Clinical input came first.

Before we wrote production code, we did qualitative research and brought clinical expertise into the design. The Assistant, the part of Alto that reads and structures a referral, was developed with that input from the start rather than retrofitted to it.

Tested against clinical judgement with the University of Galway.

We then asked the hardest question: can it match a clinician? An independent concordance study at the University of Galway measured Alto's output against expert clinical judgement and found 92.4% concordance to gold standard. There is a full report behind that figure. That beats the X% reported in the literature, but we knew that the X% of misses were real patients. 

The Judge.

From that body of work came the Judge, the second agent in Alto's architecture and the part that checks the Assistant's reasoning rather than trusting it. Our own inbuilt QA. We have continued to develop and stress-test the Judge in further work with Ulster University.

Post-hoc analysis with Munster University. 

Alto and Munster University carried out a separate post-hoc analysis of the data collected with Galway. Those findings fed directly into a new structured tagging system, sharpening how Alto categorises and reasons about each referral.

Ongoing internal simulation.

The work does not stop at any single milestone. We run our own internal simulations continuously, so the system is tested against new and edge-case referrals long before they would ever appear in a live setting.

This is not one study we point back to. It is a chain of independent and internal evidence, each step building on the last.