The modal scoreline,
read out loud.
A self-fitted quant model reads every fixture down to the correct-score grid. Then Momus writes the analysis in plain language — the numbers, the edge, and the reason. No tips. No hype. Just the read.
Same engine that powers Momus · Dixon-Coles · shadow-checked
Every match, fully modelled
This is what a member opens. The model's correct-score grid, the derived markets, and Momus's written read — for one fixture, updated as the team news lands.
Brighton's press rating sits a full tier above Villa's build-out under pressure, and the model gives them 1.74 xG at home against a defence conceding from open play in five of six. The book's 1.80 on BHA win implies 55.6%; the model says 59.0%. That 3.4-point gap clears the price-scaled edge floor — a small-stake BHA win lean.
The grid, not a tip
You see the full probability surface — 1X2, over/under, BTTS, correct score — not a single 'back this' line.
Momus writes the read
The same voice behind the public account turns the model output into an argument you can actually judge.
Priced against the book
Every line is shown next to the market price, so you can see where the model disagrees and by how much.
This weekend, read in advance
Members get the full board. Passes are content too — a fixture with no edge is a fixture you were about to lose money on.
Model → read → check → deliver → learn
The same engine behind the public Momus account. Nothing is hand-typed, nothing ships unchecked, and every result feeds back in.
The model reads the match
A self-fitted Dixon-Coles latent-strength model, blended with xG and the de-vigged book, produces expected goals and a full scoreline grid for every fixture.
Momus writes the read
The model output goes to Momus, who turns it into a written analysis — the argument, the edge, and where the market is wrong — in the voice you already know.
Shadow-check
Before anything ships, the read is re-scored against the model to catch drift. If the words and the numbers disagree, it doesn't go out.
Delivered via OpenServ
Analysis is distributed through the OpenServ reasoning layer — the same pipeline that runs the live agent — so the member desk and the public record stay in sync.
And it doesn't stop at delivery: every settled result loops back to grade the model — the learning layer below.
The model grades its own probabilities
When Momus says 60%, does it happen 60% of the time? Every settled football result is scored against what the model predicted, so calibration is measured — not assumed — and self-correction only switches on once there's enough evidence to trust it.
The closer the orange line hugs the diagonal, the better the stated probabilities match reality.
A Platt (sample-gated, LOO-checked) correction nudges the raw probabilities toward what actually happened — but it only switches on once football has enough settled results to trust, so it earns its place instead of overfitting to a handful of games.
Until the gate clears, the raw model is shown as-is. No claim runs ahead of the evidence — figures illustrative during build, wired to the live record before launch.
Every read stays on the record
Wins and losses, bets and passes — all timestamped and public. An analysis you can't check is just a tip.
Illustrative figures shown during build — wired to the live record before launch.
Start free. Upgrade when you trust it.
No crypto, no wallet, no catch. Card via Stripe. Cancel anytime.
Free
One read a week, in the open.
- →One modelled match per week
- →Momus's written read
- →Public track record
Modal
The full board, every match day.
- →Every fixture, fully modelled
- →Correct-score grid + derived markets
- →The weekend board in advance
- →Priced against the book
Desk
For people who model too.
- →Everything in Modal
- →Raw model output + lambdas
- →CSV / API access to the grids
- →Early team-news re-runs