Model settings

These change how projected xP is calculated for every player. The two dials below are the ones worth touching — each shows what it changes and a live example. Everything technical is behind Advanced and is already calibrated. Edits apply on the next compute-xp run (they won't change numbers until then).

Fixture swingedited1.00

How much an easy or hard fixture moves a player's projected GOALS. At 0 the opponent is ignored; at 1 a weak defence gives the full boost.

0 · ignore fixtures1 · full swing

Now: vs a weak defence, goals swing up to ~25% either way.

When to change: Lower it if you think fixture difficulty is overrated; raise it to lean harder into good/bad matchups.

Trust in finishingedited0.80

Scales projected GOALS by how well a player finishes vs their xG. At 0 everyone is projected at their raw xG; higher lifts clinical finishers and trims wasteful ones.

0 · pure xG1 · fully trust

Now: Darwin (17G from 11xG) → about +18% projected goals.

When to change: Raise it if you trust proven finishers to keep out-scoring xG; lower it if you think finishing is mostly luck.

Advanced parameters (9)

These are calibrated from the 2025/26 backtest and rarely need changing. Only touch one if projections look systematically off and you understand the input.

Last-season carryover strength Weight pulling last-season rate toward the position prior		default 500
Finishing shrinkage (pseudo-xG) Pseudo-xG shrinking the finishing factor toward neutral		default 15
Baseline points floor Play-weighted baseline points floor (calibration)		default 0.5
League avg goals conceded / game League avg goals conceded per team per match		default 1.4
Penalty conversion rate Penalty conversion probability		default 0.79
Penalties awarded / team / game League penalties awarded per team per match		default 0.11
Rate confidence (pseudo-minutes) Bayesian shrinkage strength for per-90 rates		default 400
Assumed starter minutes Assumed minutes when starting		default 82
Assumed substitute minutes Assumed minutes when appearing as a substitute		default 22