Research
The open problems behind Seamore.
Six questions we are working on, what we have so far, and what is still unsolved. We have no measured results to publish yet.
How little machine learning does a good forecast need?
We are building a forecasting engine that is about 80% statistical models, with a thin layer of small experts and routers. It is still in development. This is the reasoning behind it, and what we have not worked out yet.
Read →
How early can a disruption be called, and how often can it be wrong?
Seamore ranks sea, air and road routes on real-time data and watches them for disruptions. An early warning only helps if a dispatcher still believes it after the last false alarm.
Can a person correct a model’s assumption and trust the new answer?
Scenario Builder recalculates by formula when you edit an atom, so the same inputs give the same result. Each metric is still calculated on its own, and the range around a projection is not a calibrated probability.
Can software work out what a column means before it cleans it?
Source fields are mapped to business meaning, and that meaning shapes the cleaning. Mapping confidence is a workflow signal today, not a calibrated measure of accuracy.
Which rule applies to this decision, in this country, on this date?
Seamore connects to regulators such as ANAF, ASF and the SEC. Connecting a rule’s jurisdiction, effective date and exceptions to a specific shipment, supplier or invoice is the hard part.
How should an agent earn more authority?
AI Studio agents work within set scope and permissions and keep a record of what they did. What counts as a track record that justifies more authority is still open.