She replied at 6:12 AM.
*I've been staring at this for forty minutes. Can we talk today?*
They talked at noon, over video. Sera Okafor was in her mid-thirties, sharp-featured, with the kind of focused attention that Marcus associated with people who had trained themselves to listen before responding. She had the architecture diagram open on a second screen during the call. He could tell because her eyes moved slightly to the right at regular intervals.
"Walk me through the normalization layer," she said, ten minutes in, skipping over the product pitch entirely.
He walked her through it.
"And the network graph — how are you handling the entity disambiguation? Because that's always where these things fall apart. Two companies with similar names in different states, different registration numbers, same beneficial owner—"
"Three-stage disambiguation," Marcus said. "Name similarity scoring with Jaro-Winkler, cross-referenced against officer name matching, then a secondary pass using filing date clustering. If there's ambiguity past stage three, it surfaces with a confidence score rather than resolving silently."
Sera looked at him for a moment. "You built this yesterday."
"Starting from components I'd been developing for about three months, yes."
"The core disambiguation alone — I know a team of four that spent six weeks building something worse than this."
Marcus said nothing. He was watching her process it.
"Who else have you shown this to?"
"You're first."
"Why me?"
"Because you're the only person I know in this space who would understand both the technical architecture and the market application, and who wouldn't immediately try to introduce me to their brother-in-law's incubator."
The corner of her mouth moved. "Fair."
She leaned forward. "Marcus. I've been funding and advising civic tech for two years. I've seen maybe six products in that time that I thought could genuinely reach escape velocity. The data journalism market is real but it's small. The compliance market — the part where government contractors need competitive intelligence and anomaly detection — that market is not small."
"I estimated it at around four hundred million addressable in North America alone."
"Your estimate is conservative." She said it without any performance, just flatly. "If you extend to anti-corruption compliance, FCPA obligations for multinationals doing international procurement, ESG due diligence requirements—"
"I've done that extension. It gets complicated because the data sources become significantly less standardized internationally, but the architecture is designed to be source-agnostic."
She stopped talking. Looked at the architecture diagram again.
"I want to make an introduction," she said. "A specific one. There's a seed-stage firm — Lattice Ventures, you may have heard of them — they've been looking for a civic tech investment for two cycles. The managing partner owes me a conversation."
"What would that conversation look like?"
"An ask in the range of five hundred thousand for ten to fifteen percent. Enough for you to build a real team and get to a demonstration version."
Marcus thought about this. He thought about his contract at Helix, which paid him $145,000 a year, which was good money that was not thirty-one thousand dollars by September. He thought about the architecture diagram and the eight days it would take him — him specifically, the current version of him — to build a demonstration.
"What's your interest?" he asked.
"Advisory equity. Five percent if they invest."
"Standard."
"And," Sera said, "I genuinely want to see this exist. I've watched bad actors cycle public money through procurement structures for twenty years and none of the detection tools are anywhere close to good enough." She paused. "This is good enough. Possibly better than good enough."
"Give me two weeks before you make the introduction," Marcus said.
"Why two weeks?"
"Because in two weeks I'll have a working demonstration rather than an architecture diagram, and your introduction will go better."
She studied him. "You can build a working demo in two weeks?"
"Closer to ten days," he said. "I want some buffer for edge cases."
Another pause. He could see her calibrating something — not whether to trust him, but how much.
"Two weeks," she said. "Don't disappear."
"I won't."
---
He went back to Helix the next morning. He sat in his ergonomic chair under the buzzing light and did his contracted work, which took him approximately 40% of the cognitive capacity he now had available, and spent the rest of the day in a state of very disciplined patience.
The System updated twice during the workday.
First, mid-morning:
---
**PASSIVE OBSERVATION:**
Suboptimal environment detected. Current task complexity: insufficient. Inference Engine degradation risk: low for now. Recommendation: manage exposure time or supplement with higher-complexity parallel tasks.
---
Marcus thought: *yes, I know.* He opened a discreet terminal window and started outlining Threadline's API specification.
Second update, just before his afternoon meeting:
---
**EXPLOIT INTUITION — Lv. 2 → Lv. 3** *(threshold reached via passive system observation)*
New capability note: vulnerability surfaces now detectable in human organizational behavior, procedural structures, and institutional decision-making, in addition to technical systems.
---
Marcus read this one more carefully.
*Vulnerability surfaces in human organizational behavior.*
He sat with that for a moment. Then he went into his meeting and listened to Gordon Fitch talk about velocity, and for the first time he found the meeting quietly interesting — not for its content, but because he was now reading it as a system. The power dynamics, the procedural gaps, the places where the org chart did not match the actual flow of decisions. The vulnerabilities.
He wasn't sure yet what he would do with that particular ability.
But he filed it away.