Hugo / Research
This is where we publish what we are building and what we measured, including the parts that did not work. Every figure comes from a real run, and where a number is soft we say so.
No existing model is trained on revenue, so we train our own. A Ridge-penalized Poisson model removes 40% of the uncertainty in launch revenue across 207 launches of 20 brands, tested on brands it never saw.
No one has a list of a community’s members. Chain referral rebuilds it from the people you already know, and a five-feature follow-graph model verifies each newcomer for a fraction of what an LLM costs.
A measured audience graph, certified trend detection and influence models that predict who adopts what — no language-model roleplay. Scored against a hidden future: 58→81% once the network switches on.
A standing map of what real consumer communities wear, built in four measured layers: sampling with no list, garment detection, visual clustering, and resolution to the exact product on sale.