lablab.ai × Alpaca · Options Alpha Agents

The agent that grades its own refusals

Every trading bot can show you the trades it took. This one re-prices the trades it refused, and publishes whether refusing was right — including when it wasn't.

loading…

Equity

Read live from the Alpaca paper account on every cycle.

Did refusing pay?

Each trade the risk gate vetoed, re-priced at current marks. Scored symmetrically: a refusal that avoided a loss counts, and so does one that missed a profit.

RefusedSpreadCreditWould beVerdictBecause
GateRefusalsPricedCorrectCostlyNet of refusingHit rate

Same model, no gates

A naive agent — the same model, the same live chain, the same spreads — with the regime filter, challenger, risk officer, floor and position limit removed. It never submits. Its would-be trades are re-priced by the same scorer that grades our refusals.

WhenSpreadWould sizeCreditMarked nowIts reasoning

Where the edge is — and isn't

The deterministic layers run across four index ETFs at 1 and 7 days to expiry. Expected value at horizon-matched realised volatility and at the option's own implied.

SymbolDTERegimeShortδCreditPoPE @ realisedE @ impliedVerdict

Wrong if — graded

Before every trade the analyst pre-registers one concrete condition under which it is wrong. Each is parsed into a level and checked against every daily close since. A level at or beyond the short strike can only fire after the loss is locked in — that is a description, not a warning.

WhenDecisionConditionLevel vs short strikeStatus

Every order traces to a decision

Both directions: each order the broker holds must match the ledger record that submitted it, and each live submission in the ledger must exist at the broker. From the final session the cycle's ledger id is the order's client id.

Risk frontier

What the portfolio would be worth under each candidate minimum credit-to-width.

Decision ledger

Append-only. Refusals are shown, not hidden.

Orders

Every multi-leg order sent to Alpaca.

SubmittedStatusQtyLimitFilledLegs