Brant flew in from Chicago and met Marcus at a hotel bar near the airport on a Thursday afternoon, which Marcus had suggested and Brant had agreed to without modification. Brant was a large man who had let himself go slightly to the comfortable side of sixty, with the kind of unhurried authority that didn't require a particular setting to operate. He looked at Marcus for three seconds when they shook hands — the specific assessment of someone who had been meeting people his entire career and had developed precise and reliable opinions about them.
"Voss says you built the beneficial ownership tool in under a year," Brant said, when they were seated.
"The first version was faster than that," Marcus said. "The version with the semantic layer for multi-jurisdictional chains took longer."
"What was the core insight?" Brant said. He had not ordered yet. He was not consulting a menu. The question was the reason he was here.
"Officer continuity as a signal," Marcus said. "Not the entities — the entities change. The people responsible for them have career trajectories that are consistent across entities precisely because the purpose of the entity structure is to be inconsistent. The inconsistency of the structure and the consistency of the people who manage it are inversely related. That inverse relationship is the signal."
Brant looked at him for a moment. "That's exactly right," he said. "That's exactly what I found in 2008, and when I brought it to the compliance leadership it took three years for them to accept it as a valid investigative methodology." He paused. "By which time the actors we were tracking had adjusted their officer rotation cadence and I had to rebuild the baseline model."
Marcus thought about this — the specific frustration of building something right and having the institutional context that should have used it slow it down until the adversarial actors adapted. "How much of your data is still actionable?" he said.
"The officer continuity records are permanently actionable," Brant said. "Career histories don't change retroactively. I have twelve years of documented officer associations across approximately nine hundred entities in the secondary-tier network — the people who provide services to the principals without being principals themselves. Accounting, legal, registered agent services, nominee directorship services." He paused. "Those nine hundred entities are the infrastructure that makes the principal network function. The principals get prosecuted. The infrastructure reconstitutes itself under different entities within eighteen months and supports the next network."
"Unless the infrastructure people are mapped," Marcus said.
"Unless the infrastructure people are mapped," Brant agreed. "And charged separately where chargeable. And publicly documented where not chargeable. Public documentation alone changes the economics of the service relationship — a registered agent in Malta who appears in a public legal record as a documented service provider to a prosecuted network has a harder time finding the next client. Not impossible. But harder."
Marcus looked at him. "What format is the data in?"
"Old format," Brant said, with the specific expression of someone about to describe something embarrassing. "I built the data structure myself in 2007 on a relational database schema that was reasonably sophisticated for the time. The entity records are complete. The officer association records have gaps in the 2014-2016 period when the institutional conflict was at its peak and I was operating with constrained resources." He paused. "It needs to be cleaned, restructured, and validated before it would be useful as an input to anything like your semantic layer."
"How long is the cleaning and restructuring work?"
Brant looked at him. "At the rate the two of us are describing it, I have no idea. That's your domain."
Marcus thought for a moment. He was doing a real-time estimation of the data transformation problem: nine hundred entities, twelve years, a 2007-era relational schema with documented gaps, needing to be mapped to the semantic layer's entity model. He ran the problem through the specific part of his capability that handled data architecture and transformation.
"Four weeks," he said. "If I have access to the schema documentation and a sample of the records to work from, I can build the transformation pipeline in the first week. The validation and cleaning runs in parallel over the remaining three."
Brant looked at him for a moment with an expression that contained, briefly, something that was not quite disbelief and not quite admiration but shared properties with both. "You're telling me four weeks for a project that would take my old team four months."
"Possibly three weeks," Marcus said. "Depending on the gap density in the 2014-2016 records."
"I was told you were good," Brant said. "I was told that with some precision by a person who chooses her words carefully." He picked up his menu for the first time. "All right. What do you need from me to start?"
They worked for two hours at the hotel bar, on the specifics: schema documentation access, sample data export format, confidentiality framework, the question of whether the data transformation work and the resulting combined dataset would remain with Marcus or be shared with the DOJ process, and the relationship between Brant's proprietary data and his own legal obligations as a former bank employee still bound by confidentiality agreements.
The confidentiality question was the most complex, and Marcus worked through it with Marsh on a three-way call the following morning. The conclusion was workable: Brant's data could be used for methodological development and academic research purposes under a framework that separated the data from specific legal proceedings — the framework created a research context that protected Brant's confidentiality obligations while allowing the combined methodology to be described publicly in a way that would have practical downstream impact.
It was not a perfect solution. It was the kind of solution that was possible in the real world rather than in a clean model, where interests that were not perfectly aligned could still be made to move in the same general direction.
Marcus thought that was usually the best available answer, and he had learned to recognize the best available answer without spending too much time mourning the perfect one.
He began the schema analysis that evening. By the following Thursday, the transformation pipeline was running on the first batch of Brant's data.
The System updated, quiet and satisfied:
---
**Architecture Authority Lv. 9 → Lv. 10** *(legacy schema transformation and integration with active semantic layer; multi-source data architecture operational)*
**Sixth Gate: 31% complete.**
---
He noted the level increase and went back to work. There were still eight hundred and thirty-seven entities to process.