Custom scouting · percentile fit

The Forge

Build the player you're looking for. Pick up to 10 metrics, choose per 90 or season totals, and get the top 50 ranked by fit — with the percentile behind every column, computed live against your pool, never a pre-baked table.

3,000+ players Big-5 + challenger leagues, 2025/26 10 leagues, one ranking universe at a time Understat × FotMob × Sofascore

The pool

Every percentile below is computed live against exactly this pool, never a fixed leaderboard.

Player pool

Season

League

Club context

Contract

Position pool

Minimum minutes

900

Max age

Any

Percentiles are computed vs your pool — Big-5 and challenger universes are never mixed in one ranking.

The metrics

Pick up to 10 metrics from the categories your pool actually carries — every one counts equally. Set a bar on any of them to list only the players who clear it. Switch units to rank on season totals instead of per-90 rates. Start from a preset and tune it, or build from scratch below.

Presets

Defence

Midfield

Attack

Goalkeeper

Units

The candidates

The top 50 of your pool, ranked by Fit — the percentile behind every metric column, the cross-league Statz Score, and each player's market value and contract expiry alongside. Click any header to re-sort — by cost or expiry to read the shortlist as a recruiter would.

Rows

Cells

Frontier

The candidates

Pick metrics to forge your candidates

How it's computed

Fit. For each metric you pick (up to 10) a player earns a live percentile against your filtered pool, then fit = Σpercentile / n — every metric counts equally, giving a straight 0–100 score, not a single-metric maximum and not a market value.

Per 90 or season totals. A per-90 rate says how good a player is per minute played; a season total says how much he actually delivered. Switching to Raw re-ranks on totals, so a 3,000-minute season beats a brilliant 900-minute cameo — availability is part of the profile. Only per-90 counts convert: a percentage, a ratio or an already-season figure has no total form and reads the same either way, and its column says so. One caveat: for lower-is-better metrics — dispossessions, dribbled past, goals conceded — a season total rises with minutes, so inverting it rewards playing less. Those columns read against the availability story, not with it.

Thresholds filter, they don't score. A bar on a metric decides who is listed; fit still ranks whoever survives. Two consequences worth knowing. The bar always reads the metric's own per-90 value, never the season total — so switching units re-ranks the survivors but never silently reinterprets a number you typed. And percentiles are still computed against your whole filtered pool, not against the survivors: if six players clear your bars, their percentiles still say where they sit among every qualifier in your pool, which is the only reading that means anything.

A missing metric excludes. This is the one place the Forge is stricter than elsewhere. When you score, a player missing a metric is judged on the rest. When you set a bar, a player who doesn't carry that metric can't be shown to clear it, so he drops out — a winger has no gc90, and a goals-conceded bar will remove every winger.

Live, not pre-baked. The card pages show one percentile, computed once against the whole qualifying pool for the season. The Forge recomputes percentiles in your browser every time you change league, position or the minutes floor — narrow the pool to one league and the ranking shifts, because the players it's measured against shifted with it. Ties share a midrank¹, the same convention the pipeline itself uses.

Lower-is-better metrics are inverted. Metrics where less is better — dispossessions, goals conceded per 90, errors leading to a shot or a goal — are flipped (100 minus the percentile) before they're averaged, so picking one of these always rewards doing it less, never more.

A missing metric isn't a zero. Not every player carries every metric — a winger has no gc90, plenty of outfielders carry no shot-stopping entry. A player missing a picked metric is scored on the picks he does have; the average renormalizes over what's present rather than padding the gap with a zero or a league average. Each metric column shows its own pool size (n), because a metric only a few dozen players carry is noisier than one every qualifier has.

Pools never mix. Big-5 and the challenger leagues are ranked separately, always — switching pools reloads a different season file and rebuilds the picker from the metrics that file actually carries (the challenger pick list is a strict subset of the Big-5 one; picks that don't survive the switch show struck-through, not silently dropped). The one number that is comparable across both pools is the Statz Score column — the site's cross-league rating (rating_adj), re-centred so a Big-5 and a challenger-league player read on the same scale.

The contract filter is expiring deals, not free agency. Contract end dates come from FotMob's player pages as of the latest refresh — club-facing option years and unannounced extensions are invisible to it. "≤ 1 yr left" keeps players whose deal ends by July 1 of next year. A player with no contract data on file cannot be shown to expire, so an active filter excludes him and the results line says how many were dropped that way. True free agents hold no club and play no qualifying minutes, so they cannot appear in a minutes-gated pool at all. Contract and market-value data covers the current 2025/26 season only — on the 2023/24 and 2024/25 pools those two columns, the contract filter and the value frontier have nothing to read.

Small pools are flagged. Filter down far enough — one league, a high minutes floor, a thin position group — and a pool can shrink to a handful of players. Any pool under 20 qualifiers carries a visible noise warning: the percentiles are still computed the same honest way, but with that few players a single outlier can swing the whole ranking.

¹ Live percentiles use half-up rounding in the browser; the pipeline's stored card percentiles use Python's banker's rounding. The two can disagree by at most one point for the same player and pool — a documented, tested bound, not a bug.

The Forge

01 The pool

02 The metrics

03 The candidates

The candidates

Fit vs market value

The pool

The metrics

The candidates