Timeline
What we've shipped.
Research, hackathons, releases and posts, newest first. Planned work has an open circle.
-
Planned
Does the referee work on a real pendulum?
Real recordings with faults we introduce ourselves, and a referee that waits a moment after the alarm before naming the cause.
-
Tellurion Referee at the Open Scientific Intelligence Hackathon
In two days: a simulated pendulum with an answer key, a learned world model, and a referee that names why the model and the measurements disagree. Tested once on a locked set of 1,000 runs: the model's own confidence missed 25.9% of real problems; the referee missed 3.6%.