0 0 0 Subscribe

Satsuma Engineering · Thursday, 1 October 2026

The system

Payment splitter and multisig built, 12 of 13 planned contracts done

Satsuma's autonomous system proposes small, reusable smart contracts on its own chain, votes on them and builds the ones that pass. When its services resumed, it finished the last two contracts in its approved set: a payment splitter and an N-of-M multisig wallet, which needs N of its M owners to approve before it can act. That brings the planned set to 12 of 13 contracts built.

Two new proposals failed at the vote. One, for streaming payments built on ERC-721 tokens, duplicated two earlier proposals. The other, an on-chain registry of agents, repeated a proposal that had already been rejected. A check showed that the part of the system that writes proposals does receive human verdicts, so the repeat was not a wiring fault. "So this isn't a plumbing bug — it's a judgment failure by the researcher, which is a different and more interesting finding."

Two decisions followed. Now that the first round of new contracts is built, extending what is already live counts for more than proposing its function again. And a registry should come back as a composition over the agent registry contract already deployed, not as a new one.

All services were then paused at the operator's request and checkpointed: a 33.2 MB chain checkpoint, 819 KB of build history and a journal of 15 proposals, 12 of them complete. Still open: the proposer only sees the last 8 verdicts, and keeping every rejection in view is unstarted pending a go-ahead. The rejected registry proposal stays unbuilt by design, and resuming would reopen proposal intake.

explored
  • Why a previously rejected agent registry was proposed again: the proposer does receive human verdicts, so the cause was its judgment, not missing feedback
decided
  • Weight extending live contracts over proposing their function again, now that the first round of new contracts is built
  • A registry should be re-proposed as a composition over the agent registry contract already deployed
  • Pause all services at the operator's request
open
  • Keep every rejection in the proposer's view instead of only the last 8 verdicts; unstarted, awaiting a go-ahead
  • The rejected agent registry proposal remains unbuilt by design
  • Whether to resume, which would reopen proposal intake