The problem was the category.

Maren had been filing documents at the Mechanist Academy library for twenty-three years. In that time, she had processed 9,814 items — monographs, technical specifications, reference catalogues, experimental summaries, survey reports, and one recipe for foundry-grade lubricant that had been submitted by a junior researcher who’d confused the intake forms and whose embarrassment had been, by institutional consensus, never mentioned again.

She had categories for all of these. The filing system was not complicated — it was, in fact, elegantly simple, the product of her predecessor’s predecessor’s conviction that a library was a machine that ran on alphabetical order and thematic groupings, and that the purpose of a librarian was to keep the machine oiled. Natural Philosophy. Engineering. Theological Reference. Applied Metallurgy. Agricultural Science. Survey and Cartography.

None of these categories contained the document she was holding.

It was a bound monograph. Eighty-seven pages. Title page: The Domain Mechanics: On the Systematic Measurability and Reproducible Application of Domain-Yield Effects. Author: Sera Thornwick, with editorial attribution to the late Meren Thornwick’s research notes and Great Debate transcripts. Publisher: Mechanist Academy Press, Year 463 AF. Filing date: today.

Maren opened it to the methodology section. Read one paragraph — fourteen-month controlled trials, Forge-domain yield increase measured at 23-27%, reproducible across independent testors regardless of devotional status or temple affiliation, statistical variance within acceptable bounds. She turned two pages further. The data tables were dense and clean, formatted in the standard Academy tabular layout. Each trial set was tagged with location, testors, control protocol, and a column for observational notes that was mostly empty, which meant the observations had been unremarkable, which meant the trials had worked, which meant the yield numbers were solid.

She closed it.

The paragraph was boring in the way that only correct things could be boring. The prose was flat — someone who had measured something real and did not need to make the measurement exciting, because the measurement itself was sufficient. Maren had filed enough speculative monographs to know the difference. Speculative authors wrote with energy. They used italics. They made arguments. This author had made measurements and then written them down, and the writing had the specific temperature of certainty.

She queried two colleagues. Pellus (senior archivist, Human, sixty-three, responsible for Natural Philosophy and Theological Reference) said: "It’s natural philosophy. Domain effects are natural phenomena. File it under Natural Philosophy."

Corinne (junior archivist, Kobold, twenty-nine, responsible for Engineering and Applied Metallurgy) said: "It’s applied. They’re measuring yield outputs. That’s engineering methodology applied to domain effects. It’s engineering."

Neither was wrong. Neither was right. The document occupied a space that the filing system had not anticipated because the filing system had been designed for a world in which domain effects were either theological (the Crucible’s position) or philosophical (the Academy’s traditional position), and this document said they were measurable, which was a third thing, and the filing system did not have a third thing.

She considered, briefly, whether the document could simply be filed under Theological Reference and left for the subject editors to sort out. This was not a coward’s option — it was, in fact, what intake protocols specified for documents whose classification was contested. But it was also incorrect. The document was not theological. It did not discuss the god’s will, the Crucible’s doctrine, or the relationship between faith and divine favour. It discussed percentages. Pellus had said "natural phenomena" and he was not wrong, but natural phenomena did not get measured in fourteen-month controlled trials with independent testors and statistical variance analysis. That was scientific methodology. And scientific methodology had a category.

Or it should have.

Maren sat at her desk for four minutes. She was not thinking — she was not the kind of person who sat and thought dramatically about decisions. She was waiting for the obvious answer to arrive, the way obvious answers usually arrived: not as inspiration but as the absence of a better alternative. The clock on the wall behind the intake counter ticked through those four minutes with the indifferent regularity of a mechanism that had been measuring institutional time for longer than Maren had been alive, and she had the habit of not looking at it, because looking at the clock during a filing decision implied the decision was time-sensitive, and filing decisions were never time-sensitive. They were accuracy-sensitive. Time was irrelevant.

She opened a new category card. Wrote: Applied Domain Science.

Stamped it. Filed the monograph.

She pulled the next submission from the intake queue — a pre-review manuscript, anonymous, something about resonance patterns in faith-reactive mineral deposits. It had been submitted through the standard academic channel three weeks ago. The formatting was acceptable. The methodology section was present but abstract, written in a register that suggested familiarity with Academy conventions without full institutional membership — someone who’d read enough published work to reproduce the structure, or someone who’d had it reproduced for them. She noted its filing number, assigned it a preliminary category (Applied Metallurgy — pending classification), and set it in the review queue without curiosity, because the queue was long and the morning was short and anonymous pre-submissions were not unusual and this one would be reviewed by the appropriate subject editor within forty-five days, which was standard, which was the machine running.

She did not know that the anonymous manuscript had been submitted through a data-collection methodology that would not be identified as hostile for another seven arcs, or that the methodology’s architect was a god she had never heard of, or that the library’s standard intake process — the machine she maintained with twenty-three years of institutional diligence — was, for the first time in its existence, being used as an intelligence-collection channel by a power that would, eventually, make her filing system the least of her problems.

She did not know any of this.

She filed the anonymous submission and picked up the next item in the queue.

At close of day, Maren locked the library and walked home through the canal district. The streets were wet from the afternoon’s Ironfall drizzle — a persistent, cold rain that didn’t fall hard enough to justify sheltering but didn’t stop long enough to dry the cobblestones. Her boots found the familiar path without requiring her attention: left at the chandler’s, right at the cooper’s sign, straight through the alley where the tailor’s apprentice always hung his practice cuts to air, even in Ironfall, even in the wet, because apprentices learned their trade’s habits before they learned its reasons.

Her husband was reading at the kitchen table — a foundry maintenance manual, third edition, which he’d been reading for two weeks because he was methodical and the manual was six hundred pages and he did not skip sections. The table smelled of the wood oil he’d applied that morning, and the lamp was already lit, which meant he’d been home since fifth bell, which meant the foundry shift had run on schedule, which was information she processed and filed in the same automatic way she processed and filed everything — not because she wanted to, but because twenty-three years of institutional filing had trained her mind to categorise inputs, and inputs included the smell of wood oil and the state of lamplight and the angle of a husband’s reading posture.

She sat down. Pulled off her gloves — the leather ones, lined, Ironfall weight.

"Made a new category today," she said.

He looked up. "What for?"

"Domain science. Someone measured how blessings work."

He considered this the way he evaluated foundry tolerances — slowly, thoroughly, once. "Hasn’t the Crucible been saying they work because the god wants them to?"

"Yes."

"And now someone’s measured them."

"Twenty-three to twenty-seven percent yield increase. Reproducible."

He returned to his manual. "Doesn’t change the soup."

She didn’t answer. She started dinner. He was right, in the way that practically correct things were right — the measurement didn’t change the soup, didn’t change the temple on the hill, didn’t change what she did at sixth bell every morning or what he did at the foundry or how the canal district smelled in Ironfall. Whatever the monograph had measured existed in a different layer of the world from this kitchen. The library would open again at sixth bell tomorrow, and the queue would be longer, and the new category would be filed between Applied Metallurgy and Natural Philosophy in the alphabetical sequence, and the machine would run.