Warren called him in December.
Not through Cho, not through the facility scheduling system. His personal mobile again, direct, which Marcus had come to understand meant the conversation was outside the normal channel for a reason.
"The eleven entities," Warren said. "I want to give you an update."
Marcus was at his apartment, late evening, with the Threadline roadmap on one screen and the Depth project's current architecture on the other. He set both aside and gave the call his full attention.
"Three of the eight new entities have been identified," Warren said. "Two individuals and one corporate vehicle. The corporate vehicle is registered in a jurisdiction we have treaty access to, which creates a legal pathway. The two individuals are more complex — one is a national of a non-treaty country, one has dual citizenship and is currently in the United States."
"What's the status of the three entities you already knew?"
"One is in a formal legal process. Two are under active surveillance with a view toward a future process." A pause. "The remaining five new entities are still being characterized. Your semantic layer is generating leads faster than the analytical team can follow them."
Marcus thought about this. "You're resource-constrained."
"We're always resource-constrained," Warren said, with the flat pragmatism of a person who had been operating inside institutional limits for decades.
"What do you need?"
"I need your semantic layer to prioritize. Right now it surfaces everything above a 0.85 confidence threshold without ranking the results by operational relevance. Our analysts spend time on high-confidence findings that are investigatively significant and equally high-confidence findings that are, in our current operational context, lower priority."
"You want relevance-weighted output."
"I want the system to understand what we're trying to accomplish and surface findings in the order that serves that objective."
Marcus thought about the architecture. Adding relevance weighting was not technically difficult — it was a scoring modification to the existing query pipeline. The hard part was defining what "operationally relevant" meant in a way the system could apply consistently.
"I need to understand your operational priority framework," Marcus said. "Not at a classified level — at a structural level. What criteria distinguish a high-priority finding from a lower-priority one?"
"Proximity to the core principal network. Evidence of active operation rather than historical activity. Potential for the finding to unlock a blocked thread."
"Those are three distinct criteria. I can build a scoring function that weights for each of them, but I'll need your analysts to calibrate the weights against a sample set of historical findings."
"How long?"
"Two weeks for the scoring function. One week for calibration with your team."
"Three weeks," Warren said. "Can you deliver in three weeks given your current load?"
Marcus looked at the two screens he had set aside. The Threadline roadmap. The Depth architecture.
"Yes," he said.
"Good." Warren paused. "Marcus. The domestic individual — the one with dual citizenship who's currently in the country. Our legal team has made a recommendation. The recommendation is a controlled interaction rather than surveillance — engaging him through an intermediary to test his response and gather further intelligence."
"What kind of intermediary?"
"A business one. Someone in a sector where his interests and theirs overlap."
Marcus was quiet for a moment. "You're describing a sting."
"A controlled introduction. The threshold between the two depends on intent and execution."
"What does this have to do with me?"
A pause. "Nothing directly. I'm informing you because the outcome of the controlled introduction will generate data that your semantic layer will need to process. I want you to understand the provenance before the data arrives."
"I appreciate that," Marcus said carefully.
"And," Warren said, "because the sector where his interests overlap with the intermediary's is the civic data space."
Marcus went very still.
"How close to my sector specifically," he said.
"Your company will not be the intermediary," Warren said. "But the introduced context will be a company with a similar profile. A procurement anomaly detection startup."
"You're using Threadline's public profile as a template."
"We're using the existence of companies like Threadline as context. Your work created a plausible landscape."
Marcus thought about what Warren had told him at the Phase Two briefing. *Your legitimate business is an asset. Keep building it.* He had understood that comment as general. He was now understanding that it had been more specific.
"I need you to confirm that Threadline and my team are not at risk from this operation," Marcus said.
"Confirmed. The controlled introduction has no operational connection to Threadline."
"I need that in writing."
A pause. "I'll have Marsh's office receive a document by end of week."
"Thank you."
The call ended.
Marcus sat in his apartment with both screens dark, thinking about the way things connected. The semantic layer, the eleven entities, the controlled introduction, the civic data landscape that his work had helped create. He was not inside the operation. He was adjacent to it in ways that were precise and, apparently, useful to people who moved in rooms he was still learning to see.
The Fourth Gate glowed faintly at the edge of his vision.
*The architecture of trust is harder than any system you have made.*
He thought he was beginning to understand what the Gate was really asking for.