Automated Trade Alerts Need a System-State Audit TrailState Street SPDR S&P 500 ETFBATS:SPYquantsignals_alphaAn automated trade alert is an output. It is not an audit trail. To evaluate an automated process, separate four layers: 1. Observation: What market data did the system read, and at what timestamp? 2. Decision: What rule or model converted that observation into an eligible setup—or rejected it? 3. Execution: Was an order proposed, submitted, acknowledged, partially filled, filled, cancelled, or left uncertain? 4. Reconciliation: Does the internal position and order ledger match the broker’s record after fees, partial fills, and delayed updates? Without these layers, a chart marker can create false certainty. A “buy” label may represent only an idea. A submitted order may never have reached the venue. An ambiguous timeout may hide a fill. Retrying before reconciliation can duplicate exposure. A neutral system-state checklist Before interpreting any automated action, record: - instrument and session; - source timestamp and data age; - paper or live execution mode; - model or rule version; - eligibility and invalidation conditions; - position-size and loss limits; - order identifier and state transitions; - partial fills, costs, and slippage; - stop or kill-switch state; - broker reconciliation time. Data health should also be explicit. “Current,” “stale,” “reconnecting,” and “no new data” are materially different states. A frozen last value should never be presented as a current one. Why refusals belong in the record A complete audit trail includes actions the system declined to take. Risk-cap refusals, conflicting evidence, closed sessions, stale inputs, and unresolved prior orders affect the opportunity set. Removing those states makes the remaining trades look more selective and more certain than the actual process. Performance review should therefore use the complete distribution: wins, losses, refusals, inactive periods, costs, drawdown, and timestamps. Isolated outcomes cannot establish robustness. Invalidation Trust in the process weakens when mode labels disappear, timestamps are missing, state transitions cannot be reconstructed, broker records disagree with the internal ledger, or stale data continues to drive actions. It strengthens when an independent reviewer can replay the same evidence chain and reach the same recorded state—even if the trade lost money. Educational information only; not investment advice. Automated systems can fail, data can be delayed, and orders can behave differently from simulations. Trading involves risk of partial or total loss.