A live paper trader for the genuinely different hypothesis from Turtle Trading: instead of riding a price move that's already proven itself (a breakout), fade a price that has moved unusually far from its own recent average, betting on reversion. Real BTC feed, real fees, same leverage-capped position sizing as every other live trader on this site.

What is mean reversion?

Track a rolling average of price and how much it typically wobbles (its standard deviation) over the last N minutes. When the current price strays unusually far from that average -- measured in standard deviations, a z-score -- bet that it snaps back. A z-score of 3.5, this trader's actual threshold, means price would need to be 3.5 standard deviations from its own 200-minute average before it takes a position: a rare, extreme deviation, not a everyday wobble.

  • Entry. Price closes 3.5+ standard deviations below its 200-minute average -- go long, betting it reverts up. 3.5+ above -- go short, betting it reverts down. Both directions, unlike the long-only turtle traders.
  • Scaling in. If price keeps moving further from the average after entry, add another unit -- the opposite trigger from turtle pyramiding, which adds as a trend runs further in its favor. This is the classic "buy more as it gets cheaper relative to the average" approach.
  • Exit. Close the position once price has reverted all the way back to the average (this genome's exit_z evolved to exactly 0.0 -- no early profit-taking), or a fixed stop-loss 6×ATR from the average entry price if the reversion never happens and price just keeps going. Unlike turtles' trailing stop, this stop doesn't ratchet -- mean reversion has no "let the winner run" phase to protect, the stop exists purely to cap the risk of being wrong about reversion.

Why this exact genome, and the honest caveat

This is the actual winning parameterization from a genetic algorithm evolved against real 1-minute BTC data (4× 45-day windows, 2019-2022) -- see Processing's "GA Parameter-Search Experiments" table for the full run. It is not a cherry-picked nicer-looking config. It's deployed here specifically because the honest result deserves to be shown running for real, not because backtesting found an edge:

0% win rate on every closed trade, in all 4 backtested windows. Not one completed trade was ever profitable. Whatever positive-looking numbers appeared in testing came from either (a) a position still open and unresolved at the end of the 45-day test window -- unrealized, unproven paper P&L, not a real result -- or (b) losing less money than an even-worse buy-and-hold benchmark during a crash, which is not the same thing as making money. 3 of 4 training windows, and the holdout window too, ended with the position still open.

What that means live: entries are rare (a 3.5-standard-deviation move doesn't happen often) and, based on backtesting, positions may stay open for a long time waiting for a reversion that may not come within any reasonable window -- expect long stretches of "flat, waiting" punctuated by occasional trades that, on the evidence so far, are as likely to sit open indefinitely or hit the stop as to actually revert and close at a profit.