RaoscaffResearch
Prediction Series · Lock · Issue P-17
Prediction Series · P-17

The Tour, raced 200,000 times before it starts.

We ran the 2026 Tour de France General Classification 200,000 times in parallel — a market-anchored Plackett-Luce engine with a crash/abandon model — and read off the full distribution. Pogačar wins 56.9% of simulated Tours and stands on 91.8% of podiums; Vingegaard is the clear #2. The win odds sit just below the market by design (we add three-week attrition on top); the podium and top-5 decomposition is what the bookmakers don't quote. Locked before the Barcelona Grand Départ.

Type · Prediction Lock · Monte-Carlo field simulation (200k runs) Locked · 2026-06-21 · before the Jul 4 Grand Départ Resolves · 2026-07-26 · final GC, Paris Scored · Brier on the win + podium distribution
Tour de France 2026 · our simulated distribution
POG 56.9%
P(win) · 91.8% podium · 200,000 simulated races

Market-anchored (Pogačar 1/2, Vingegaard 5/2, Evenepoel 12/1, de-vigged) Plackett-Luce + an attrition layer. Vingegaard 25.2% win / 81.1% podium; Evenepoel 6.8% / 43.8%. Pogačar's 56.9% sits just below the book's ~58% by design — we add three-week crash risk on top of the market. The podium/top-5 split is a market-anchored extrapolation. Reproducible: fixed seed, committed script.

— 1 · The Locked Forecast

200,000 simulated Tours → one calibrated distribution.

We simulate the GC 200,000 times and lock the resulting distribution. Pogačar is the clear favourite, but a three-week Grand Tour has real attrition — which is why his podium probability (91.8%), not certainty, is the honest headline.

Tour de France 2026 GC · simulated distribution (200,000 races, seed 2026)
RiderP(win)P(podium)P(top-5)
Tadej Pogačar56.9%91.8%93.9%
Jonas Vingegaard25.2%81.1%91.1%
Remco Evenepoel6.8%43.8%76.9%
Isaac del Toro2.6%19.1%49.4%
Juan Ayuso1.7%13.1%37.1%
Field (others)6.8%
— 2 · The Engine

Plackett-Luce + a crash model — the race, run in parallel.

Each simulated race draws every rider's performance as strength + Gumbel noise (a Plackett-Luce ordering), then applies an independent crash/abandon probability that removes a GC contender from contention — the structural reality of a three-week Tour (Pogačar ~6%, others ~10–13%). Strengths are anchored to the de-vigged outright market (Pogačar 1/2, Vingegaard 5/2, Evenepoel 12/1). Because the market already embeds some attrition, our crash layer is a deliberate extra haircut — not a claim of independent information — so the simulated win probabilities sit a touch below the book (Pogačar 56.9% vs ~58%). The podium and top-5 numbers, which the market does not cleanly quote, are a market-anchored extrapolation that falls out of the 200,000-race ordering.

Verified 2026-06-20: 113th Tour, Barcelona Grand Départ → Paris, 4–26 July; 184 starters; route includes an opening team time-trial and a double Alpe d'Huez (stages 19–20). Sources: market odds + ASO route/startlist.

— 3 · The Variant View

Pogačar dominant — but the podium is the honest story.

The market makes Pogačar an overwhelming favourite; our simulation agrees on the win (56.9%) but quantifies what dominance really means at GC level: a 91.8% podium probability, with Vingegaard (81.1% podium) as the near-certain second man. The ~8% chance Pogačar is NOT on the Paris podium is almost entirely crash/illness tail — the variable a static odds line hides and a simulation surfaces.

— 4 · Methodology

Monte-Carlo field simulation: reproducible by anyone.

The simulator is a committed, fixed-seed script (scripts/sim/tour-de-france-2026.py) — 200,000 parallel races, deterministic. We synthesise and calibrate public signals; we do not claim a proprietary cycling-physics model. Scored by Brier against the official final GC in Paris.

— 5 · Pre-Committed Post-Mortem

How the distribution would be wrong — named in advance.

Pogačar off the podium (our ~8% tail) — a crash, illness, or a mountain implosion across the double Alpe d'Huez. A first-time GC winner (~10%) — del Toro / Ayuso / Seixas breaking through if both Pogačar and Vingegaard falter. We score the locked distribution honestly against the result, win and podium alike.

200,000 races locked before the Grand Départ — scored against the Paris GC.

RAOSCAFF locks P-17 on 2026-06-21, before the July 4 Grand Départ. The commit timestamp is the cryptographic lock, and the simulator is committed for reproducibility. We score the locked win + podium distribution by Brier against the official final General Classification in Paris.

Locked
2026-06-21 (commit timestamp on origin/main)
Resolves
2026-07-26 — final GC, Champs-Élysées / Paris
Source
ASO / letour.fr official final General Classification
Scored by
Brier on the win distribution + podium (top-3) outcomes

The simulator runs before your eyes — fixed seed, committed script — so the distribution is checkable, not asserted.