Mandate & projects
Top-level research authority is separate from agent autonomy.
| Project | Objective | Autonomy | Status |
|---|
Closed-loop research system
Owner mandate → research → deterministic evidence → adversarial review → shadow/paper → feedback
Research Owner + Mandate
Sets goals, universe, economics, risk and allowed autonomy.
Research Team
Director · Formula Engineer · Quant Researcher · Data Investigator · Failure Analyst
Evidence + Challenge
Runner · independent implementation · Quant Critic · isolated Reviewer · G0–G11 Guard
Prospective Loop
Shadow → paper-only deployment → reconcile → monitor → repair / transfer / retire.
LLM team
Drag a model from the pool onto an agent, or use the dropdown. Assignments are versioned and audited.
This dashboard has no password by owner request. Public mutation is intentionally limited to assigning an already-approved model to an LLM role. There is no shell, credential, live-order, deployment, or raw database endpoint.
Approved model pool
OpenCode runtime · current free model pool
Formula readiness
G0–G8 are required for shadow. Reproducibility alone is not promotion.
| Formula | Family | Lifecycle | Shadow decision | Gates |
|---|
Task queue
Scientific verdict and execution state remain separate.
| Task | Role | Type | State | Scientific |
|---|
Research memory
Negative evidence and migrated findings stay visible.
Owner configuration still required
The platform can be built while these remain unset, but dependent execution must fail closed.
Safety state
Recent audited decisions