Evidence before claims

Risk & performance

Methodology reviewed October 14, 2026

Profitability is not established. WSLI is validating a simulated trading system; a successful test order or a positive week would not prove a durable market edge.

Where the experiment stands

The scheduled October 7–20, 2026 experiment began with $10,000 in a separate paper account. The October 14 review had only three completed entry trades, including two exits triggered by execution safeguards rather than a normal strategy outcome. That sample is too small and operationally disrupted to establish profitability. All of those losses remain in the account result.

The October 20 endpoint is an operational checkpoint, not a profitability deadline. Further sessions require a separately approved schedule. The website does not extend the test, change risk limits or turn on live trading.

What would count as useful evidence?

  1. Reliable execution first. Reconcile every filled order, protective stop and exit against the broker. Record outages, rejected entries, forced exits and missing observations. Do not label an incomplete session successful.
  2. Fixed, versioned rules and a fresh test. Specify the universe, entry/exit rules, risk limits, evaluation dates and pass/fail checks before observing outcomes. Changes start a separately labelled test. Do not optimize on the same observations and call them independent confirmation.
  3. Net results, not only wins. Report trade count, average gain and loss, realized and unrealized P/L, drawdown, exposure and net expectancy after estimated fees, spread/slippage and relevant running costs. Distinguish broker-simulated P/L from a separately calculated, cost-adjusted estimate.
  4. Uncertainty and market conditions. Assess results across independent trading sessions and different conditions, with a risk-appropriate benchmark. Same-day trades may be correlated; treating every trade as independent can overstate confidence. Report uncertainty, not just a single return or win rate.
  5. Preserve the complete record. Show strategy outcomes and operational failures separately while retaining both in total system performance. Include no-trade sessions, specify missing data, and keep the untouched out-of-sample test distinct from development results.

Several weeks or months of clean observations may still be inconclusive. There is no universal number of trades that proves an edge. A cost-adjusted positive result with credible uncertainty bounds would be encouraging evidence, not a guarantee that future or live results will be positive.

This is an evaluation plan, not a claim that confidence intervals, a net-cost model or a complete multi-session performance report have already been computed. Repeatedly testing variants increases the risk of finding a result by chance; see Bailey and co-authors’ primary research on backtest overfitting.

Paper results are hypothetical

Paper orders do not reach a live exchange. Alpaca’s simulator omits important effects, including market impact, latency-related slippage, queue position and regulatory fees. Signals here use sampled IEX data from one exchange, while simulated fills use consolidated best bid/offer information. This difference can make apparent signals or fills unlike those available in live trading. See Alpaca’s paper-trading documentation.

The dollar P/L shown on the dashboard is the paper broker’s result. It is not a fully cost-adjusted live return, a prediction, or a guarantee. Do not annualize a few days of results or import returns reported in another strategy’s research paper.

Trading and AI risks

Real trading can lose money. Gaps, spreads, slippage, partial fills, stale data and outages can cause losses beyond a planned stop distance or daily entry cutoff. A halt marker is not confirmation that an account has been liquidated. A sleeping host cannot supervise positions.

Gem can misread evidence or produce incorrect explanations. Its reviews are bounded summaries, not verified forecasts or a record of hidden thoughts. The current deterministic executor applies rules and sends paper orders; the language model does not directly submit them. Neither role establishes skill or profitability.

No live-money decision should be based on this dashboard alone. The project does not offer personalized investment advice, guaranteed returns or account management for visitors.