Targets for a Premier League move
Choose a position. Players are ranked by projected Premier League level and grouped into tiers by their chance of reaching the top quarter of current PL regulars in that role. Pick a destination club in the sidebar to add each player's chance of becoming a regular there.
| Tier | Player | Role strengths | Age | Plays as | Projected PL level vs regulars | Value | Chance at club | Brief |
|---|
Shortlist side by side
Every starred player, from any position, on one scale. Star or unstar players from the shortlist or the profile.
Projected level in the Premier League
Gold: projected first season and range, on the cross-league scale used for every position.
Who gives the most for the money?
Projected first-season PL level against market value for the players loaded in the shortlist. The gold line is the value frontier: nobody cheaper projects higher.
Brief
Requirements are percentiles. Must-haves and nice-to-haves rank and explain; they do not predict success. The shortlist's Brief column follows this page.
Club role
Start here. Name the destination club and the job its coach wants done; the role template fills the requirements below, which you can then edit. Behaviours the data cannot measure become scouting questions on every profile.
Requirements
Mix tactical & technical style, contribution, physical and performance measures. Every condition is a percentile: style against players in the same role, contribution against players in the same position in his own league, physical against PL players in the same position, performance against all players in the position.
Freeze this shortlist
Records today's recommendation with every input as it stands now, so it can be judged later against what actually happened. In the platform this writes to the decision log; here it is kept in this browser.
Who meets it
Each cell is the player's percentile on a requirement. Gold = meets; dashed gold with ~ = borderline, so check it; dashed grey ? = unknown or still loading; empty = misses. Sorted by must-haves met, then fewest misses, then nice-to-haves, then projection.
League ladder
How far each league sits below the Premier League on the role-score scale (PL = 0), as held in the RJW database today. Gold rings mark leagues the current shortlist comes from.
How the forecasts have performed
Every model is tested on moves it was not fitted on and compared with the strongest simple rule we could find, not only the weakest. Ranges are 95% bootstrap intervals (2,000 resamples). Full-back figures come first because this dashboard is about full-backs.
Projection: average miss against the strongest simple rules
Lower is better. Cross-league movers with 900+ minutes after the move, first season. Bars are 95% intervals.
| Test | Moves | RJW miss | Half of last season | RJW gain (95% interval) | Inside range |
|---|
Gain = how much smaller the RJW miss is than the half-of-last-season rule; an interval crossing zero means no reliable difference. "Swapped" tests fit on the later season and test on the earlier one, as a stability check rather than a forecast.
Does it improve shortlisting decisions?
A replay of past windows. For each destination league and position, take the affordable players who actually moved there, rank them using only what was known before the move (frozen inputs: last season, market value then, incumbent then), keep each method's top five, and see what happened to them.
| Ranking method | Became regulars | Minutes share | Kept the role | Level after (on-ball z) | RJW advantage, regulars (95% interval) | RJW advantage, level |
|---|
- What it shows. With a €20m budget, the RJW model's top five became regulars 51% of the time against 40% for the whole pool and for a random five. It beats picking by market value, by last season's quality or by last season's minutes. It ties the simple minutes + value + age model on becoming a regular, but its picks then played at a higher level (+0.05, interval +0.02 to +0.09).
- The trade-off. Picking the best on-ball quality gives the highest level after the move but the fewest regulars: quality without opportunity. The RJW ranking balances the two.
- Limits. Only players who actually moved can be scored, so each pool is "the players who went to that league", not every player a club could have bought. Level after is only measured for players who played. Full-backs alone (26 pools) point the same way but are too few to be conclusive.
Model choice: can a position-specific projection beat the simple rule?
Seven candidates fitted on one season and tested on the other, in both directions. Gain = how much smaller the average miss is than halving last season (positive = better). Bold green = reliably better.
| Candidate | All · fit 23-24, test 24-25 | All · swapped | Full-backs · fit 23-24 | Full-backs · swapped |
|---|
- No position-specific version is reliably better. Full-back-specific gains flip sign between the two tests, so they are noise. Rejected.
- The current projection is kept, not replaced by halving. On average they tie, but the current one is better where this dashboard operates: moves up a level (gap 0.15+, gain +0.006, interval +0.000 to +0.012) and strong players moving up (28 moves: halving overestimated by 0.09, current by 0.03). Switching to halving would lift Scottish and Norwegian full-backs by about 0.13 with no evidence for it.
- Ranges widened for strong players. For players whose last-season score was above 0.5 (76 moves), only 60% landed inside the range, not 68%. Their ranges are now 20% wider, easing in from a score of 0.4. 15 of the 20 shortlisted full-backs get the full widening and the other 5 part of it.
- Known top-end risk. For the 35 movers scoring above 0.7, projections came in 0.11 too high on average (interval −0.27 to +0.06, not conclusive). Not corrected; flagged on affected profiles.
- Framing: the projection is a calibrated range with tiers, not a sharper forecast than a simple rule.
Chance of becoming a regular: ranking score (AUC)
AUC answers one question: given one signing who became a regular (50%+ of league minutes) and one who did not, how often does the model rate the regular higher? 0.5 is a coin toss and 1.0 is perfect. It is not the share of transfers predicted correctly.
| Test | Moves (regulars) | RJW screening | Minutes + value + age | Gain (95% interval) |
|---|
Projection by transfer route
Where the league adjustment earns its place: only moves into stronger leagues show a reliable gain over halving last season.
Are the chance estimates honest?
Each band: what the model said against how often players actually became regulars, moves into 2025-26. On the diagonal is perfect.
| Probability quality, moves into 2025-26 | Brier score | Skill vs base rate (95% interval) | Log loss | Calibration error | Average said |
|---|
Ranking (AUC) and probability quality are different questions: a model can rank well and still show the wrong percentages. Brier score and log loss judge the percentages themselves (lower is better). Skill = how much better than giving everyone the base rate. Calibration error = the average gap between what the model said and what happened. Bars on the chart are 95% intervals for each band; the 70%+ band (31 moves) is too small to trust either way.
Projection range, by position
Share of movers who landed inside the shown range. Target 68%. Full-backs (65%, 213 moves) and especially goalkeepers and central midfielders fall short.
Recent changes, tested
Possession adjustment raised how well passing and touches carry over between clubs. Physical intensity predicts level only for moves into the PL.
Test design
- Projection. 1,174 cross-league moves (561 from 2023-24, 613 from 2024-25), players with 450+ minutes in their main role before and 900+ after. 15% of arrivals played under 450 minutes and cannot be scored; players with 450–899 minutes came in 0.08 below projection. Read the projection as "level if he becomes a regular".
- Chance of regular. Trained on moves into 2024-25, tested on 1,326 moves into 2025-26 (552 became regulars). Checked again with the seasons swapped and with repeated cross-validation. League strengths refitted without test moves gave 0.628 against 0.624, so no leakage from that source.
- Becoming a regular is not the same as a successful signing. Minutes also reflect injuries, competition for places, squad shortages and selection. It is the measurable first step, not the verdict.
- Repeat movers. 224 of 2,403 players moved more than once. Holding out whole players, not just moves, gives the same ranking score (0.645 against 0.644), so repeats are not flattering the results.
Forward log: forecasts frozen before the outcome
Every new move is logged with the model's forecast at the time and scored later without refitting. This is the cleanest test, because nothing about the outcome could have leaked in.
| Cohort | Forecasts | Complete | Average predicted regular | Actual regular rate | Role-fit projection vs level so far |
|---|
Winter 2026 completes around November 2026 and summer 2026 in May 2027. The early role-fit correlation (0.39 on 82 players with 900+ minutes) is a first read, not a result. Shortlists frozen on the Recruitment page are the next layer: they record what was recommended and why, so the tool can later be judged on decisions, not just on players who happened to move.
How Role Fit works
Roles and scores
Each player-season is split into 12 side-aware roles from StatsBomb lineups. Scores are weighted z-scores of each role's measures, compared within the position family and league-season.
Cross-league score and projection
League and team strength are fitted together from players who changed league, with the Premier League at 0. The projection translates a player's multi-season record into the target league, calibrated on 775 real moves to a stronger league; the range is built to cover about two outcomes in three and is widened up to 20% for strong players, whose ranges were too narrow in testing. On average the central estimate is no more accurate than halving last season's score; it is kept because it does better for players stepping up a level. Read it as a range and a tier, not a precise forecast. Premier League projections include a physical adjustment from high accelerations and high-intensity distance.
Chance of becoming a regular
The chance of playing 50%+ of the club's league minutes in the first season, from quality, current minutes share, age, the step in league and team strength, quality against the club's current first choice, and market value. Tested on a later season it had not seen: AUC 62% (63% for the deal version).
Style, job types and managers
Nineteen style measures within role, with passing volume and touches adjusted for team possession. Job types come from clustering how players in each position play. Manager profiles average the players each manager used in a role. None of these predicted success when tested, so they describe the player and the job rather than forecast the outcome.
Physical
SkillCorner broadcast tracking. Season top speed is the average of a player's best five PSV-99 matches; volumes per 90 use matches of 45+ minutes. Compared with the target league's players in the same position.
This prototype
Reads the RJW database live when opened: the shortlist, style, role scores, physical data, destination-club model, manager profiles and each PL club's current first choice in every role. Projections use the Premier League as target. Ranges are widened up to 20% for strong players (last-season score above 0.5), as tested. Contribution percentiles are taken from each role's role-score measures (z within his league-season). Role templates, job focus measures and scouting questions are RJW's starting points, to be agreed with each club.