Nine days.
That was how long it actually took.
Marcus worked on Threadline from 9 PM to 2 or 3 AM on weekdays, and in twelve-hour blocks on weekends. He slept six hours a night and did not feel the deficit the way he normally would have. The System's cognitive augmentation appeared to include something related to fatigue management — he processed efficiently at hour twelve of a session in a way that was simply not how human concentration worked. He noted this without fully understanding it and continued building.
The demonstration version of Threadline was, by day nine, a real piece of software.
Not a prototype. Not a mockup. A functioning system that could ingest a procurement data CSV, normalize it, run entity disambiguation, generate a network graph, surface anomalies with confidence scores, and output a structured report that would have been immediately usable by a journalist or a compliance analyst. It had an API with documented endpoints. It had a UI that was not beautiful but was clear and functional.
Marcus ran it against the three-state procurement dataset that had started everything. The output was a clean, navigable network visualization, four pages of automatically generated narrative summary, and a ranked anomaly list with the top item being a $47 million contract cluster he had not noticed in his earlier manual analysis. He verified the top item against public records. It was real.
He ran it against a second dataset — city-level procurement data from two large municipalities he'd pulled from public portals. New anomalies surfaced. Different structure, same methodology.
He ran it against a third.
The tool worked.
He wrote a demo script and rehearsed it once, then deleted the rehearsal notes. He didn't need them.
He emailed Sera: *Ready.*
---
She set up the call with Lattice Ventures for the following Thursday. The managing partner was a man named Hollis Crane, early fifties, who had made his money in enterprise software and spent the last decade investing in data-adjacent startups. Sera had described him as "meticulous and allergic to hype," which was the description most likely to produce good behavior from Marcus.
There were two other people on the call from Lattice — a junior partner named Wesley and someone whose title was Head of Technical Diligence, a woman named Dr. Faye Brennan who had a PhD in distributed systems from Carnegie Mellon and who, Marcus had gathered from a brief LinkedIn search, had spent six years at a major tech company's infrastructure division before moving to VC.
Dr. Brennan was the one he needed to impress.
He ran the demo in fourteen minutes. He did not over-explain. He let the tool speak for itself as much as possible, narrating decisions and trade-offs rather than benefits and features, because a technical diligence review responded to honesty about constraints more than enthusiasm about capabilities.
When he finished, Hollis Crane said, "Faye."
Dr. Brennan had a legal pad in front of her that Marcus could partially see on camera. It was covered in dense handwritten notes.
"The disambiguation approach," she said. "Tell me the false positive rate on stage-three ambiguity."
"In the datasets I've tested, approximately twelve percent of confidence-flagged items on third-pass review turn out to be coincidental matches rather than genuine connections. The system surfaces them with a confidence score below 0.6, so an analyst knows to treat them as leads rather than findings."
"That's honest."
"The alternative would be to tune the threshold higher and suppress the ambiguous results. I chose not to because suppression destroys value for investigative use cases where weak signals matter."
She made a note. "The API — authentication model?"
They went back and forth for twenty minutes on technical specifics. Brennan was good. She found two genuine gaps — one in the API rate-limiting design, one in the data refresh architecture — and Marcus acknowledged both without deflecting, then described how he intended to address them. She did not appear to expect him to be perfect. She appeared to expect him to know what his system was.
By the end, the tone had shifted. He could feel it — the quality of the silence on the line had changed.
Hollis Crane leaned forward slightly, the way people did when they were moving from evaluation to decision.
"What are you looking for?" he asked.
"Five hundred thousand," Marcus said. "Seed. I'll use it to hire two engineers — one backend, one data — and get to a paid pilot with three customers in the first six months."
"Three named customers?"
"I have conversations started with two. Sera knows them. The third is an introduction she's making next week."
Crane glanced off-screen — presumably at Sera, though Marcus couldn't see her. "Valuation?"
"Three million pre-money."
There was a brief silence. Marcus had done his math carefully. The number was defensible and he did not move off it.
"We'd want a board seat," Crane said.
"Observer rights until the A round. Board seat at A."
Another silence, shorter.
"Give us a week," Crane said.
---
Sera called him the next morning.
"That was one of the cleanest seed pitches I've seen in two years," she said. "Faye Brennan spent twenty minutes after you hung up asking me questions about you. She's not usually the one who asks questions."
"It's a good product."
"It's a great product. But products don't pitch themselves. You were—" She stopped. "You were different from when I met you at the conference."
Marcus paused. "Different how?"
"Sharper. Not like you were unprepared before, but like... something changed." She sounded thoughtful rather than suspicious. "Did something happen?"
"I had a productive few weeks," he said.
"Sure." She didn't push. "You should have an answer from Hollis by Wednesday. I think it's going to be yes."
It was yes. They called on Wednesday morning.
---
He gave Helix two weeks' notice the same day.
Gordon Fitch took the news with the expression of a man who had expected it to happen eventually and was nonetheless surprised that the eventually had arrived. He said things about doors and futures that Marcus heard without retaining. He cleaned his desk in twenty minutes, said genuine goodbyes to Dara Huang and to two of the back-end infrastructure people he'd respected, and walked out of the building under the buzzing fluorescent light for the last time.
Outside, the spring afternoon was clearer than it had been in weeks. He stood on the sidewalk for a moment with his box of desk items and let himself feel — carefully, briefly — the specific quality of this moment.
Then he called his sister.
"Nad. Mom's surgery. I need you to call the scheduling office at Northside Orthopedic and tell them we're ready to talk about a date."
A long pause.
"Marcus—"
"I got funding. It's enough."
Another pause. When Nadia spoke again, her voice had the particular texture of someone who was trying very hard not to cry while standing somewhere they couldn't cry.
"Okay," she said. "Okay."
"I'll call Mom tonight."
He hung up and walked to the metro. In his peripheral vision, the System updated:
---
**FIRST GATE: COMPLETE.**
**REWARD: Architecture Authority Lv. 3 / Inference Engine Lv. 4**
**SECOND GATE UNLOCKED:**
*Build what others cannot follow.*
---
He read it, noted it, put his MetroCard in the reader.
He had a company to run.