Three live paper traders running the rules explained in Turtle Report against the real BTC price feed. Leonardo and Donatello are long-only -- short breakouts are detected but never traded (BTC perpetual funding costs aren't modeled in this project's data pipeline, so shorts are out of scope for them, not an oversight). Michelangelo trades both directions using a genetic-algorithm-evolved parameterization instead of the classic 20/55/10/20 values -- see "Turtle-Parameter GA" on the report page for how those parameters were found. Entries trigger once per day at the daily close, exactly as backtested; stop-loss and channel exits are monitored continuously against live price.

What each one actually needs to see

All three bootstrap from the full historical archive (2017-08 to present, ~3,300 daily bars) so their channels and volatility read are correct from day one, live. Once running, the minimum rolling window each decision actually needs is much smaller than the full archive -- for Leonardo/Donatello, 55 days of prior closes for the slow entry channel (20 for the fast one), 20 days of true range for ATR, 10 days for the exit channel; for Michelangelo, the GA-evolved equivalents (30/17/14/18 days). Donatello's classifier is the exception among the three: it was trained once, at startup, on every breakout in the full ~9-year archive (walk-forward validation only matters when testing a moving-forward accept/reject rate over time; a one-time live deployment just wants the most data it can get).