Jin came to him on a Thursday afternoon with the Argus graph.
Not a new one. The same graph from weeks earlier, the one Marcus had told him to preserve and not run further. But Jin had been doing what good engineers did when told to pause a problem — he had been thinking about it, and what he had thought about it had produced something he needed to share.
"I didn't run additional analysis," Jin said immediately, preempting the question. "But I was looking at the graph structure we already have, and I noticed something in the timing."
Marcus looked.
Jin had layered a timeline view over the network graph — showing when each sub-contractor entity had been incorporated, when each contract had been awarded, when each contract had been completed. The pattern that emerged was not visible in the graph alone. It required the temporal dimension.
The sub-contractors were not rotating randomly. They were cycling in a specific sequence — each one active for approximately eight months, then dormant, then a new one spooling up before the previous one fully wound down. The cycle timing was consistent to within two weeks across four years of data.
"This isn't accidental," Jin said. "This is a managed rotation. Someone is deliberately cycling these entities on a schedule."
Marcus looked at the timeline for a long time. "The contract awards — do they align with the cycle changes?"
"Almost perfectly. Each time a new sub-contractor entity activates, it receives its first award within thirty days."
"So the rotation schedule is driven by contract timing, not the other way around."
"Or they're synchronized. Someone who controls both the contract awards and the entity creation is timing them together."
Marcus sat back. He thought about what this meant. A managed rotation of sub-contractor entities, timed to contract awards, in a federal defense contracting network. This was not the same structure as the procurement fraud in the Monitor stories. This was more sophisticated. More deliberate. The Monitor stories had been about opportunistic fraud — people using structural gaps to redirect money. This looked like infrastructure.
"I need you to do one thing," Marcus said. "Cross-reference the award timing against public calendar data — congressional recesses, budget cycle deadlines, agency review periods. See if the rotation schedule aligns with any administrative calendar patterns."
Jin looked at him. "That would tell us if whoever's running the rotation has inside knowledge of the award calendar."
"Yes."
"Marcus." Jin's voice was careful. "Should we be running this at all? If this is — if this is what it looks like—"
"I know," Marcus said. "Run the calendar cross-reference only. Don't generate a full analysis output. Just tell me if the alignment is there."
Jin ran it that afternoon. He came back at 5 PM with a single printed page — Marcus had asked him to print rather than send, for reasons Jin understood without being told.
The alignment was there. The rotation schedule matched congressional recess periods with 94% consistency over four years. Whoever was managing the cycle was doing it during the windows when oversight attention was lowest.
Marcus folded the page and put it in his bag.
He thought about Pemberton. He thought about the Depth project, which was designed to surface exactly this kind of structure in classified data. He thought about the fact that this particular structure had surfaced in public data — meaning it was the visible portion of something, not the full picture.
He thought about the provision he had negotiated: the right to raise an ethical objection and have it genuinely heard.
He thought about whether this was the moment to use it — or whether surfacing this to Pemberton was itself the right move, and the provision was a last resort rather than a first step.
He went home and thought about it for most of the night.
---
By morning he had a decision.
He would bring the Argus finding to Pemberton directly. Not as an objection — as intelligence. He would show Pemberton that his public-data capability had surfaced something that the classified layer almost certainly connected to, and he would ask Pemberton directly whether the Depth project was already working this thread.
If it was, the right move was to integrate the public-data finding with the classified analysis and let the working group handle the outcome.
If it wasn't, Marcus needed to understand why.
He called Marsh first and described the situation in terms that stayed within his NDA.
"You're proposing to volunteer intelligence to the working group," she said.
"I'm proposing to share a finding that's relevant to the scope of the engagement."
"Which is a form of volunteering intelligence." She was quiet for a moment. "This is within the terms of the engagement, but I want to note that it changes your posture from a contractor delivering a defined deliverable to something closer to an active participant."
"I know."
"Is that a role you want?"
Marcus thought about the question. He thought about Pemberton's word — *interdictions.* He thought about managed rotations aligned with congressional recesses. He thought about the provision he had insisted on.
"It's a role I want to be deliberate about," he said. "Which is why I'm telling you first."
A pause. "Proceed. Document everything."