Strategy and research
How Invest uses high-recall LLM research with deterministic selection for autonomous paper-trading candidates.
Research is broad; decisions are constrained
Invest uses an LLM as a high-recall candidate generator across US options and equities research. The generator may surface possibilities worth examining, but it does not select positions, validate structures, or approve orders. Those responsibilities belong to deterministic systems.
Research principles
- Evidence first — a candidate must be grounded in available market and portfolio context, not treated as a standalone prediction.
- High recall, bounded authority — broad generation can improve coverage, but generation is never a substitute for deterministic selection.
- Explicit uncertainty — missing, stale, or contradictory inputs are reasons to defer or reject a candidate.
- Lifecycle learning without self-authorization — monitoring can inform future research while controls remain responsible for deciding whether a new entry is allowed.
What is still being proven
The system's north star is for deterministic selection to decide and break loops around LLM generation. That selector is still being proven load-bearing. Invest does not represent the current research loop as finished, reliable for live trading, or suitable as investment advice.
For the decision boundary behind the research loop, see how Invest works. For the controls that can block an entry, see risk controls.
Risk in practice
The implemented companion to Invest's conceptual risk controls. Two equity-relative ruin fences, a final-submit backstop, deterministic sizing that ignores model confidence, fail-closed operator knobs, and a plain statement of which controls are armed and which run shadow-first.
Operator console
How a human watches and steers Invest while it stays inside its paper boundary. The console's panels, the single audited pipeline every knob change flows through, operator identity, and the Telegram bot that mirrors alerts and selected writes.