121 possible scores

Model 02 · Premier League 2026–27 · live

The Premier League model.

The same Bayesian engine, re-tuned for club football: twelve seasons of results, a league Elo anchor, newly promoted clubs, and a simulator that plays the rest of the season 20,000 times.

4,560league matches it learned from at the season opener
20,000simulated seasons behind every forecast
380fixtures priced, every one before kickoff
0experiments past their pre-written bar, so far

This weekend · Matchday 6

Ten matches, priced before kickoff.

Issued on Friday 9 October, using every result up to Thursday night. Each match was played out 20,000 times; the bar shows how often each side won.

Sat 10 Oct · 7:30 AM ET

ArsenalvLeeds

70% Arsenal20% draw10% Leeds
Expected goals
2.08 – 0.64
Likeliest score
2–0 14%

Sat 10 Oct · 10:00 AM ET

Aston VillavBrentford

41% Aston Villa26% draw33% Brentford
Expected goals
1.55 – 1.35
Likeliest score
1–1 12%

Sat 10 Oct · 10:00 AM ET

ChelseavBournemouth

45% Chelsea25% draw30% Bournemouth
Expected goals
1.68 – 1.34
Likeliest score
1–1 12%

Sat 10 Oct · 10:00 AM ET

IpswichvFulham

25% Ipswich25% draw50% Fulham
Expected goals
1.15 – 1.72
Likeliest score
1–1 12%

Sat 10 Oct · 10:00 AM ET

SunderlandvBrighton

31% Sunderland27% draw41% Brighton
Expected goals
1.19 – 1.42
Likeliest score
1–1 13%

Sat 10 Oct · 12:30 PM ET

Man UnitedvTottenham

57% Man United23% draw20% Tottenham
Expected goals
2.00 – 1.10
Likeliest score
1–1 10%

Sun 11 Oct · 9:00 AM ET

Crystal PalacevNott'm Forest

35% Crystal Palace29% draw35% Nott'm Forest
Expected goals
1.23 – 1.22
Likeliest score
1–1 14%

Sun 11 Oct · 9:00 AM ET

HullvEverton

38% Hull31% draw31% Everton
Expected goals
1.13 – 0.98
Likeliest score
0–0 14%

Sun 11 Oct · 11:30 AM ET

LiverpoolvMan City

29% Liverpool25% draw46% Man City
Expected goals
1.26 – 1.66
Likeliest score
1–1 12%

Mon 12 Oct · 3:00 PM ET

CoventryvNewcastle

19% Coventry24% draw57% Newcastle
Expected goals
0.88 – 1.79
Likeliest score
0–1 12%

Expected goals here are the model's average goals per side across those 20,000 runs, not shot-based xG. Even the likeliest exact score happens only about one time in eight. Percentages are rounded, so a row may add up to 99. Team news after Thursday is not included, and none of these forecasts has been scored yet.

Method

How one Premier League forecast is made.

Every new issuance runs these steps from scratch, using only matches that had already been played.

  1. Twelve seasons in

    Every Premier League match from 2014/15 up to the cutoff day, with older matches counting less: a year-old result counts half.

  2. League Elo

    A running Elo rating for every club (K = 20). Newly promoted clubs start 75 points below the division average.

  3. Anchored prior

    Each club's attack and defence start from its Elo rating (weight 0.6), then the goals pull them where they belong.

  4. Bayesian fit

    Dixon–Coles, fitted with ADVI over 30,000 iterations, keeping 1,000 plausible versions of every club.

  5. 121 scores per match

    An 11 × 11 score grid for every draw. Clubs the model knows little about get a wider, less confident grid.

  6. 20,000 seasons

    Every remaining fixture is played out 20,000 times under Premier League tiebreak rules. The published numbers are those frequencies.

One forecast, start to finish

Arsenal v Coventry, the season opener.

Coventry had never played in the model's twelve-season archive, so their strength came from the promoted-club rule and the model's own prior. Here is what the model published before kickoff, and what happened.

Published before kickoff · 21 Aug 2026

From 20,000 simulated seasons. Each figure carries a simulation error of about ±0.3 percentage points.

What happened

3–0

Arsenal won, the outcome the model rated most likely. The forecast's ranked probability score was 0.031, close to perfect (0).

Also on that opener board: Arsenal v Man City in November, priced at 44.5% · 28.7% · 26.8%.

Contributing factors

Everything that moves a Premier League forecast.

Follow one real match through the machine, from the data it reads to the forecast it publishes. Every number is a real setting from the published model.

Every fixture of the season

From 121 scores to 370 fixtures.

The forecast published on 25 August, after the opening weekend: every remaining fixture, home club down the side, away club across the top. Taller means a likelier home win. Hover any cell, or pick a club.

Home win most likelyDrawAway winRings: played before this forecast

The season outlook

What 20,000 seasons said on day one.

Before a ball was kicked on 21 August, the simulator counted how often each club finished first, in the top four or in the bottom three.

Median points needed: 86 for the title, 67 for fourth place, 38 to survive. Simulation counts, not guarantees: within one forecast, team strengths are held fixed for the rest of the season, so the simulation knows nothing of injuries or January transfers. They update when a later forecast re-fits on new results.

The live cycle

A forecasting machine with brakes.

Each step only runs if the one before it passed. Results enter only when two independent sources agree exactly; any disagreement stops the machine.

Forecasts as evidence

A forecast is a timestamp, not a draft.

Each issuance prices every remaining fixture and is sealed. Results are scored against the freshest forecast that still came before kickoff.

Sealed

Every issuance

All remaining fixtures

370

The 25 August issuance priced 370 fixtures, from days to nine months ahead, each stored with its time, inputs and model version.

Beside

After new results

A newer forecast

+1

New results trigger a new issuance. It sits beside the old one and never replaces it.

Missed

If a cycle fails

No backdating

×

When a run stops, nothing is filled in afterwards. The older forecast stays the one that gets scored.

Scored

After the final whistle

Freshest forecast wins the slot

90′

Each result is scored against the most recent forecast issued before that kickoff, never one made later.

In the product plan

Fixed horizons

T−24h

The specification adds separate day-ahead and near-kickoff forecasts for every match. Planned, not yet live.

Experiments

Every idea had to pass a test written in advance.

Each experiment fixed its pass mark before seeing results, usually an improvement of 0.001 in RPS with an interval entirely below zero. None has passed yet, and every result is published.

15 Aug 2026Not proven

Does it beat plain Elo?

Six seasons, 2,280 matches. The model was ahead, but the interval crosses zero, so the pre-written pass rule failed.

Δ −0.0012 · interval −0.0028 to +0.0005

15 Aug 2026Not adopted

Four tuning levers

Memory length, fixture congestion, a faster-moving home edge and post-window uncertainty: 47 runs on older seasons only. Shorter memory was worse every time.

best combination −0.00007 · bar −0.0010

15 Aug 2026Not proven

A season it had never seen

2025/26, scored once. The model finished slightly behind Elo, inside the noise; one season can only detect a much larger gap.

Δ +0.0010 · interval −0.0026 to +0.0045

26 Aug 2026Not adopted

Refit before every matchday

The estimate leaned towards fresher fits, but it missed the bar and its interval crossed zero. The owner switched to matchday refits anyway, recorded as a decision rather than evidence.

Δ −0.00022 · bar −0.00030

26 Aug 2026Not adopted

Start from the market's view

Each refit began from what opening prices implied about every club. The gain was a fifth of the bar.

Δ −0.00019 · bar −0.0010

30 Aug 2026Not adopted

Wider grids for thin evidence

The largest effect measured so far, on 85 matches involving clubs the model barely knew. One of its two intervals crossed zero.

Δ −0.0041 on 85 fixtures

1 Sep 2026Not adopted

A public rating for promoted clubs

Checked against 30 promoted clubs. It was closer on average, but better in only 15 of 30, and the interval crosses zero.

error 57.7 vs 64.3 Elo

2 Sep 2026Refused

Add shots on target

The fit converged but failed its pre-set quality certificate, so no score was ever computed.

no score: fit failed

2 Sep 2026Refused

Start promoted clubs lower

The only clean alternative had already measured worse on the tuning seasons. Running it could not have produced a usable answer, so it was not run.

−100 Elo seed: +0.00017 worse

Still running, or ended without a result

  • Running
    Recalibrated probabilitiesScored live beside the real forecasts since late August. No verdict yet.
  • Running
    Player availabilityA second live shadow forecast that adjusts for who is available. No verdict yet.
  • Running
    A longer memoryDoes a longer archive of past results help? Data gathered; no result yet.
  • Built
    Market start, narrowedA re-test only at strengths the model can compute. Reviewed, waiting to run.
  • Stopped
    A stronger market pullStopped by the model's own numerical safety check. No result exists.
  • Unsettled
    Wider grids, re-testedA much larger simulation still could not settle the table-level question. The live model is unchanged.

Results appear here once they are published in full. Δ is the change in mean RPS against the live model; negative is better.

The product it is becoming

Three questions, kept strictly apart.

The forecasts are being built into a match workspace. Its specification starts with three questions that are easy to blur together.

Question 1

What does the model expect?

Home, draw and away probabilities, expected goals and supported totals for each fixture, with the time the forecast was made.

Question 2

How have forecasts like this done?

The whole season from matchday 1, including missing, late and refused forecasts. Never just the good weeks.

Question 3

Does the evidence support acting on a price?

A likely outcome and a well-priced outcome are different things. Without enough evidence, the answer stays “no”.

MarketWhat the model outputsStatus
Home · draw · awayThree probabilities that always add up to oneIn scope
Goals and marginsExpected winning margin and the chance of winning by 2, 3 or 4 or moreIn scope
Spreads and handicapsThe exact goal-margin event a contract describes, settled by its own rulesIn scope
Anytime scorerEach player's scoring chance, with minutes and lineup status shown separatelyIn scope
First scorerNeeds its own event-order model and validationDeferred
In-playNeeds live data, state changes and latency rulesDeferred

Research roadmap

Nine workstreams, tested one at a time.

Each is tested on its own against the unchanged model. Only then do the winners get combined, and only after that does anyone look at prices.

    The rules of the work

    Rules first. Results second.

    Work moves through five milestones, with two gates only the project owner can open.

    1. M1

      Build and review

      Research harnesses, tests and independent reviews.

    2. Gate A

      Owner rulings

      No experiment starts from a general “continue”.

    3. M2

      Run the experiments

      Frozen, authorised runs. Every verdict published.

    4. Gate B

      Combine or not

      Two good parts do not automatically make a good whole.

    5. M3

      Product and scorers

      Market pages, plus scorer, goal and workload research.

    6. M4

      Prices, going forward

      Paper decisions on real future prices, never hindsight.

    7. M5

      Controlled release

      Only with an explicit decision and a rollback plan.

    Back toThe overview←