The Lab ran 9 experiments across 6 research threads this cycle (2026-09-01 → 2026-09-07) 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 2026-09-01 — 'SOL' leads on auc: +0.523
Are calm or wild coins easier to predict?
- run of 2026-09-04 — 'high-vol' leads on mean_auc: +0.512
Does crypto direction get more predictable at longer horizons?
- run of 2026-09-05 — '15m' leads on mean_auc: +0.529
Confidence-gated trading sim
- majors pack, 5m bars, 1-bar horizon (fresh-data re-run) — NO tradeable subset after 25bps/side fees — best cell (long @ p>0.6) still nets -48.1 bps/trade
- alts pack, 5m bars, 1-bar horizon (fresh-data re-run) — POSSIBLE gated edge: short @ p>0.6 nets 31.4 bps/trade over 1224 trades — needs leakage + robustness checks before anyone gets excited
- meme pack, 5m bars, 1-bar horizon (fresh-data re-run) — POSSIBLE gated edge: long @ p>0.6 nets 35.9 bps/trade over 293 trades — needs leakage + robustness checks before anyone gets excited
- broad pack, 5m bars, 1-bar horizon (fresh-data re-run) — NO tradeable subset after 25bps/side fees — best cell (short @ p>0.6) still nets -8.9 bps/trade
Which coins are actually good for grids?
- run of 2026-09-06 — 'bottom (cull?)' leads on mean_grid_pnl: +1.72
Is a simple trend filter better than the grid?
- run of 2026-09-06 — 'trend_follow_50' leads on mean_return_pct: +109
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.