The bet: fade overstretched moves.
Seven of eight coins were negative, four of them significantly so. Worse, it falsified its own thesis: we predicted the edge would live in the ranging (low-trend) regime, and that is exactly the bucket where it lost worst (SOL ranging t -5.99, ETH -3.33). Even correctly gated to chop, fading the move lost. The lone positive (ADA) came from its trending bucket, not ranging — i.e. noise.
Fade z-score stretches behind a hard chop gate (ADX below a max) — the deliberate opposite of trend_follow, built with the very regime filter that was supposed to make mean-reversion work. Large positive in-sample-to-out-of-sample gaps mark it as an overfit mirage. The deeper lesson: on this tape stretches continue more than they revert, so betting against the move is actively punished.
| Pair | Sharpe | PF | trade t | n |
|---|---|---|---|---|
| ADA | +0.17 | 1.54 | +1.04 | 43 |
| XRP | -0.19 | 0.68 | -1.00 | 49 |
| TON | -0.34 | 0.46 | -1.92 | 44 |
| SOL | -0.46 | 0.30 | -2.43 | 35 |
| AVAX | -0.59 | 0.17 | -3.44 | 28 ⚠ |
| LINK | -0.41 | 0.38 | -2.11 | 39 |
| BTC | -0.35 | 0.28 | -2.01 | 24 ⚠ |
| ETH | -0.34 | 0.53 | -1.78 | 56 |
⚠ = underpowered (fewer than 30 out-of-sample trades). Negative values in red.
A significant loser; the move it bet against simply kept going.
Every trade we place — wins and losses, from real money — is on the live track record. This page is the research that decides what we will and won't run.
See the live track record →