Forlais Research
Hermes
We set out to find whether accuracy, reuse, efficiency and refusal could advance together rather than be traded off. In the assessed work, they could.
The question
Greater capability through accuracy and restraint
The question was whether intelligence could become more accurate and more efficient at the same time, while declining the cases it could not properly support and keeping only what stayed useful. Most of our instincts said something would have to give.
Research journey
How the work unfolded
Hermes delivered exact or accurate outcomes across varied work, again and again, while declining cases that were not adequately supported. We came to treat that refusal as part of dependable performance rather than as a gap in it.
Selected advances reduced the work needed to reach a supported outcome, and did so without giving up capabilities the evidence had already established. Keeping both was the whole point.
Earlier discoveries let later related work begin from less information. We only claimed that where later evidence confirmed the benefit still applied, because a shortcut that quietly stops working is worse than no shortcut.
We removed retained material we no longer needed and watched to see what broke. Assessed capability, verification and refusal all held.
Turning point
What changed the direction
The turn was realising that efficiency and reuse are selective, never automatic. Drawing that line protected the accuracy, and it made the limits of reuse part of the finding rather than a footnote to it.
Achievement
What Forlais established
Hermes established that exact reach, selective reuse, greater efficiency, refusal and narrower custody can advance together across assessed families of work.
Significance
Why it matters for AI
Hermes is our evidence that capability and resource discipline are not opposing goals, and that measured restraint can be a strength rather than a concession.
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.