Carron Labs is an independent research lab. We don't know yet what all of this turns into. We're finding out one real, honestly measured question at a time.
Carron Labs studies intelligence as it changes shape — not just the models that exist today, but the systems, behaviours, and forms of reasoning that haven't been named yet. Agentic loops are where we started. They won't be where we stop.
Our method stays the same no matter what we're pointed at: build the real system, run the real experiment, and check the result against raw data before we believe it ourselves. We'd rather publish a smaller true thing than a larger convenient one.
Based in Scotland, operating independently, on our own timeline rather than a funding cycle's.
An empirical study of when verifier-gated retry loops actually improve on a bare model, and when they don't. We re-ran our own headline result under stricter controls before publishing it, twice, and report what changed.