如何閱讀賽事預測？
入門 · 約 5 分鐘閱讀
predicts.website 為每場世界盃賽事提供統計模型預測與 AI 文字分析。本文說明各欄位代表什麼，以及早盤、臨場預測該如何配合使用。
1. 在哪裡找到預測？
打開首頁後，每張賽事卡代表一場比賽。點擊卡片可展開詳情，包括陣容、早盤／臨場預測、AI 分析師摘要，以及（如有）價值分析表。
預設顯示即將開賽的賽事，亦可切換「賽事中」或「已完場」。若只想看已有模型輸出的場次， 可勾選「只顯示已有預測的賽事」。模型通常會在開賽前約一週內，為即將開打的比賽自動生成預測。
2. 早盤 vs 臨場預測
本站提供兩個時間點的預測，用途不同：
- 早盤預測 — 約在開賽前 48 小時 產生。當時陣容、傷停或天氣資訊可能仍未完全確定，適合觀察模型在資訊較少時的「基線」看法。
- 臨場預測 — 約在開賽前 60 分鐘 更新。會納入較新的陣容、傷停、對賽往績調整及官方賠率，通常比早盤更接近開波當刻的形勢。
若同一場賽事同時有早盤與臨場，建議並列比較：機率若明顯轉向（例如主勝由 40% 升至 55%），通常代表最新陣容或數據令模型調整了主客入球率 λ。
3. 勝和負機率（全場／半場）
預測面板會顯示全場主客和與半場主客和的百分比，例如「主 45% · 和 28% · 客 27%」。這代表模型估計該結果在法定時間（含補時、不含加時與 12 碼）發生的機會。
解讀時請注意：
- 三項百分比相加應接近 100%（四捨五入後可能差 1–2%）。
- 機率最高的一項是模型「最可能」的結果，但不代表一定會發生 — 35% 的主勝仍意味約 65% 機會不是主勝。
- 半場機率反映上半場結果；與全場機率並不一定一致（例如半場和局機率可以高於全場和局）。
4. 入球大細、雙方入球、比數 Top
除勝和負外，模型還會估算：
- 入球大細 2.5 — 總入球多於或少於 2.5 球的機率。
- 雙方入球 — 兩隊是否都有入球的機率。
- 比數 Top — 模型認為最可能出現的幾個比分及其機率（例如 1-1 佔 12%）。
比數機率通常較分散：即使最高的一個比分，往往也只有約 10–15% 機會，因此比數預測難度高，宜作參考而非定論。
5. AI 分析師面板
展開賽事後，「AI 分析師」區塊會以文字摘要說明模型對該場的看法，包括全場／半場推介、可能比分，以及（如有官方賠率）價值較高的投注選項。
點開「模型點樣預測？」可看到方法概要：本站使用 Poisson 入球模型，並以 Dixon-Coles 修正低比分相關性；在有數據時會加入對賽往績、傷停與 AI 因素微調。
6. 實用閱讀流程
How to read match predictions
Basics · ~5 min read
predicts.website publishes statistical model outputs and AI summaries for World Cup 2026 fixtures. This guide explains each section and how early vs late picks fit together.
1. Where to find predictions
On the home page, each match card represents one fixture. Tap a card to expand lineups, early/late predictions, the AI analyst summary, and (when available) the value analysis table.
Upcoming fixtures are shown by default; you can also switch to Live or Finished. Enable “With predictions only” to filter matches that already have model output. Predictions are usually generated within about a week of kickoff for upcoming games.
2. Early vs late predictions
- Early pick — Generated about 48 hours before kickoff. Lineups and weather may still be uncertain; useful as a baseline view.
- Late pick — Refreshed about 60 minutes before kickoff, incorporating fresher lineups, injuries, head-to-head tweaks, and bookmaker odds.
When both exist, compare them side by side. A meaningful shift (e.g. home win 40% → 55%) usually reflects updated team news or adjusted goal rates (λ).
3. Win / draw / loss probabilities
Panels show full-time and half-time 1X2 percentages, e.g. “Home 45% · Draw 28% · Away 27%”. These are model estimates for regulation time (including stoppage, excluding extra time and penalties).
- The three outcomes should sum to roughly 100%.
- The highest percentage is the model’s lean, not a guarantee — 35% home still means 65% chance it is not a home win.
- Half-time probabilities can differ strongly from full-time (a high HT draw % does not require a high FT draw %).
4. Goals, BTTS, and top scores
- Over/Under 2.5 — Probability of more or fewer than 2.5 total goals.
- Both teams to score — Yes/no probability.
- Top scores — Most likely exact scorelines with individual probabilities (often 10–15% for the top pick).
5. AI analyst panel
The analyst block summarizes the model’s view in plain language: full-time/half-time lean, likely scorelines, and (when odds exist) selections with positive expected value. Expand “How does the model predict?” for methodology: Poisson goals, Dixon-Coles low-score correction, and optional AI factor adjustments.
6. Suggested workflow
- Find the fixture on the schedule.
- Read late (or early) full-time and half-time probabilities.
- Read the AI summary for context (injuries, form, etc.).
- For odds vs model gaps, see our EV guide.
Questions? ai@predicts.website