Forlais Research
Odysseus
We pushed reasoning out of the material it grew up on into unseen real language. It travelled further than expected, and we mapped how far.
The question
Reasoning that travels while its limits remain visible
We wanted to know whether dependable machine reasoning could reach across varied work, unseen real language and real source information, with every claim we made staying inside what had actually been assessed. That last part is the difficult one.
Research journey
How the work unfolded
We established repeatable results first, across varied reasoning, language and information handling work. The weaker routes stayed in the record as limits. It would have been easy to quietly drop them and present a cleaner picture.
Then we left the familiar material behind for unseen real language. Odysseus performed substantially better than a strong comparison that had matched it on everything familiar, which established transfer within the tasks and languages we reviewed.
A separate study on real source information achieved fully correct retrieval across its controlled body of work, and stayed correct under independent checking.
Some of what looked like success stopped holding as conditions changed. We corrected the record and used those limits to decide where more work was justified and where it was not.
Turning point
What changed the direction
The turn was discovering that strong looking results leaned on the evidence available to them. We kept the real success and drew a firmer boundary around how far the claim could travel.
Achievement
What Forlais established
Odysseus produced a substantial body of bounded reasoning, language and information handling capability, including transfer to unseen language material and fully correct retrieval inside a separate controlled study.
Significance
Why it matters for AI
Odysseus is our argument that reasoning research advances through repeatable outcomes, independent checking and visible limits, rather than through broad claims that nobody can bound.
Public scope
Forlais Research is the public research surface for Forlais Group. Related official routes include the company site, EvaEsi overview, Project Eve, and selected research programme pages.