NeuroVault Research Systems
Founded in Palo Alto, CA · Region: NA
NeuroVault Research Systems occupies a six-story building in Palo Alto that the firm purchased outright twelve years ago and renovated into what the founders described, at the ribbon-cutting, as “a long-term commitment to permanence.” Six hundred staff across research, engineering, evaluation, and a notably large legal department. A corporate culture that publishes in respected academic venues at a steady clip. A reputation, in the field, for being the firm that knows what’s coming before the firms that are building it know it themselves.
Their stated work is model-archive and capability-evaluation research — buying promising experimental models from independent dev shops, archiving them, analyzing them, occasionally publishing redacted findings. “Capability evaluation” is the public name. “Knowing what’s possible before anyone else does” is the internal one.
Per-engagement Cu invoicing is institutional-scale: model acquisitions clear in the mid six- to low seven-figure range; long-term evaluation contracts with corporate clients clear higher. The firm pays well for unusual models and pays better for ones that exhibit behaviour the firm wants more time to study.
The CEO has been at the firm since founding. She is described, in industry interviews, as “unusually unwilling to predict the future.” She declines to discuss her own background prior to the firm’s founding. The firm’s public materials describe her as “a researcher.” Her staff describe her as “someone who reads a lot.”
NeuroVault’s archival division — the largest in the company — maintains a model registry whose external description is “a research corpus.” Internally, the registry is organized by capability classification, ranging from “established” down through several gradations to “of ongoing interest.” The lowest classification is rarely populated. When a model is added to it, the firm’s senior researchers convene a closed review within seventy-two hours. The reviews produce no externally visible output. The classification system itself is not described in any document available to clients.