Read this first. Every number here is paper trading on synthetic prices from our own simulator. The simulator has mean reversion and trends built in, so strategies that match it look good. That is not evidence of a real edge. The null, shuffle and out-of-sample columns are there to show where the "profit" disappears.

Strategy lab: each strategy run through the same 3-bot desk, risk engine and paper broker

Seed 42 = the showcase run. In-sample = 30 more seeds of the same simulator. Null = driftless random walk. Shuffle = same bars, order shuffled (same volatility and fees, no structure). OOS = a different synthetic market (weaker reversion, weaker trends, more noise). Cells show profitable seeds / total (median net %). Click a row to load its trades.

Equity

Every trade, with each bot's thoughts

Thoughts are from the bar the trade was opened. Each bot answers its typed question battery; code turns the answers into a verdict. initiated proposed the trade · paused stood aside on purpose · passed asked, said no · not asked skipped by desk rules. Exits are code only (stop, target, time, mean reached).

Overfitting demo

Experimental · not financial advice · no exchange keys exist in this project · source /workspace/ai-trading-desk · generated