How good are these forecasts?
Every forecast scored on seasons the model never saw while its settings were chosen.
Held-back seasons
| Forecast | Forecasts | Log loss | Favourite won |
|---|---|---|---|
| The model | 437 | 0.630 | 76% |
| Base rate | 437 | 0.653 | 73% |
| Recent form | 437 | 0.656 | 75% |
Against the betting markets
Each market's price before the start, scored on the same finished games as the model. A negative figure means the market was more accurate.
| Market | Since | Priced | Model log loss | Market log loss | Model vs market |
|---|---|---|---|---|---|
| Bookmakers (Pinnacle, DraftKings, bet365) | 24 Jan 2026 | 31 | 0.557 | 0.514 | −8.5% |
| Polymarket | 24 Jan 2026 | 120 | 0.644 | 0.688 | 6.3% |
| Kalshi | 14 Feb 2026 | 46 | 0.585 | 0.662 | 11.7% |
Ratings use every game in the related competitions (see Sources); only Top 14 games are scored in the track record.
The simulated table adds bonus points the union way: four for a win, two for a draw, one for losing by seven or less, one for four tries (at each side's usual rate).
International call-ups during the Six Nations and test windows weaken clubs unevenly; the model doesn't know who is away.
Questions and answers
How accurate are PodiumOracle's Top 14 predictions?
On 437 games from held-back seasons (2024–25 to 2026–27), which the model never saw while it was tuned, the side it favoured won 76% of the time. Its log loss was 0.630, against 0.653 for the base rate and 0.656 for recent form. Against the betting markets on the same finished games, its log loss was 0.557 against the bookmakers' 0.514 over 31 games, 0.644 against Polymarket's 0.688 over 120 games and 0.585 against Kalshi's 0.662 over 46 games (lower is better).