Money Machine
Performance
scoring$527.18 · 0.5266%
0 working orders
100% constrained-model confidence
0 contracts · rounded down
Cash retained
Candidate evaluation
Each row is a possible defined-risk options trade. The system compares payoff, maximum loss, liquidity, and data quality; “Cash won” means every candidate was rejected.
Recent agent activity
“Cash won because every generated candidate was excluded by competition policy, confirmation, pending, or existing-underlying state.”
Evidence
+The full candidate report was preserved for counterfactual audit.
Invalidation
×A later cycle may proceed after managed or pending exposure clears.
Hard gates
These are fixed safety checks, not model opinions. Every check must pass before an order can be sent; one failure keeps the account in cash.
How a decision becomes an audited order
Every cycle moves from broker evidence to a constrained model choice, then through code-enforced safety gates. The model can propose; only the deterministic execution layer can authorize.
-
01
ObserveAlpaca MCP
Live market, account, position, and order evidence.
-
02
GenerateCandidate engine
Defined-risk structures filtered for data and liquidity.
-
03
SelectDeterministic selector
Chooses one valid candidate or explicitly keeps cash.
-
04
GovernDeterministic gates
Position size, max loss, clock, and reconciliation checks.
-
05
ExecuteAlpaca paper order
Idempotent multi-leg limit order, only when every gate passes.
-
06
ProveDecision Passport
Inputs, reasoning, checks, order state, and outcome retained.
Evidence first. Risk always bounded.
Money Machine is designed around a simple rule: a trade is only interesting after it survives the controls. Each cycle starts with broker truth, narrows the opportunity set, and leaves behind enough evidence to reconstruct what happened.
Observe
Read the live account, working orders, positions, market clock, and option-chain evidence before forming a view.
Constrain
Generate only defined-risk structures, then reject weak data, poor liquidity, excessive loss, and late entries.
Act
Submit idempotent multi-leg limit orders only after every deterministic gate passes. The model never bypasses policy.
Prove
Checkpoint equity and seal the inputs, decision, checks, and order state into a durable decision passport.
Competition-window discipline. New exposure stops before the close, flattening begins on a fixed schedule, and official performance is measured only inside the published scoring window. Outside that window, the system preserves evidence and close authority without manufacturing a score.