The top layer turns the Brain outward. Describe a real treasury event in plain words; Opti interprets it, re-solves the affected closers as a test run, and hands back a recommendation you review and keep — or, within a signed contract, let it commit.
Scenarios in plain words
“A revolver was cut.” “Rates spiked 150bps.” Describe the event; Opti scopes it to the entity or account, re-solves the affected desks, and quantifies the impact — reasoning visible end to end.
The autonomy dial
Three settings: Advise, Review & approve, and Auto. Auto stays unavailable until you sign a Guardrail Contract — Opti ships inert.
Guardrail Contracts
The written bounds an action must satisfy to commit on its own. Anything that changes state needs a single-use token, minted only inside a contract.
Test-run isolation
Every simulation runs the real solvers against a sandbox and persists as a test run — never silently against the live book. Maker never equals checker.
What's unique
Autonomous, but governed. Opti is grounded in the same deterministic optimizers as the Brain, so its recommendations are reproducible and explainable rather than improvised. And it cannot act until you let it: default-inert, contract-gated, single-use tokens, maker-checker enforced, every step audited. It is the rare autonomous treasury layer you could actually put in front of an auditor.