3x3 Speedcubing: The Ultimate Limits Forecast
History · Methods · Hardware · Math · Biomech · Top Cubers · Training · Statistics — toward a single & average forecast.
Statistical Modeling: 4 Independent Models
Four candidate models:
- Exp + floor. Grid search L. Pro: floor interpretable. Con: single-floor assumption.
- Gompertz decay. S-shaped; adds an inflection vs exp+floor.
- Power law. No floor; hits 0 long-term.
- GEV reverse-Weibull. Shape ξ < 0 gives finite endpoint; fit on 1982-2026 WR yields τ̂ ≈ 2.0 ± 0.4 s.
Walk-forward backtest (train 2003-2020, forecast 2021-2026):
| Model | 2026 forecast | Actual | Error |
|---|---|---|---|
| Exp + floor | ~3.50 s | 2.76 s (Zajder) | +0.74 |
| Gompertz | ~3.20 s | +0.44 | |
| Power law | ~2.90 s | +0.14 | |
| GEV reverse-Weibull | ~2.85 s | +0.09 |
Key finding. Magnet-era acceleration (post-2017) makes GEV most accurate. BMA: GEV 0.55, Exp-floor 0.30, Gompertz 0.15.