoning takes numbers. The numbers were on the road.
One more packet on the road.
The media wouldn't wait.
The second packet's data had been released in full, checksum frames and all, on the public server. Nine days after release the first long piece came out, from the desk that had once written "the universe's reply to us needs no translation." This one's title was shorter: we write; it erases. The piece set the two profiles side by side with the difference map, and the difference map was one color only; in the article that color was described as "the vanishing half-meter." Half a meter was the wrong number—it was 0.5 meters; no difference in precision, a great difference in the reading. No one corrected it. The first comment under a correction would have been: call it zero point five or half a meter, whatever you like—it's the one that gets to say what this is called.
The Sleepkeepers picked the sentence up. They had been quiet two years; the S table's flatline left them nothing to say; the scratch gave them something. Their Friday assembly started again, broadcast, and that steady-voiced man no longer spoke of pretending sleep—he spoke of mechanics. Gentlemen: a thing dead three hundred million years has wiped away, in fourteen months, the half-meter humanity carved. It does not need to hate us. It does not even need to know us. Our highest art, on its body, is a scratch that heals. He paused, and gave the conclusion. He said: this isn't a failure. It's a reply. People listening for a reply ought to stop.
The passage was cut to forty-seven seconds; plays broke sixty million.
The poll was the framework information office's, quarterly. Support for the word "bone" fell from twenty-nine percent to twenty-two. The top opposition reason changed: what had ranked first was too expensive; now first was it won't keep. Too expensive dropped to second; third was a sentence new to the options, entered whole: even it is doing the erasing—why do we still carve.
Hu Yisu stood on neither side. His account's last post sat three years back, a link to the alignment module's open repository. A reporter found him and asked about the scratch. On the phone he spoke twenty-seven characters: the scratch is eleven point three light-years out; the rent isn't. What you should be asking is whether the third packet has arrived. Then he hung up. The twenty-six characters were forwarded twenty thousand times; most used them as mockery, a few as a calendar—counting forward by ninety days a packet to the third packet, and marking it: wait.
Kramer's student had been running forward models since the second packet.
The student's name was Brouwer, third-year doctorate; his topic had been cooling models for white-dwarf shells, and when the scratch data came down the topic changed its name, to numerical simulation of carved-mark degradation. He ran two versions. Version one was pure diffusion: thermal motion of lattice atoms filling the groove's high edges into its lows, rate set by temperature. This version computed fourteen months of shallowing at 0.1 meters; the measurement was 0.5; off by five times. He added the strain field at the groove floor and brought out version two, and version two fitted the second packet, error eight percent. He posted the preprint; its conclusion line read: degradation is driven by the strain gradient; the rate will fall as depth shallows; the third packet is predicted shallower by no more than 0.3 meters; if exceeded, the model is void.
Kramer forwarded the preprint into the group with the same half-sentence attached: we'll talk when the model can't keep up with the data.
Zhou Yan spent two evenings on that version of the model. She didn't check the physics—on the physics she trusted them—she checked the structure of that sentence: draw the line of falsification first, then wait for the data to step on it. That is clean work. Her own S table was written the same way, five thousand-odd lines, each line one chance to be falsified, and not one line stepped on. She saved the preprint into 3t, beside the folder named 2471.
In that year and a half she checked the S table once a week. The table was no longer hers to compute; it belonged to the duty sequence, printed one line a day, and she signed after checking. One weekend she turned to the printout's page foot and found the young people of the duty sequence had added three columns of their own: the day's running line count, days of flatline, days to the third packet. The third column came down by one each day; once it fell below two hundred, she would stand a while longer after checking and signing.
The ears were going down. The left ear's four-kilohertz line had dropped two notches in two years; at meetings she sat front row, read lips, and where lips failed, guessed. Old Tang had retired to the interior; on the phone his accent was getting past her, and twice she said type it, and Old Tang typed, and the typing Old Tang said more than the phoning Old Tang. The right ear's hearing aid was never fitted—there was nothing to fit; the left one she wore, its usefulness marked down month by month, and the fitter said the same word each visit: compensation. The last time she asked how much room for compensating was left. The doctor looked at her once and wrote on the chart: lows still serviceable; highs approaching the compensation limit. She said: I know those two characters. Another day, she said. The doctor didn't laugh. This one didn't know her, and so didn't count the two characters with her.
Saturday video. Zhou Yao was twenty-seven; behind her, the print-drying wall of the studio. She said the studio had been quarreling over something: a client wanted a carving cut deep, saying carved deep keeps longer, and the master wouldn't take it. Zhou Yao said: the master says carve a block deep and the wood fibers break inside it; by two hundred prints it cracks. Whether it keeps isn't a matter of how deep you carve; it's a matter of how deep the wood's own grain will take. Zhou Yan read the passage through the captions, paused, and typed one line: ask the master for me—how does he know how deep a piece of wood will take. Zhou Yao said: he says, ruin three, and you know the fourth.
That autumn she went back to the station once. The S table's originals lived in an archive box at the station archive; at handover it had been agreed she would verify them once a year, and she did the verifying in passing. Lines 1 through 3,200-odd in the box were still her handwriting; on the pages furred from turning the archive had laid interleaving tissue. The verification sheet signed, she walked around to the little room behind the machine room. The tank was still there, the water still changed, fine sand laid on the tank floor. The technician didn't know her; seeing her stand at the tank, he said there used to be a crayfish in this one, kept twelve, thirteen years. She said she knew. The technician said Master Eli had left instructions: the water must not stop; if it stops, he'll come back and check. She asked whether he still came back. The technician said once last year; he didn't come in; stood at the door a while, looked at the feed lines on the roof, and left.
The shell was in the specimen box, the box on the shelf's third level, the label in Eli's hand: Lao Wu. Raised at the station. The shell is proof that it grew. She took the box down and opened it one crack; the notch still two millimeters; the shell more brittle than when he was alive; at the edge a small ring of whitening—weathering. She put the box back, photographed it once, didn't store it in 3t—stored it in the phone album. On the ride back she took it out and looked again. Where the road crossed the pass the phone lost signal; she shut the screen, leaned back, and went over the number once more in her head: eleven years and four months. The road can't open a gap.
The day the third packet arrived was Julian date 2465303, 14 October 2037, 3:51 in the morning.
On duty again, Engineer Ma. This time he called people before the checksum finished; when he'd called them he posted nothing to the group but the time. By the time Zhou Yan reached the machine room the difference map was already up, and from where she stood she couldn't see the depth column—the first thing she saw was the lip. The groove was still there, 210 meters long, 6.1 wide; the extra 0.1 of width came in rounded from the two sides. The polyline of the groove lip had rounded like a knob worn long in use—the graduations still there, the edges gone.
The depth column: 2.2 meters.
The second number came out slower; Engineer Ma had done it by hand, and when it was done he read it out: eighteen months, shallower by 0.7. The stretch before was fourteen months, shallower by 0.5—0.036 a month; this stretch, 0.039 a month. He read it out and laid the pen down. No one in the room picked it up. The rate hadn't fallen. The model had it falling by half; it had risen a little.
Brouwer's preprint stayed up one hour forty minutes. He took it down himself, and with the withdrawal attached one line: model void, as agreed. Kramer answered in the group with one word: good.
Mateo's account, gray a year and a half, lit at seven that morning. He said nothing; first he stacked the three profiles into one image and posted it, and then he said: call a video meeting.
Ursula spoke last at the video meeting. She said she had waited for the third measurement; now she could speak. She spoke three points, one sheet of paper to each, held up before the camera, handwritten.
First: the white dwarf's shell is not dead through. Forty-one hundred degrees; in the lattice, carbon and oxygen atoms trade lattice sites; until the temperature falls to absolute zero, the trading doesn't stop. That is the noise floor—uniform; it can't flatten a six-meter groove; it only ages every surface at the same speed.
Second: the driver is the strain gradient. In a groove smashed out by a plume, the floor's lattice is bent; a bent atom sits higher in energy than an atom on flat ground, and the excess pushes atoms toward the groove's flanks. Here she stopped a moment and said: this one isn't my invention; it's in Brouwer's version two. His error was taking the strain for a constant. The shallower the groove, the shorter the slope the same drop in depth spreads over, and the steeper the gradient. So the rate doesn't fall—it rises. He drew his line of falsification correctly: onto himself.
On the third sheet, one number: two hundred. Strain energy decays exponentially with depth; the decay length is computed from this star's own lattice constant and shear modulus—around two hundred meters. Deeper than two hundred meters, the groove's root sits in the range that cannot crawl, and the time to flatten it climbs by the exponent, up past the age of the universe. Shallower than two hundred, the whole groove, top to bottom, soaks in the range that can crawl—depth doesn't matter, only sooner or later. She stacked the three sheets and said the last line: depth is the only independent variable. Everything else is a coefficient.
Someone asked: and between three meters and two hundred. She said: everything in between is called shallow. Another asked: why is it exactly two hundred meters. She said: this star doesn't know two hundred meters. Two hundred meters was fixed the day it was born. People have only just measured it.
Before the meeting broke up, Mateo had the engineering consequences laid out.
The main etch's design depth was 230 meters, plus or minus one hundred. That plus-or-minus-one-hundred had been margin left for the unknown: storage time 10^12 years, error of two orders, riding on that margin. Now the measured rate was in the door, and the margin was gone. He computed it for everyone: extrapolating at the third packet's rate, a 230-meter groove—below the root it can't crawl, but thirty meters above the root soaks in the range, and storage time falls from 10^12 years to 10^9. Ten to the ninth years, he said—short of the age of the universe by an order of magnitude. Long enough; not the number he had once given.
The plan came out with it, fifty-three black-and-white pages as before. Deepen to 320 meters: ninety meters more below the root, and of the stretch that soaks, thirty meters kept as loss allowance. Impactors go from twenty-nine to forty-four: of the seventeen calibration bodies one has already been refitted to the shallow cut, sixteen stand unchanged; the thirteen content bodies stand unchanged; fifteen are new, in two batches. Schedule, four years more. The budget page he turned fastest, and read out one number only: plus forty percent on the approved total.
The budget committee's man was online, and asked where the burn stood now. He said: a third of the supplemental is burned; three bodies in the main belt are being pushed, pushed halfway—stop now and it's all waste. What the burned third bought was one item only: knowing where the two-hundred-meter line sits. Someone said it on the spot, and the sentence later made the headlines: a third of the money, for one number. Mateo said: right. He said: in a concentration plant this happens every day. Drill ten wells; nine come in dry; the account goes on the well that gave water. Nobody laughed.
The hearing was scheduled, the following spring; the topic line read: main-etch design depth and budget amendment. Opinion had tipped before the schedule was even published. The support number was not issued that quarter; the information office's reason: no longer asked as a separate question. On the social platforms the sentence pattern closed its loop—erase then carve, carve then erase; either way, it erases. One image traveled widest: a purple rendering of the groove, and pressed beneath it that line—it is keeping our drafts for us.
Old Tang's message came on the ninth day after the third packet. Typed.
He said that before retiring he had cleared the old man's study once, with the family there; the things went into three classes—donated, kept, pending. Pending held two items. One was the two tape reels, long since given to the archive. The other was a kraft-paper envelope, and on its face, the old man's own hand, three lines. Line one: the day the shallow cut gives a result. Line two: for Zhou Yan. Line three, in smaller characters: do not mail.
Old Tang said that back then he hadn't known what "gives a result" meant; the old man hadn't explained; so he kept it. Kept it six years. Days before, he saw four characters in the news—the scratch is disappearing—and he took the envelope out of the cabinet. He said: I'll send it registered tomorrow. The no-mailing was the old man's fear of losing it; registered doesn't get lost. When it arrives, answer me one character.
The envelope came the third day. She opened it alone in the office. Inside, three pages. The first two were originals from the 1968 calibration logbook, the paper gone yellow, the tables ruled by hand—the groundwork of that system made public long ago. The third page was the statement, one page; that pen, the pressure still heavy, the ink drinking through to the back. The first line read:
The conclusion first: those five places in the fair copy, I altered. The direction of the alterations was to record the noise smaller.
Below were the five original values, one to a line, laid out like a table. The first line was the six o'clock slot of March 9, and at its end one number, 0.7 larger than the fair copy's.