研究
预测如何构建——以及为什么值得信赖
每一个概率在发布前都经过 8×90 天前向回测门控验证，然后在赛事期间与真实结果进行实时评分——方法论、回测和局限性全部公开，包括那些失败的实验。
23 篇短文，22 篇方法论文档，23 篇研究与回测笔记。
- 3 个独立模型，取平均
- 8×90 天前向回测门控
- 失败实验均已发布
- 仅基于公开数据构建
你能信任这些数字吗？
为验证而生
决定这些已发布概率是否值得认真对待的因素：它们与真实结果的对照表现、与成功并列发布的失败记录，以及每一个数字背后的版本化记录。
完整论证 · 免费
Why trust these numbers
A probability publication is a credibility game. Anyone can publish numbers; the question is whether those numbers track outcomes once the matches finish. This page collects the d…
实时校准追踪器
数字能追踪结果吗？
按层级划分的 Brier score 和校准度，根据真实结果评分并在整个赛事期间持续更新。一个被标为 70% 的结果大约应有 70% 的时间发生——在此处进行验证。
阴性结果
那些失败的实验
每一个未通过门控的模型变体，连同完整判决一并发布。未发布的结果与成功的结果同样可见。
模型变更日志
每一个版本，公开记录
模型的版本化历史——每一次重新训练和架构变更，均标注其发布时 Brier score 并链接至完整笔记。每个页面上的数字都可以追溯到此处的某一行记录。
方法论要点
从这里开始
如果你想了解模型如何运作，首先要阅读的三份文档。完全免费阅读。
How we make predictions
How our 2026 World Cup prediction model works
Our 2026 FIFA World Cup forecasts come from a statistical prediction model that blends three approaches — an Elo rating system, a Dixon-Coles Poisson goals model, and a hierarchic…
How we make predictions
What we predict and how
For every prediction target — match outcomes, goal totals, scorelines, individual player events — there's a standard modelling approach and a set of input variables. This page cat…
Behind the scenes
Where our data comes from
The quality of any prediction depends on the data behind it. This page maps every data source we use — from free public archives to commercial feeds — and explains what each one p…
我们尝试了什么
研究笔记
模型构建过程中的决策日志：假设、回测、结果、发布/不发布判决。失败实验与成功实验一同记录在案。
Not shipped · 3 June 2026
A within-match chase layer "passes" the headline gate — and the placebo proves it shouldn't
The feasibility probe found that, after controlling for team strength, only
Shipped · 31 May 2026
Testing our approach on the Champions League final
The `/test/live/<slug>/` route renders the live-tracker pipeline
Not shipped · 29 May 2026
Is composite *coverage* the lever for the player-strength offset? (No)
player-composite's match coverage — whether honestly (point-in-time WC
最新文章
最近有什么变化
来自最近模型运行和研究发现的短篇笔记。
11 June 2026 · OnThePitch Staff
Five places the model disagrees with the consensus
Models don't know narratives. They read results, schedules, and xG rates. Here are five places ours diverges most from the consensus, from Ecuador over Germany to Raphinha as the #1 anytime scorer, Iran at 81%, Spain an…
11 June 2026 · OnThePitch Staff
Argentina and Spain: 0.6 points apart, nothing else in common
The model's two most likely World Cup winners sit at 17.5% and 16.9%, a gap well inside simulation noise. They get there via opposite paths. Argentina runs through the best defensive rating in the field and a penalty ad…
11 June 2026 · OnThePitch Staff
Why prediction markets keep underrating World Cup draws
Prediction markets, including Polymarket, consistently compress draw probabilities in World Cup openers. The model disagrees. Here's why the structure of prediction markets makes draws hard to price correctly, and what…