Principal Yusuf Adair, Science Corps (Human)

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.