The document on Pemberton's desk was four pages.
It had been drafted by someone in the oversight office — career staff, not political, the kind of person who wrote in careful bureaucratic language that was nonetheless precise about what it meant. The subject line was a case number. Below the case number, in the body, was a section titled OPERATOR ASSESSMENT, and below that was Marcus Vane's name.
The assessment was thorough. It covered his educational background, his professional history, the Threadline founding and the Series A, the Monitor stories and their downstream effects. It covered the Northern Virginia engagement in language that was appropriately classified. It covered his performance in that engagement, which was described in the careful vocabulary of an evaluation: *demonstrated capability substantially exceeds initial projection; architectural approach to ontology layer is novel; timeline ahead of schedule.*
Then it covered something else.
There was a section titled ANOMALY, and in that section the document described a pattern that the oversight analyst had identified in the Depth project's development logs. Not a security breach. Not a protocol violation. Something more subtle: the rate of Marcus's architectural output was inconsistent with what was possible for a single developer working the hours he had logged at the facility. The gap between logged hours and documented output was not explainable by efficiency alone. The analyst had not drawn a conclusion. They had raised a question.
Below the question was a recommendation.
The recommendation said: *Suggest expanded evaluation of Operator capabilities before advancing to Phase Two access.*
Pemberton had read the document twice. He had set it on his desk and looked at it for approximately five minutes. Then he had picked up the phone and called Dr. Chen.
"Have you read the oversight assessment?" he said.
"Yes."
"Your read?"
A pause. "The gap is real," she said. "I've reviewed the development logs myself. What he's building — the semantic layer specifically — is not something I could produce in that timeframe. I don't think a team of three could produce it in that timeframe."
"Explanation?"
"I don't have one that fits the standard model." Another pause. "He's either the most capable systems architect I've encountered in twenty-two years in this field, or something else is happening that I don't understand."
"Which do you think it is?"
The pause was longer this time.
"I think," Dr. Chen said carefully, "that those two options may not be mutually exclusive."
Pemberton had looked at the document for another moment. Then he had filed it in a folder he kept in the lower left drawer of his desk, the one with the combination lock. He had not acted on the recommendation yet.
He had decided to watch instead.
---
Marcus did not know about the document.
He knew, with the Institutional Mapping sub-domain now fully active, that he was being evaluated in ways that went beyond his technical output. He could feel the shape of the additional attention — in the slight variation in how Cho greeted him at the facility, in the one instance of a camera angle change in the secure workroom that he had noted without remarking on, in a question Dr. Chen had asked about his personal workflow that was technically reasonable and also not entirely about workflow.
He did not know the specific content of what was being assessed. He knew the assessment was happening.
He had made a decision about it two weeks earlier, sitting in the secure facility at 7 AM with the semantic layer open on his workstation and the System display hovering in his peripheral vision at Architecture Authority Level Six. The decision was: behave the same way whether or not he was being watched, because the alternative — calibrating his output to appear more normal — would require him to build worse things, and he was not going to build worse things.
There was a second part to the decision, which was: if they asked him directly, tell them the truth. Not all of it. The System was not something he was prepared to explain to Pemberton, partly because he had no adequate explanation and partly because a capability he couldn't explain was a capability that could be treated as a threat. But the truth of his capabilities — the fact of what he could do, if not the mechanism of how — that he would not lie about.
He had thought about this with the full weight of his current Simulation Depth and had concluded that honesty, carefully structured, was the dominant strategy. Deception required consistent maintenance and created vulnerability surfaces. Honesty, even partial honesty, was architecturally simpler.
He went to the facility on Monday morning and built the next layer of the semantic engine.