Real results through of 21 stages — and model predictions for everything still to come.

_{📊 Data: Wikipedia (cites letour.fr / Tissot timing), refreshed daily at 08:00 UTC | 🤖 ML: trained on 2020-2025 + this year's form | ✨ AI commentary: baked at build time}

---

🏆 The Battle for Yellow — GC after Stage #

---

📋 The Story So Far — Stage Results #

---

🔮 Next Up: Stage — → #

_{The table ranks the full startlist for this stage — page through or search for any rider. The chart shows the first page's top 10. Rank and win % can disagree: the ordering comes from a ranking model (LambdaMART) that weighs current-tour form and full finish-order structure, while the percentages are the calibrated win/podium probabilities from the classifier — see the methodology.}

---

🎯 Stage-by-Stage Predictions for the Rest of the Tour #

Every rider's win probability (%) on every remaining stage — one column per stage, shaded per column so each stage's favourites stand out. Search for any rider, or sort by a stage's column.

_{💡 Click any row for the full per-stage breakdown — every rider's rank, win and podium probability. The pick is the ranking model's #1; its win % is the classifier's calibrated probability, so a consistent placer can out-rank a rider with a higher peak win % (e.g. on sprint stages).}

---

👑 Final Podium Forecast #

_{⚠️ Honest caveat: with only six Tours of training data, beating the "standings stay as they are" baseline on exact positions is genuinely hard — see the methodology for the numbers. The podium probabilities are the more reliable signal.}

---

👕 Jersey Projections — Who Wins in Paris? #

_{🚏 Green-jersey points are also scored mid-stage — the intermediate sprint pays 20-17-…-1 — and at the finish for places 4-15. Neither comes out of the podium models, so the projection adds each rider's observed rate of those points this Tour (the third bar segment). For the current green leader that segment is most of the story.}

---

🔍 Rider Analysis — Form vs Expectation #

---

📜 Historical Context (2020-2025) #

---

🎲 Model Scorecard — Honest Numbers #

_{🧪 Tour de France stages are genuinely hard to predict — long breakaways decide many of them. These numbers are cross-validated on held-out years, not cherry-picked. The headline KPIs belong to whichever model currently supplies the published stage ranking — the classifier and a LambdaMART ranker compete for that role on every retrain. Full details on the methodology page.}

---

📡 Data Freshness & Sources #

_{⏰ Last refreshed: — pipeline: scripts/fetch_tdf_data.py (Wikipedia wikitext → Parquet) then scripts/train_predictions.py (train + predict). Built with Dashdown.}