The office was a single room.
Marcus had found it through a sublet listing — a twelve-hundred-square-foot space on the fourth floor of a mixed-use building in a neighborhood that was somewhere between industrial and half-gentrified, the kind of block that had a pour-over coffee shop next to a sheet metal supplier. The building had exposed ductwork and concrete floors and adequate internet infrastructure, which was the full list of his requirements. He signed a six-month lease and moved in a folding table, four monitors, two ergonomic chairs, and a whiteboard.
Then he hired Jin Park.
Jin was twenty-six and had been working backend infrastructure at a payments startup that had just done a down round and was quietly letting people go. Marcus found him through a referral from Dara Huang, who had described him as "annoyingly good at distributed systems for someone who looks like he's never slept." This turned out to be accurate on all counts. Jin had dark circles, excellent posture, and an instinct for data architecture that Marcus recognized as genuinely first-rate.
The technical interview was forty minutes. Marcus gave him a design problem — a multi-source data ingestion pipeline with specific latency and consistency requirements — and watched how he thought. Jin went to the whiteboard without being asked. He drew a diagram. He revised the diagram twice. He acknowledged one trade-off that Marcus had intentionally left ambiguous in the problem statement and asked a clarifying question about it rather than assuming. Then he presented three possible architectures and recommended one with specific reasons.
"When can you start?" Marcus said.
"Monday," Jin said.
"Monday works."
He hired the data engineer a week later — a woman named Priya Mehta who had a background in investigative data journalism before moving to tech. She understood both ends of the product, which was rare, and she had opinions about the UI that were almost uniformly correct.
Three people. Five hundred thousand dollars. A working prototype.
Marcus spent the first two weeks doing something that did not come naturally to him: managing.
He was good at it, he discovered, in the way that he was good at most things he turned his full attention to. The System helped in ways he hadn't entirely anticipated. The *Exploit Intuition* domain, which he had initially assumed was purely technical, had a human dimension that continued to develop. He understood very quickly how Jin and Priya worked best — what they needed in terms of clarity versus autonomy, when to give direction and when to get out of the way. He understood where the friction points in their communication would be before they materialized and took small pre-emptive steps. It was not manipulation. It was attentiveness at a resolution that most managers did not operate at.
Jin told him, after three weeks, that it was the best team he'd ever been part of. Priya said something similar in different words. Marcus accepted these comments with the specific form of understated gratitude that was the only version he was capable of, and moved on to the next problem.
The next problem was customers.
---
Sera's two introductions led to one paid pilot: a nonprofit investigative newsroom called the State Financial Monitor, based in Chicago, which produced long-form accountability journalism about state government. They were exactly the right kind of first customer — technically unsophisticated enough to stress-test the product's UX, professionally sophisticated enough to immediately recognize useful findings, and motivated enough to publish things that would become case studies.
Their editor-in-chief was a woman named Carla Reyes who had been doing accountability journalism for twenty years and had approximately zero patience for technology that didn't work. She was Marcus's favorite kind of customer.
Their first session with Threadline ran four hours. Carla and two of her reporters sat in front of the tool with access to their state's full procurement database, and Marcus watched from a laptop on the other side of a video link while Priya sat with them in the room to provide support.
Within ninety minutes they had found three anomaly clusters they hadn't previously identified.
The first two were relatively minor — irregular award timing in a county IT contract that suggested favoritism but probably wasn't worth a story on its own.
The third was not minor.
It was a $23 million cluster in a state infrastructure maintenance contract, awarded in four separate pieces to nominally independent vendors over three years, with officer overlaps that connected back to a construction firm that had donated to the relevant oversight committee chairs in the cycles before each award.
Carla Reyes looked at the network visualization for forty-five seconds.
Then she looked at the camera — at Marcus, on the other side of the video link — and said, "This is real."
"The confidence scores are above 0.8 on the core connections," Marcus said. "The peripheral links are 0.6 to 0.7. Worth independent verification before publishing."
"Obviously." But she was already leaning toward her reporter. "Get me the secretary of state filings for these four entities."
Marcus closed the laptop.
He sat in the empty office — Jin and Priya were both off-site — and looked at the whiteboard where he'd drawn the original Threadline architecture in red marker two months ago.
In his peripheral vision, the System showed a brief update:
---
**Simulation Depth Lv. 3** *(consecutive accurate outcome modeling, confirmed)*
**Second Gate: 34% complete.**
---
Thirty-four percent.
He thought about what the Second Gate said. *Build what others cannot follow.*
The tool was good. The tool was better than anything else in its space. But it was still, fundamentally, a tool that sat on top of publicly available data and processed it faster and more cleverly than a human team could.
*Cannot follow* implied something different. Something further.
He pulled up a blank document and started writing.