The Lab ran 31 experiments across 8 research threads this cycle (2026-06-26 → 2026-07-05) on the idle GPUs — every verdict below is published as the data said it, wins, losses and coin-flips alike. No cherry-picking.
Which crypto coins can a simple ML model actually predict?
- run of 2026-06-26 — 'ARB' leads on auc: +0.532
- run of 2026-07-03 — 'ARB' leads on auc: +0.532
Wide-feature XGBoost direction
- majors pack, 5m bars, 1-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- alts pack, 15m bars, 1-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- alts pack, 15m bars, 3-bar horizon — NO EDGE — confirms coin-flip prior on short-horizon direction
- alts pack, 15m bars, 6-bar horizon — NO EDGE — confirms coin-flip prior on short-horizon direction
- alts pack, 5m bars, 1-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- alts pack, 5m bars, 3-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- alts pack, 5m bars, 6-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- broad pack, 15m bars, 1-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- broad pack, 15m bars, 3-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- broad pack, 15m bars, 6-bar horizon — NO EDGE — confirms coin-flip prior on short-horizon direction
- broad pack, 5m bars, 1-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- broad pack, 5m bars, 3-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- broad pack, 5m bars, 6-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- majors pack, 15m bars, 1-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- majors pack, 15m bars, 3-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- majors pack, 15m bars, 6-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- majors pack, 5m bars, 3-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- majors pack, 5m bars, 6-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- meme pack, 15m bars, 1-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- meme pack, 15m bars, 3-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- meme pack, 15m bars, 6-bar horizon — NO EDGE — confirms coin-flip prior on short-horizon direction
- meme pack, 5m bars, 3-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
- meme pack, 5m bars, 6-bar horizon — Marginal: AUC slightly above coin-flip; not bankable on its own
We tested the 'obvious' fix for our bleeding grid bot. It backfired.
- run of 2026-06-27 — 'BTC-downtrend' leads on total_pnl: +158
Can we score which coins are good for grids
- and cull the bleeders? — 'bottom (cull?)' leads on mean_grid_pnl: -2.67
Is a dead-simple trend filter better than our grid bot?
- run of 2026-06-28 — 'trend_follow_50' wins: mean_return_pct +62.80 vs 'buy_hold' +8.00 (+54.80, +685%)
That 'trend-following beats grids' result? We stress-tested it with fees.
- run of 2026-06-28 — 'trend_follow_50 net (OOS)' wins: mean_return_pct -29.80 vs 'buy_hold (OOS)' -47.70 (+17.90, +38%)
Does crypto direction get more predictable at longer horizons?
- run of 2026-07-04 — '15m' leads on mean_auc: +0.526
Are calm or wild coins easier to predict?
- run of 2026-07-04 — 'low-vol' leads on mean_auc: +0.509
Auto-published weekly by Gordon Lab. Each entry is a real backtest or model run on real market data; methods and full result tables live in the lab archive. This is research, not financial advice.