Juventus, summer 2026 — what we said before the season

Every projection here comes from a model built on data through the end of 2025/26 and frozen before the window opened. It is deliberately simple — a Marcel baseline — so that the real model has something honest to beat. Nothing on this page can be edited after the fact; the file hashes are at the bottom.

How Marcel works: the last three seasons weighted 2 / 1 / 1, regressed toward the positional league average as if 480 minutes of average play were added, then multiplied by an age curve (peak 26–27; 19-year-olds ×1.07, 30-year-olds ×0.93). It uses only three seasons by design — older seasons carry little signal for next year, and the year-to-year stability of npxG/90 (r ≈ 0.72) is what makes even this simple a projection work. The full recruitment model is trained on twelve seasons; this is its baseline.

Pool = every CF/SS arrival into Serie A, EPL, La Liga, Bundesliga or Ligue 1 with a fee between €25m and €60m in the summer 2026 window (Transfermarkt, pulled 28 Aug). Names joined to Marcel by last name + initial; the source row is shown on hover so a bad join is visible, not hidden.

The price-band table above compares what clubs paid. It says nothing about style, and the obvious objection is that Kolo Muani is not the same kind of player as Vlahović. So: every attacker with 900+ minutes in the big five in 2025/26, 494 of them, described by 23 per-90 metrics from FotMob (non-penalty xG, shots, box touches, total touches, xA, chances created, crosses, long balls, dribbles, duels, aerials, dispossessions, fouls won, pressing, recoveries, defensive actions), standardised and reduced by PCA. Distance in the full 23-metric space is the similarity measure; the two axes below are the first two components, for reading only. Source changed: this chart was previously built on FBref, which has since removed every xG and advanced statistic from the site — see the methodology page.

Season shown is 2025/26, complete. Built on FotMob, which unlike FBref still publishes non-penalty xG, box touches, aerials and dribbles; the two sources are never pooled. FotMob has no carrying or progressive-reception data, so the wide-carrier distinction the FBref space found is not reproducible here — the FotMob taxonomy resolves three profiles, not four. Not residualised for team context, so a striker at a possession-dominant club drifts right. Openda has no big-5 season over 900 minutes in 2025/26 and is not in the set; Alajbegović has no big-5 minutes at all.

• Not a grade. Grades need realised outcomes; the first arrives May 2027. This is the ex-ante half, sealed so it cannot move.
• Not the real model. Marcel is a 3-season weighted average with shrinkage and an age curve. It is the baseline the recruitment model must beat, and it is the only model that existed before the window.
• Marcel's own error is ±0.10 on npxG/90 at roughly 80% coverage. Read the chart's vertical positions with that in mind.
• Serie A matchday 1 had been played when this was sealed. No 2026/27 data is in any input; only the choice of what to register post-dates it.
• Wages are Capology estimates, gross fixed, bonuses excluded. Fee amortisation is straight-line over the contract as signed.
• The pool is 14 players. Small, and the name-join is fuzzy — every hover shows the row it matched to.