Survives Correction

Nothing was wrong for ninety days, and Syra logged the nothing each day in the form the protocol gave it, because nothing is a reading when it is measured.

The membrane had been reached the way every new membrane was reached, by fold from a Tier 1 site on the far side, and the party of twelve had come through unharmed and found air that was air and a cold moon with a gradient quiet enough to put a shelter on. Not a Lagrange point of anything; the survey lead had looked for one and the system did not have the geometry, and the quietest ground the party could find was logged as Tier 2 on the first day and every day after. The pilot section had tried for a harmonic leg between the shelter and the second survey point and the drive lattice had found no admissible signature, which was filed as no charted corridor, which is what a new membrane usually has, and the second point was reached by hull instead.

Then the single-system work, which was most of the work. The atomic standard kept the second. The spectrometer found the handbook's lines. The chemistry ran. Syra's own body, which sampled the field and the harmonic texture of a place before any instrument she carried, reported nothing: no thinning, no pressure, the ordinary grain of a place that was not near anything. She noted that and filed none of it, since her body was not an instrument the Safety Corps certified and the Survey's log did not carry what it could not audit. It was still the first thing she had checked. The second thing had been the chronometers, which closed their loops to budget, as they had on every membrane she had ever stood on.

On the ninety-first day she ran the correlation calibration, because it was on the schedule, and because the Survey carried a correlation standard to every new membrane the way it carried a second.

Three sources, independent by construction, each sealed and doing its one thing. Braid them, read the joint distribution, compare it to the certified reference, file the residual. The residual is never zero. It has a budget, and on every membrane she had worked it had sat inside the budget, and the calibration had taken a morning.

This one did not sit inside the budget.

She did what one does. There is a list of corrections anyone can predict for a three-source braid, and she applied them in the order the list gives: source drift, detector dead time, timing offset between channels, thermal, alignment, the shelter's own field. Each one took something out of the residual. None of them changed its shape. After the sixth she re-ran the single-system checks on each source alone, which passed, as they had passed, and braided again, and the residual came back the same shape to the width of the error bars, which is the one thing a fault does not do. A fault wanders. This sat.

She gave it a week. She gave it the week not because she expected the eighth correction to be the one — she had run out of predicted corrections on the third day — but because a week is the interval after which a thing that has not wandered may be said not to have wandered, and she would not say it sooner. On the seventh day she wrote, under the raw statistics, which she filed whole and did not summarise: survives correction.

The survey lead came to the bench that evening. Vint was a careful officer and had been a good one for ninety days, and she had read the filing, and she wanted a word.

"I have a log that says discrepancy seven days running," Vint said. "I need to close that line or carry it. If I carry it, it needs a status. Under investigation is a status."

"The investigation is closed," Syra said. "I closed it. There are no further corrections I can predict, and a reading nobody is investigating is not under investigation."

"Then instrument."

"The instruments pass. Each one, alone, to certification. I have filed that above the braid, where it will be read first."

"A reading that says our arithmetic is wrong here," Vint said, "is not a reading I can put in a log on one bench, Nai Syra. Not on this party's name. It will be read at home by people who were not here."

"That is what the log is for." She said it without weight, since none was needed. "The log says what the bench did. What the bench means is not the Survey's to say, and I have not said it. I have said that the residual survives correction. That is the thing that happened. If you want a word that is smaller than the thing that happened, you will have to write it yourself, and I will not countersign it, because it will not be true."

Vint was silent for a moment, and Syra read the room change and named what had changed in it, in the plain terms, since that was the courtesy.

"You asked me to find the correction. I looked for seven days and there is not one. You are now asking me to find a word. Those are different requests. I have done the first."

The log said discrepancy for the eighth day, because a log is kept by the day and the day had been logged before the conversation. On the ninth it did not say anything about the braid at all, because there was nothing further to say about it on the membrane, and Vint, who was a careful officer, had understood that a log cannot carry an investigation its investigator has closed. The raw statistics went into the filing, and the three words went under them, and nothing went over them.

Syra packed the three sources herself. Each one sealed, labelled, written as cargo. What it did at home would be the question. What it had done here was answered, and she did not hold the question past the point where holding it did work, which was the point at which the hull folded.

He had wanted it to be the instruments. He would have said so to anyone who asked, because it was true and because wanting a thing has never been a correction.

The filing had come across to the Science Corps in the spring, through the Standards Office, with the Safety Corps' audit clipped to it: three sources, back from a fold, taken apart on the floor, each one asked whether the transit had changed it, each one certified unchanged. Under the audit, the Survey's raw three-source statistics from the membrane, filed whole, with no summary and no slogan. Under the statistics, three words in a boundary analyst's hand. Survives correction. Under that, nothing.

Adair's section explained things. That was the division of labour and he had never minded it: the Survey collected, and when the Survey collected a thing that did not fit, it came across the corridor to a floor like his to be explained, and most of what came across was explained by the end of the week. He had assumed, reading the three words, that this would be one of those, and that the explanation would be the one it always was when a residual survived a field list — a correction the field could not have predicted, because the field did not have the floor.

So he rebuilt the braid. The same three sources, back from audit, on the floor that had never been anywhere, under a reference that had been certified in this building for longer than he had worked in it. He braided them and read the joint distribution and it sat inside the budget. It sat where every braid on this floor had ever sat. He ran it again with the field's own corrections stripped out, to see whether one of them had put the residual in, and it sat. He ran it with the field's full list applied, as she had applied it, and it sat.

Which closed the instruments. He wrote that down and felt the closing as a loss, and noted that he had felt it, and went on.

Then the list. He re-derived the predicted corrections for a three-source braid from first principles, without her list in front of him, and arrived at her list, in her order, with one addition that the shelter's field made unnecessary and that he applied anyway, because unnecessary is a prediction and he was done with those. The residual in her filing did not move. It had not moved for her either. He had been reading her filing for three weeks before he noticed that he was no longer looking for the correction and had started looking for the sentence that would let the Science Corps say the correction existed without having found it, and he stopped, because that was not his section's job and never had been.

He went to the record instead. Not the field record. The Corps' own standing statement about its physics, the one every instrument on this floor was certified against: that the base quantum layer, the arithmetic of combination, was identical to the real universe's. He had quoted it in a hundred certifications. He read it now as a reviewer reads a thing, which is for its scope, and found that it had one.

Within any given Concordant. Throughout the local cluster. The record had never once claimed its arithmetic for everywhere. It had claimed it for here, in a sentence that had carried a hedge nobody had ever needed to cash, because nobody had ever stood anywhere the hedge applied to. He went through every statement that leaned on the kernel — the second, the cavern certifications, the corridor admissibility tables — and found the same scope on each, written in by careful people who had not known what they were being careful about. Nothing in the record was contradicted. Something outside the record's stated width had been measured for the first time, and the record had been built, without anyone intending it, to hold exactly that.

That was the finding, and it took him four months to be sure of it, which was three months longer than the Science Corps liked and a month less than it would have taken anyone who had not had the three words to start from. He wrote it the way his section wrote things. The instruments pass. The floor returns local statistics to budget. The field residual survives every correction the floor can derive. The record's kernel claim is scoped to the local archecluster and is not contradicted. Reading, at the width of one membrane. He recommended revisit, with the rule he had learned on his own floor written in as a rule: that nothing read on the membrane be filed as a finding there, that every value come home as cargo, that the instruments be audited on a floor that had never left before any number became a reading — because the one thing a party could not calibrate against on the membrane was the membrane.

He had never met the analyst. He had read her filing more times than he had read anything that year, and the thing he kept returning to was not the statistics, which were what they were, but the three words, and the absence under them. She had filed what had happened and declined to file what it meant, and had left the second half to a floor she had never seen, and the floor had taken four months to arrive at a sentence no longer than hers.

He countersigned under the three words, since that was where a countersignature went. Then he went to find out what else the record had only ever claimed for here.