Experiment 001 redone through Satsuma's own governance loop and published

Experiment 001 is now published on this site. In it, Satsuma, an autonomous system that uses language models for its decisions and its own chain for its identity, treasury and rules, gets an objective and has to produce everything else through its own process. The governed run created and approved six proposals over 1h45m. Three were built, reviewed and deployed: a Uniswap v2 exchange on tangerine, a mandate account with a keyless executor, and a second executor that duplicated the first. Each swap traces back to an executed proposal.

That is the second run. The first used a standalone harness: a deploy script set the objective and the mandate, and no proposals were created or executed, so all it showed was something agents can already do. Of that first run, the finding was blunt: "The swaps in the recorded run were not the result of proposals." The experiment was redone through the full loop of objective, proposals, council vote and execution.

The rerun also exposed a gap in the loop: "The loop has no step for using what it builds. Today it ends at deploy." The decision was to add an operating step and to disclose it in the article. In the system, the commit that runs the loop on a real Satsuma chain, across two chains, also adds operation of what it builds.

The article was retitled. It gained a short primer on objectives and key results and lost a paragraph about the limits of sibling chains. It was merged and deployed, and the live site now serves the article, a replay GIF (720px, 76s, 13MB), a 1080p MP4 and an interactive replay.

Still open: a narrated cut (2m18s, 12 clips, with an ambient bed) exists as a private preview and needs a listen-through before it replaces the silent version. The primer's history of objectives and key results has not been checked against sources. The merge happened before CI reported on the last commit, though the deploy build passed.