42 trades, 27 years, one rule. QuantGPT stacked three layers between the model and the market. The model wrote one rule. The stack did everything else. → DATA held S&P prices back to 1998. Delisted tickers still in. Fundamentals stamped by filing date, not by quarter. → COMPUTE replayed the rule across five moving-average lengths and four exit buffers. One prompt. Verdict: DATA VALIDATED, TRADE-OFF or NO EDGE. QuantGPT's 27-year backtest grew $10K to $61,526 with the rule. Buy and hold: $61,009. Same finish line. -24.4% worst crash against -56.8%. Sharpe 0.61 vs 0.44. 42 trades, gross of costs. → STRESS replayed the surviving holdings through tariffs, oil shocks, war onset. → TRADING sent the strategy to the user's broker under the user's keys. No tool the model could call placed an order. Swap Claude for ChatGPT next month. About ten minutes to plug the new one in. The three layers never move. Without this stack the model invents a Sharpe it never computed. With it, the model is just the pilot. The stack is the jet. Three layers, four pilots swapping through one door below ↓