Forlais Research
Icarus
We wanted capability that accumulates without dropping what it could already do. Icarus grew, kept what it had earned, and refused one tidy conclusion.
The question
Capability that grows without losing earlier success
The idea was accumulation. Could useful capability build up from experience while every earlier success was preserved and unsupported conclusions were still refused? Growth that costs you what you already had is not growth.
Research journey
How the work unfolded
We brought earlier findings together into a dependable capability we could call on. It answered every supported case in its assessed set and rejected every unsupported control we put in front of it.
Later growth widened its supported reach without losing what had come before. Every advance had to keep what was already earned, and that condition did more work than any single advance did.
Different assessed learning histories produced different readiness and different exact outcomes. Within the assessed work, looking only at the final material was not enough to judge whether something was ready.
Separate retained conditions produced different strongest results under the same bounded assessment. We wanted one universal reading and the evidence would not give us one, so we left the unresolved cases unresolved.
Turning point
What changed the direction
The turn was accepting that the same bounded assessment did not support one universal account. We swapped an attractive general conclusion for a sharper finding tied to conditions, which is less satisfying and considerably more true.
Achievement
What Forlais established
Icarus established cumulative exact capability, meaningful sensitivity to history, differences that depend on conditions, and continued honest refusal as its supported reach expanded.
Significance
Why it matters for AI
Icarus is why we judge growth by what is retained as much as by what is added, and why refusal protects the evidence record precisely when capability is expanding fastest.
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.