—
Completed league fixtures ●
THE INTELLIGENCE DESK EPL 2023—27
DATA OVER HYPE ↗
SEE THE GAME
BEYOND THE SCORE.
Turn real Premier League results into transparent matchup probabilities, smarter scouting priorities, and decisions you can actually explain.
01 / THE BIG PICTURE
Season intelligence
—
Full-time total, both teams
—
Share of completed matches
—
Relative to historical baseline
02 / THE WORKBENCH
Matchup studio INTERACTIVE
Pre-match probabilities · Goal models · Scenario controls
FIXTURE PROJECTION
Who has the edge?
POISSON MODEL
HOME WIN—
DRAW—
AWAY WIN—
PROJECTED GOALS STANDARD CONDITIONS
Illustrative adjustment to home goals only; not an estimate of a specific tactical intervention.
DECISION SUPPORT
READOUT / 01Scouting briefing
MATCH UNCERTAINTY
Calculating…
Likelihoods help prioritise research, not replace it.
More predictableLess predictable
WHAT THE NUMBERS SAY 03 SIGNALS
01
Waiting for match data…
02
Waiting for match data…
03
Waiting for match data…
METHOD NOTE Predictions use only matches already played. They do not use injuries, lineups, or player tracking.
03 / SEASON CONTEXT
The league, in context
Season-specific results · Points · Recent form
RESULTS-BASED STANDINGS
2025/26 league snapshot
| POS | CLUB | PL | W | D | L | GD | PTS | LAST 5 |
|---|---|---|---|---|---|---|---|---|
| Loading standings… | ||||||||
Computed from completed matches in the selected season. Does not include disciplinary point deductions.
04 / THE EVIDENCE
Hold the model accountable
Chronological out-of-sample style replay · Every forecast scored
MODEL VS BASELINE
LOWER IS BETTER ↓Probability forecast error
SCOUTBOARD MODEL—
HISTORICAL BASELINE—
Model scoring will appear when results load.
Metric: multiclass Brier score = ½ × sum of squared errors across home/draw/away outcomes. Earlier seasons inform each prediction; that match's score is never visible at forecast time.
TRUST, NOT JUST ACCURACY
CALIBRATIONConfidence reality check
When the model says its most likely outcome has a given chance, how often does that outcome actually happen?
100%75%50%25%0%
Waiting for results
Predicted confidence Observed accuracy
Groups with fewer than five predictions are omitted. Short seasons produce unstable estimates.
05 / WHERE TO LOOK NEXT
Fixture priority queue
Ranked by model uncertainty · Current season
Checking upcoming fixtures…
06 / OPEN METHODS
No black box. No secret sauce.
01
Open, auditable data
English Premier League fixtures and final scores sourced from the OpenFootball public-domain dataset ↗. The browser retrieves the latest repository snapshot; the supplied offline subset is clearly labelled.
02
Explainable probabilities
Team home/away scoring and conceding rates are shrunk towards fixed league priors (six-match equivalent), then combined in an independent-Poisson model. We report the full home/draw/away distribution.
03
Transparent evaluation
Chronological rolling-origin backtesting compares multiclass Brier scores and log losses against a simple historical-league-frequency baseline, with same-day matches evaluated before updates.
RESEARCH LIMITATIONS
Not calibrated for betting, medical or financial decisions. Ignores team injuries, lineups, player-level quality, red cards, league point deductions and tactical systems. Historical results are not proof of future performance. Changing the home-attacking slider illustrates sensitivity, not causality.