Blog
Notes on the research as the experiments progress: what we are testing, what the runs show, and why it matters. Designs and results live on the experiment cards in the registry and blog posts discuss them. Claims in posts are labeled: REGISTERED when they trace to a registered run, EXPLORATORY otherwise.
August 14, 2026
Why KamiBench for continual learning
What a real test of agent learning requires — persistence, rules too rich to memorize, other adaptive actors, and a score the agent cannot grade itself on — and how a persistent on-chain world provides all four by construction.