World Cup 2026 · Predictions live
The line,
in the open.
Calibrated win/draw/loss probabilities for every 2026 World Cup match — published
before kickoff, shown against the market line, and graded in public. Not tips, not
hype: an honest model, validated against 20 years of World Cups, with every call
laid bare on the ledger.
What this is — and isn't
These are
calibrated probabilities, not claims to beat the bookmakers.
When the model says 60%, things happen about 60% of the time — verified across 320
World Cup matches (2006–2022), Brier score 0.150. Every prediction is timestamped and
graded on the
Open Ledger.
See
How it works.
The interesting part
Where we differ from the market
Live
These are the matches where our model's read diverges most from the betting market.
We're not claiming the market is wrong — it usually knows more than we do (injuries,
lineups, late news). These are simply where our model is most testable:
we've published our number, the market has published theirs, and the result settles it.
Click any match for the full breakdown.
Watch the ledger.
| Date | Match | Our read (W/D/L) | Market (W/D/L) | We lean | |
|---|
| Loading… |
All upcoming matches
Full slate
Model win/draw/loss probabilities for every fixture, with the market's de-vigged
implied probabilities alongside. Host nations (USA, Canada, Mexico) carry a
home-field adjustment.
| Date | Match | Our read (W/D/L) | Market (W/D/L) | |
|---|
| Loading predictions… |
The Open Ledger
Every prediction, graded
The whole point. Every match prediction is published before kickoff and graded here
afterward — wins and losses both, permanently. The git history is the proof: each
prediction is timestamped in a public repository before the match is played.
Matches graded
0
live, since June 11
Brier score
—
lower = sharper · 0.25 = coin-flip
Top-pick rate
—
3-way · ~50% is strong
Backtest Brier
0.150
320 matches, 2006–22
Public grading begins June 11
0 — 0
The ledger starts empty and fills automatically as matches are played. Every
prediction was posted before kickoff and is graded against the actual result —
no retroactive edits. The backtest validation (Brier 0.150 across five World Cups)
is on
How it works.
| Date | Match | Result | Our probs (W/D/L) | Pick | Hit |
|---|
Methodology
How this works
True Line is built on one principle: publish before, grade in public, claim only what
the numbers support. The model isn't exotic — what's rare is the honesty about what it
is and isn't.
The model
An Elo rating system built from the full international match record
(every men's international since 1872, ~49,000 matches), updated match-by-match with
a margin-of-victory weighting. Win probability comes from the rating gap; an
empirical draw model splits it into win/draw/loss; host nations get
a home-field adjustment.
Calibration
The model is
calibrated, not edge-claiming. Validated across 320
World Cup matches (2006–2022): when it says 60%, the outcome happens about 60% of the
time.
Brier score 0.150 (0.25 is a coin-flip; lower is sharper) —
consistent across all five tournaments tested.
| Predicted | Actual | Matches |
|---|
| ~14% | 19% | 36 |
| ~28% | 30% | 41 |
| ~42% | 39% | 60 |
| ~57% | 63% | 48 |
| ~72% | 68% | 71 |
| ~88% | 84% | 64 |
What it is not
This is not a claim to beat the bookmakers. The model's probabilities
are honest and well-calibrated — roughly in line with efficient market prices, which is
exactly what an honest model should be. We publish sound probabilities and grade them in
public. We don't sell the fantasy of a secret edge.
Known limits
The model rates team strength from results history, weighted for recency and the
importance of each match. It does not account for confirmed lineups,
injuries, squad rotation, or qualification scenarios — factors that matter more in
late-group and knockout matches, where a team's situation can change how they play.
This is part of why we publish where we differ from the market: the market often prices
in information our model can't see, and the result settles who was closer. The model is
on its firmest ground in the opening round, before tournament context takes over.
The open ledger
Every prediction is committed to a public repository before kickoff,
so the timestamp is verifiable — no cherry-picking, no quiet deletions. After each match,
it's graded against the actual result. Win or lose, it stays on the record.
Roadmap
World Cup first. The same transparent, calibrated approach extends to club soccer and
other sports after the tournament — always published-before, graded-in-public.
Disclaimer:
Predictions are generated by a statistical model for informational and entertainment
purposes only. They are probability estimates, not guarantees, and are not a claim to
beat any bookmaker. Past performance does not guarantee future results. If you choose to
bet, never wager more than you can afford to lose. If gambling is causing problems for you
or someone you know, call 1-800-GAMBLER.
MLB
In development
Paused
True Line started as an MLB run-line model. Here's the honest status — including why
it isn't live.
What happened
The run-line model
backtested well (+10.58% ROI, positive across four
historical periods). But the test that matters is forward, not backward —
and in
live forward-testing the edge did not hold. Backtested edges in
efficient betting markets very often fail to persist; catching that with discipline,
rather than publishing it anyway, is the whole point of this brand.
The honest read: a backtest that looks like an edge and a forward
test that says otherwise is the most common outcome in sports betting. We'd rather
show you that than sell you a record that won't repeat. The MLB model is paused and
in development — it isn't posting picks, and it won't until forward results justify it.
Why it's here
Transparency cuts both ways. If we're going to grade our World Cup predictions in
public, we're going to be equally honest about the model that didn't work out.
That's the standard.