World Cup 2026 · Market vs Model

Where the market and the math disagree

The full Polymarket board (all 48 teams, every stage) set against two venue-neutral simulations of the real bracket: an Elo Monte Carlo and a Voros McCracken run. Where the market and both models disagree is the whole story.

Market: Polymarket, 48 teams, 7 stagesModel 1: Elo Monte Carlo, 40k simsModel 2: Voros McCracken, 40k simsread through your betting principles

Polymarket board is LIVE, pulled June 28 2026. The interactive value board, stage ladders, and edge bars below update from the live Polymarket feed; the written findings are a dated analysis from the initial snapshot. Elo via eloratings.net. This report supersedes the Polymarket and Elo gaps flagged in the first ratings analysis.

Knockout stage. The interactive value board, stage ladders, and edge bars below are live for the fixed knockout bracket (Polymarket market vs the two bracket sims). The written analysis that follows is a pre-tournament (June) snapshot, kept for the record — its findings (Argentina back, the host premium, group-stage framing) predate every group result, which the market now prices in and the pre-tournament model does not. For the current model-vs-market read, see the live Futures & Group Plays board.
0

Bottom line

One back, several fades, and a host premium you can see. With the complete market and a real-bracket Elo model side by side, the cleanest signal is Argentina: the market prices them 5th to win it (8.6%) while the model, which has them as the world's 2nd Elo side, makes them a co-favorite (18.9%). The name-brand favorites England, Portugal, Brazil, Germany all price above what the model supports. And the venue-neutral model exposes the host premium the market is paying, most starkly for the USA. A second model, the Voros McCracken engine, now runs beside Elo as a cross-check (section V): it's far less top-heavy, and it narrows the live disagreements to Argentina (both models back) and Portugal / France (both fade).
48 / 48
market & model now complete
Argentina
model's biggest "back" (+10.3)
USA
R16: market 46% vs model 14%
0.879
Fogle vs Elo (now all 48)

Honest caveat up front: a pure-Elo simulation is structurally top-heavy (its top two, Spain and Argentina, hold 44.7% of the title vs the market's 24.3%), so treat the favorite-side "edges" as a strength read, not a license to hammer the chalk. Your own principle, fade the shortest price, is the counterweight.

1

What changed since the first analysis

The two data gaps the first report flagged are now closed.
  • Polymarket: 9 teams → all 48, full stage ladder. The first analysis used a Poly Win% column that had been typed into the workbook for only 9 favorites, and those nine were normalized among themselves, so Spain read as 33%. The complete board (Win Group, Make R32, R16, QF, SF, Final, Win) puts Spain and France tied at 15.7% to win it all, which actually agrees with Fogle's 2.8 tie and reverses the earlier "market splits them 2 to 1" reading.
  • Elo: 35 teams → all 48. Completed from eloratings.net (the same World Football Elo as the workbook's existing values, verified identical on the overlap: Spain 2155, USA 1726, Ivory Coast 1695). Fogle vs Elo over the full 48 is r = 0.879, Spearman 0.901 (the partial-35 figure was 0.887 / 0.915).
  • The "9-team tournament" claim is retracted entirely. It was an artifact of an incomplete capture, exactly as the first report's adversarial pass suspected.
2

The model: a venue-neutral Elo simulation of the real bracket

To compare against the market at every stage, the ratings need to become probabilities. That requires a tournament simulation.

  • Engine: World Football Elo win expectancy per match, 1 / (1 + 10^(-ΔElo/400)), with a draw model in the group stage that widens for even sides and shrinks for mismatches.
  • Structure: the actual WC2026 bracket from the schedule. 12 groups of 4 (round robin), top 2 plus the 8 best third-placed teams advance, then the real Round-of-32 group-slot matchups cascade through R16, QF, SF, and Final.
  • Runs: 40,000 Monte Carlo tournaments. Title probabilities are stable to about ±0.2%.
Deliberately venue-neutral. The model gives no team home advantage. That's a choice, not an oversight: it means the gap between a host's market price and its neutral-model number measures the host premium the market is paying, rather than burying it inside the model. Other simplifications, flagged honestly: group ties are broken by Elo rather than goal difference, third-place teams are assigned to bracket slots by eligibility matching rather than FIFA's exact lookup table, and a pure-Elo model is top-heavy relative to real markets. None of these move the headline gaps, but they are why the model is a lens, not an oracle.

3

The value board

Every team: market, the Elo model, and the Voros model on the title. Click a row to expand the full stage ladder (market, Elo, Voros). Sort any column. The 2-Model Read flags where both models move the same way against the market: BOTH BACK (both at least 1pt over market), BOTH FADE (both at least 1pt under), or SPLIT (the models disagree, stand down). HOST and +600 to +1500 still apply.

"Edge" is model minus market in percentage points on the title. Positive (green) = the model makes the team more likely to win than the market prices. Negative (red) = the market prices it higher than the model supports. Host rows: a large market-over-model gap is the priced home premium, not a fade.

Title-odds edge, biggest disagreements

V

The Voros cross-check: a second, independent bracket run

The same 40,000-sim bracket, re-run with the Voros McCracken match engine instead of Elo. Group games use Voros win/draw/loss; knockouts use P(win) + ½·P(draw). Venue-neutral, like the Elo model, so the two are directly comparable. The point isn't a third vote; it's a disagreement screen.

Voros is far less top-heavy than the Elo model, which is its single most useful contribution. Where the Elo Monte Carlo concentrates 45% of the title on its top two (Spain 25.7%, Argentina 18.9%), Voros spreads it: its top two (Spain 14.6%, England 12.6%) hold just 27%, actually flatter than the market's 31%. That directly tempers the top-heaviness the Elo model was dinged for, and it pulls the favorite-side "edges" back toward earth.
14.6%
Voros Spain title (Elo 25.7%)
27%
Voros top-two (Elo 45%, mkt 31%)
0.93
Voros vs Elo team-strength r
2
teams both models back, 2 both fade

Two models, read as a veto, not a trigger

Voros and Elo correlate 0.93 on team strength, so they're not independent witnesses; they're two calibrations of the same results-trained signal. Per your own integration memo, the right use of agreement is to stand down when they split, and to treat agreement as a reason to look harder, never as an automatic bet. Reading the title board through both:

ContenderMarketElo MCVorosTwo-model read
Argentina8.6%18.9%11.9%BOTH BACK Voros tempers Elo's inflation, still over market
Portugal9.4%4.6%3.2%BOTH FADE name-brand premium, both models well under
France15.7%13.6%8.9%BOTH FADE co-favorite both models price lower
England10.8%7.2%12.6%SPLIT Voros backs, Elo fades, stand down
Brazil8.1%5.1%9.5%SPLIT Voros backs, Elo fades, stand down

What the re-price says about your four flagged contenders

  • Argentina survives as the one contender both models put above the market, the cleanest signal on the board, though Voros's 11.9% is a soberer back than Elo's 18.9%.
  • Portugal is the clearest fade: market 9.4% (+964), both models in the 3 to 5% range. The market is paying for the badge.
  • England and Brazil are model disagreements, Voros likes them, Elo doesn't. Where your own two models split, the memo says wait, not bet.
The caution that comes with it. When both models gap the market the same way (Portugal, France), remember they share 0.93 of their signal, so "both models fade Portugal" is closer to one opinion than two. The market may be pricing a squad or form read both rating systems are blind to. Agreement narrows the watchlist; it doesn't validate the bet.
🦉Owl, on the two-model boardwhat survives

Voros's real gift here is humility: it flattens the Elo model's top-heavy favorites toward the market's shape, which makes the surviving disagreements more trustworthy. Argentina (both back) and Portugal / France (both fade) are the durable signals; England and Brazil are noise until the models stop splitting. And every one of these stays a lens, because two results-trained models are not the lineup-and-injury read the market already has.

4

Five findings

1. Argentina is the one clean "back"

Elo rates Argentina the 2nd-strongest team in the world (2114, behind only Spain). The market prices them 5th to win it (8.6%, about +1063). The model makes them a co-favorite at 18.9%. They sit in your +600 to +1500 value band, and unlike the other teams in that band, the model is above the market, not below.

2. The market pays up for name brands

Every favorite the model fades is a marquee side: Portugal (market 9.4% vs model 4.6%), England (10.8 vs 7.2), Germany (5.5 vs 2.4), Brazil (8.1 vs 5.1). Elo has them 6th, 4th, 10th, and 5th. This is your "betting reputation over form" warning rendered as a price: the market is a tier more optimistic on the famous teams than their Elo supports.

3. The host premium, finally visible

The venue-neutral model is the measuring stick. USA is the headline: the market gives them 46.3% to reach the Round of 16, the model only 14.2%, because by Elo the USA is the weakest team in Group D (behind Turkey 1911, Paraguay 1833, Australia 1777). That 32-point gap is almost entirely the home premium the market is paying. But it doesn't generalize: Mexico (model 59.9% to reach R16 vs market 52.1%) and Canada (47.8% vs 40.4%) are genuinely well placed in soft groups, so the model likes them even with no home edge. The "hosts" aren't one bet.

4. The value band, filtered

Your principles say value clusters at +600 to +1500 (a 6.25% to 14.3% implied chance). Four teams live there: England, Portugal, Argentina, Brazil. The model backs exactly one of them, Argentina, and fades the other three. The band is where to look; it's not the edge itself.

5. The favorite tension is real

The model's two largest "backs" are Spain (+10.0) and Argentina (+10.3), the two shortest non-France prices. But a pure-Elo model concentrates probability on the best teams more than markets do, and your own rule says fade the shortest price. So Spain's +10 is mostly the model being top-heavy, while Argentina's +10 also carries a genuine market-vs-Elo tiering disagreement. If you trust one favorite-side number, trust Argentina's, and treat Spain's as "strong, not underpriced."

5

Through your betting principles

  • "Elo over FIFA rankings." The model is Elo. Where Elo and the market disagree (Argentina, the host gap), that is the actionable space.
  • "Avoid the shortest price; favorites win less than expected." Directly cautions against the model's top-heavy love of Spain. Honored: Spain is framed as strong, not value.
  • "Value clusters at +600 to +1500." Used as the filter in finding 4. Argentina is the survivor.
  • "Hosts historically overperform and are underpriced." The market clearly agrees and is already paying for it (USA most of all). A neutral model can't confirm hosts are still value after that premium, it can only show the premium exists, so this stays a judgment call, not a model output.
  • "Assess bracket path; soft route underpriced." The model uses the real bracket, so path is in the numbers. Spain's high figure is partly group H plus a friendly draw; Argentina's is partly a tougher road, which makes the market's low price on them more curious, not less.
6

Owl, potato, judge: a quick adversarial pass

The process you like, applied to this model before anyone bets it.

🦉Owlwhat the numbers say
Think like an owl, slow and analytical: what are the hidden factors?

The market is a tier high on name brands and is paying a clear home premium for the USA. Argentina is the one place where a complete market and a results-based model genuinely disagree on tier, not just on a point or two. The host story splits cleanly: USA is premium, Mexico and Canada are real.

🥔Potatono politeness
Become a hostile critic. Find what's wrong.
  • The model is top-heavy and you know it. Spain at 25.7% and a combined top-two of 44.7% is a pure-Elo artifact. Half your "backs" are the model failing to price upsets, not the market being wrong.
  • Venue-neutral is a cop-out dressed as rigor. The USA "premium" might be the market correctly pricing home advantage that the model refuses to model. Calling the gap a "premium" assumes the market is wrong; maybe the model is.
  • One Elo snapshot, no xG, no injuries, no form. The whole model rests on one rating list, the same single-source fragility the first report warned about.
  • Group ties by Elo, third-place by approximation. Advancement numbers for the middle tier are softer than the decimal places suggest.
⚖️Judgethe ruling
Rule on it. Which attacks name a real, checkable fact, and which are just noise? "It's complicated" is not a verdict.

Conceded: Spain's edge is mostly top-heaviness, so it's downgraded to "strong, not value." The host gap is correctly labeled a premium that exists, not proof the USA is a fade or a back. Mid-tier advancement numbers carry wider error bars.

Survives: Argentina as the one tier-level market-vs-Elo disagreement (true even after discounting top-heaviness, because it's a relative gap between two favorites, not an absolute-favorite inflation). The name-brand fades (Portugal, England, Germany, Brazil all priced above Elo). And the USA premium as a real, quantified feature of the board. The model is a lens for finding disagreements, and three of them are worth a second look.

7

Caveats and files

Model caveats (repeated so they travel with the numbers)

  • Venue-neutral (no home advantage); host gaps are premiums, not value calls.
  • Pure-Elo, so top-heavy versus real markets; favorite-side edges overstate value.
  • One Elo snapshot; no xG, form, injuries, travel, or rest, all of which your principles call for before an actual bet.
  • Group ties broken by Elo, not goal difference; third-place slotting by eligibility match, not FIFA's exact table.

Inputs

  • Market: Polymarket normalized stage odds, pulled live June 28 2026 from the Polymarket gamma API (all 48 teams).
  • Model strength: World Football Elo, eloratings.net, June 2026.
  • Bracket: WC2026_Schedule.xlsx (real R32 through Final).

Files in ratings_analysis/

wc2026_ratings_v2.csv // 48 teams: Fogle + complete Elo + full Poly ladder model_vs_market.json // model vs market, every stage, every team build_v2.py / build_model.py // dataset + Elo Monte Carlo (40k sims) build_mvm_html.py // this report

Companion: WC2026_Ratings_Analysis.html (the first report: distribution, confederations, group-of-death, and the owl/potato/judge pass that first flagged these two data gaps).