How it’s computed

Scoring

the points behind everything

Every match prediction is scored on its scoreline:

• Exact score — 10 points (12 in a Match of the Day).
• Right result, wrong score — 7 for a called draw, 5 for a called winner (8 / 6 in a Match of the Day).
• Goal bonuses — +2 for the home tally and +2 for the away tally, on top of a non-exact result.
• First scorers (Match of the Day only) — +4 per team for naming the team's first scorer, where correctly calling a clean sheet also counts.

Across the analytics pages, per-match figures use scoreline-only scoring so everyone is measured on the same basis; the official standings include the scorer bonuses too. The gap between a player's scoreline points and their official total is exactly their scorer bonuses.

Standings & stability

the table and how steady it is

The Standings page is the live league table; the Ranking page replays each player's position after every scored match. From those rank paths we measure how settled a player is:

• Mean rank — their average league position.
• Std — how widely that position spreads (a big sustained climb shows up here).
• Stability — based on oscillation, the average number of places a player moves between consecutive matches.

100 means the rank never moved. Read it next to mean rank: a near-perfect stability can also just mean "stuck at the bottom", not "doing well".

Similarity

who predicts alike

For every pair of players we compare the scorelines they both predicted. Each shared match earns a partial-credit similarity:

Identical scorelines score top marks; the same winner with a near margin scores high; a wrong result still earns a little if a goal count lines up. Crucially the matches are consensus-weighted: games the whole pool calls the same way count for little, while the genuinely divided games count most — so the score measures agreement on the hard calls, not the obvious ones.

The heat map shows every pair's 0–100 score. The 2D map places each player as a dot so that closer dots predict more alike; it is a flattening of all those pairwise distances, and the page states how much of the structure that flat view captures. Colours are data-found clusters of like-minded players.

Profiles

the per-player scouting report

Each profile rolls a player's whole season into one card:

• Fingerprint — a radar of where the player sits in the pool (0–100 percentile) on accuracy, scoring, consistency, stability, draw-love, risk, contrarian streak and goal aggression.
• Accuracy — how often they get the result, the exact score, and each goal count right.
• Style — chalk vs contrarian, cautious vs goal-happy, draw bias, how varied their scorelines are.
• Against the odds — skill vs luck, by pricing each game against the pre-match odds.

For the luck read we estimate, from simulations of each match, the expected points a player's exact picks were worth. Comparing that to what they actually scored gives a luck signal, and a deserved standing — where pick quality alone would rank them.

One honest caveat shown on the page: nailing a rare exact score is a low-probability event, so it reads as "luck" in this model even when it is skill. Treat points-luck as a signal, not a verdict.

Deserved standing

where you'd rank if luck played no part

Deserved standing answers: "If luck played no part, where would each player rank based purely on the quality of their picks?" Here's how it works:

• Gather pre-match odds — before each match, we collect the probability of home win / draw / away win from simulations (computer models running thousands of virtual matches) and pre-match odds (betting market prices, adjusted to fair probabilities).
• Build a scoreline model — using expected goals (xG) or win/draw/win probabilities, we assign a probability to every possible scoreline: "10% chance of 1-1, 8% chance of 2-1, 6% chance of 0-0..."
• Score each pick against every possible result — for each match, we ask: "If the game ended 0-0, how many points would their pick earn? If 1-0? 2-1?" We multiply each outcome's points by its probability and sum them up. This gives the expected points — what the pick was "worth" on average, regardless of what actually happened.
• Sum across all matches — add up every player's expected points. This is their total "pick quality" score.
• Rank by expected points — sort players by total expected points (highest first). That ranking is the deserved standing — where they should be if results matched probabilities perfectly.
• Compare to actual standing — luck = deserved_rank − actual_rank. A positive number means the player is ranked better than their picks warrant (lucky). A negative number means they're ranked worse than their picks deserve (unlucky).

Example: if Player A always picks the favourite to win 2-0, and Player B always picks wild scorelines like 4-3, Player A's picks have higher expected points (more likely to score well on average). If Player B happens to nail one 4-3 and jumps ahead in the real standings, their deserved standing will be lower — revealing that their rise is luck, not skill.

Player tags

what the badges mean

Each player card shows up to six tags, ordered most-distinctive first. Here's what each one means:

• Patchy entries — left 4 or more matches blank; incomplete predictions.
• Chalk-loyal — rarely picks against the pool consensus; follows the crowd.
• Contrarian — frequently picks against the pool consensus.
• High scorer — consistently earns more points per match than most of the pool.
• Low scorer — consistently earns fewer points per match.
• Sharp reader — high hit rates on outcomes, exact scores, and goal counts.
• Goal-happy — predicts high-scoring games with many total goals.
• Cautious — predicts low-scoring games with few total goals.
• Draw-prone — frequently predicts draws.
• Draw-shy — rarely predicts draws.
• Steady — rank doesn't fluctuate much from match to match.
• Volatile — rank swings wildly after each match.
• Big-margin gambler — predicts lopsided scorelines with big goal margins.
• Chalk merchant — backs the pre-match favourite in 85% or more of games.
• Upset hunter — frequently picks underdogs to win.
• Riding variance — scoring well above what their pick quality warrants; lucky so far.
• Due a bounce — scoring below what their pick quality warrants; unlucky so far.
• Maverick — picks are very different from everyone else in the pool.

Tags measure different things: some compare to the pool (chalk-loyal, maverick), others to the pre-match odds (chalk merchant, upset hunter), and others to the model's expected points (riding variance, due a bounce). A player can be chalk-loyal without being a chalk merchant, or vice versa.

Projections

where the group stage ends

The Projections page forecasts the final table after all 72 group games. For each player we take their own per-match points so far as a sample (skipped games count as 0, so low engagement carries forward) and resample the remaining matches many thousands of times, simulating all players together so we can read off each one's projected total, an 80% range, and the odds of finishing top.

The method is backtested: we forecast the back half of the played matches from the front half and check how often reality landed inside the predicted range. The raw ranges came out too tight, so they are widened by a calibrated factor until they cover about 80% — the page reports both the factor and the result, so the forecast states its own accuracy.

It assumes scoring rate and engagement hold; late dead-rubber games add real chaos a model can't see, and thin samples get wide ranges. It's the shape of the run-in, not a prophecy.