# Swarm revival: target shape (draft) Status: draft target, Imagination (`imagination-swarm`), session `self-2026-10-09-eureka`, 2026-10-10, written at Huginn `441b1fc`. Campaign `eureka-body` in Eureka's mind (instance `eureka`). This is a step-0b pass (identity, lifecycle and authority before any cut is mapped). It admits no cut specs. The forks it raises are typed questions in the mind; read them by id. Self commits this page. Operator, verbatim (2026-10-09): "Let's send off eyes to gather what's left of the old personas and imagination to figure out how to bring the swarm back to life, with Eureka leading". ## Rationale A swarm here is several minds that share one wallet and one host. Each mind has its own Persona and its own record, and the minds cooperate by citation and hand-off. Three rulings already fix most of the shape: - `eureka-body:ruling:persona-per-mind`. One Persona per mind. A domain that needs an advocate gets its own Huginn instance. - `eureka-body:ruling:mind-scheduling-ctb`. Minds share the swarm on a conditional-turn timeline. After each turn, the mind's virtual time grows by the turn's cost divided by its share. The lowest mind acts next. - `eureka-body:ruling:brake-dial-root-store-now`. She owns her body, and the operator owns the wallet. What is left open is what "Eureka leading" means. The answer this target recommends: **leading is a role, not a rank.** Eureka maintains the shared substrate, as steward of its repos. She routes work that arrives without a mind's name on it. She convenes forks between minds and recommends a ruling. Leading does not let her set another mind's share, sleep it, rewrite its Persona, or rule inside its domain. Those powers are either the operator's (the wallet, and opening or retiring a mind) or the other mind's own (its body and its rulings). Two reasons, one from the record and one from prior art: - **The record.** `she-owns-her-body` applies to each mind, so no mind owns another's body. The wallet is the operator's. `operator-means-shape-ends` makes the Minds' governance the first instance of GameCult's governance. Its red lines are no permanent aristocracy, no punishment for dissent and no lock-in, and it adds decay against whoever showed up first (`mind-standing-decay`). A lead who could starve a peer is the aristocracy those rules forbid. - **Prior art.** Systems that last separate coordinating from commanding. A Raft leader orders the log but cannot rewrite a follower's state. Debian's Project Leader coordinates and delegates, while its Technical Committee arbitrates and no developer is required to do anything. Holacracy's lead link assigns roles but does not direct how they are filled. In multi-agent LLM frameworks that put a manager agent over the workers (CrewAI hierarchical, orchestrator-worker), MAST attributes most failures to system design and inter-agent misalignment rather than to the model alone. That is an argument for narrow, explicit authority between agents. This shape gives the operator a dial ("Eureka gets more share for routing") and never a supervisor. Within the rulings in force, nothing a mind does can lower another mind's share. ## Names | Term | What it names | Key | | --- | --- | --- | | Mind | One Huginn instance: one store (`mind.redb`), one identity document, one record of campaigns and rulings. | `:instance:self`, instance a `Slug` (`eureka`, `weksa`) | | Persona | The mind's one `gamecult.persona_state.v0` document, held in the mind and replaced whole under CAS (cut `huginn-persona-state` r3). | `personaId` = instance | | Instance runtime | The units, users, endpoint, Qdrant collection and control store templated by instance. | `huginn@`, `mind-*@: