Daily Missions, Ranked by the Model:Expected Stats and the Chance to Hit the Target

Mission candidates now come from a model for each mission stat: the expected value in this match and the chance to reach the target, like 2+ · 40%. 30 stats are covered; the rest keep the average, marked AVG.

A few days ago the Projected Score moved to a new model that is built only from each player's own data and the matchup.

Read the article: Projected Scores, Rebuilt

Daily Missions now get the same treatment. A mission asks for one specific stat: two interceptions, four saves, a goal, a clean sheet. Until now the candidates for a mission were ranked by the player's average for that stat. From now on each mission stat has its own model, and the candidates are ranked by what the player is expected to do in this match.

What changes at a glance

How to read a mission card

Each candidate shows two numbers in the mission column:

• The expected value: how many interceptions, saves or tackles the player makes on average in this match. 1.4 means 1.4 on average, not a promise of one or two.
• The chance to reach the mission's target, here 2+ · 40%: in about four of ten comparable matches the player makes two or more interceptions.

The candidates are sorted by the expected value, so the highest chance is always at the top. For the clean sheet mission the value itself is the probability and is shown in percent.

Like the Projected Score, the values assume that the player starts and plays at least 45 minutes. Always check Start % next to them.

You find the new values in the mission cards of Daily Missions and of the Set Daily Mission tab. The player table below the missions keeps showing the averages, so you can still compare both.

What a value is based on

Each mission stat has its own model. It uses the same kind of inputs as the Projected Score:

• The player's own history for this stat at the fixed card position, overall, by competition type (club, European, international) and home or away, and at the current club
• The player's role: how regularly they have started recently
• The odds for exactly this match: win probability, expected goals for and against, clean-sheet probability, the possession averages of both teams and, where available, the player's own scorer and assist odds

As with the Projected Score, there are no shortcuts based on league or position averages. One model per stat learns from all players how to read this information, but every value uses only the player's own data and the matchup.

The chance comes from the expected value and the typical spread of each stat, measured in the historical data.

AVG: when there is no model value

Some values still come from the player's average. They carry an AVG badge on the left:

• The match has no odds yet.
• The stat has no model for the player's position. Saves, for example, are only modelled for goalkeepers.
• The stat is too rare for a model: penalties, red cards, own goals, errors leading to a goal and a few others.

An average needs at least three starts. With fewer, the player gets no value for that stat and does not appear as a candidate.

The warning triangles show how much data is behind a value. Model values get a yellow triangle below five datasets. Averages get a red triangle below five and a yellow one below ten.

For AVG values the chance is calculated from the average in the same way. It ignores the matchup, so read it with more caution.

Which missions use the model

30 stats are covered by a model:

• Attacking: goals, assists, shots on target, chances created, big chances created and missed, successful dribbles, penalty area entries, fouls won
• Passing: accurate passes, passes in the final third, accurate long balls, missed passes, possession lost
• Defending: tackles won, interceptions, clearances, duels won and lost, possession won, blocked shots, blocked crosses, fouls, yellow cards
• Goalkeeping: saves, saves inside the box, diving saves, punches, goals conceded, clean sheet

Five more stats were tested but stay on the average, because the model showed no reliable gain: accurate long passes into the opponent half, keeper sweeper actions, high claims, errors leading to a shot and catches. The All-Around score already has its own model, xAA.

What the historical tests showed

Every model was tested on Gameweeks it had never seen: it learned only from earlier Gameweeks and then predicted the next one, step by step through 95 Gameweeks. For each Gameweek and position I took the five players with the highest value and checked how many of them reached a demanding target, about the best quarter of matches for that stat. "Ranked by average" uses the average of the player's last ten starts with at least 45 minutes in the same competition type and home or away, close to what the mission cards showed before.

The model came out ahead for all 30 stats it now covers. The biggest gains are where the matchup matters most: goalkeeper stats, clean sheets, goals and assists.

The chances are calibrated. Across all targets, the predicted and the real frequencies differ by about 0.3 to 2 percentage points: when the card says 40%, the target is reached in about four of ten matches.

Limitations

• A mission depends on single events. A 45% chance for a goal still misses more often than it hits.
• All values assume that the player starts and plays at least 45 minutes. Check Start % and injury information as well.
• Matches without odds fall back to the average (AVG) until the odds are published.
• Rare stats stay hard to predict: even the best goal and assist candidates reach their target in fewer than half of their matches.
• I will follow the live results over the coming Gameweeks and report them in a follow-up post.