Real Oviedo
Dashboard Guide

xG — Expected Goals

The sum of shot probabilities for all shots taken, based on historical conversion rates for shots from similar positions, angles, and situations. Measures chance quality and volume combined.

npxG — Non-Penalty xG

Expected Goals excluding penalty kicks. The cleanest measure of open-play attacking threat, particularly relevant for comparing forwards across teams with different penalty records.

xA — Expected Assists

The xG value of the chances created by a player's passes. A high xA means the player created high-quality opportunities regardless of whether the receiver converted them — isolating creative output from finishing quality.

Goals − xG

The difference between actual goals scored and xG. Positive = clinical finisher performing above expectation; negative = underperformance. Persistent negatives typically revert toward zero over time.

xG Chain

The sum of xG from all attacking sequences in which the player was involved at any point — shots, assists, build-up passes. A holistic measure of attacking contribution beyond just shots and assists.

xG Buildup

xG Chain minus the player's direct shot xG and assist xG. Isolates pure build-up and progressive passing contribution — value added in sequences before the final actions. Key for evaluating deep-lying forwards and ball-playing midfielders.

xG/Shot — Shot Quality

Average expected goal value per shot. A higher value means the player consistently attempts shots from better positions or with better technique. Elite forwards in La Liga typically maintain xG/Shot above 0.12.

G/Shot — Conversion Rate

Goals divided by total shots attempted. A direct measure of finishing efficiency independent of shot quality. Compared against xG/Shot to determine whether a player finishes better or worse than their chances warrant.

SoT% — Shots on Target %

Percentage of shots that were on target (requiring a save or resulting in a goal). Low SoT% indicates poor execution or decision-making; very high SoT% alongside low conversion may indicate strong goalkeeper opposition.

Key Passes

Passes that directly lead to a shot attempt by a teammate. Unlike xA, key passes count all shots generated regardless of quality. High key passes + high xA = a player creating both volume and quality chances.

Per 90 Normalisation

Dividing any counting stat by the number of 90-minute intervals played, enabling fair comparison between players with different amounts of playing time. Minimum 900 minutes is typically applied to filter small samples.

Scout Score (Composite)

A 0–100 index calculated as a weighted average of within-population percentile ranks: npxG/90 × 0.40 + xA/90 × 0.30 + Shots/90 × 0.15 + xG/Shot × 0.15. Designed to reward both clinical finishing and chance creation.

Within-League Percentile

A player's rank expressed as a fraction of the league's player population in the same metric. A 90th percentile npxG/90 means the player outperforms 90% of comparable players in that competition.

TklW+Int/90

Tackles Won plus Interceptions per 90 minutes. The primary defensive activity indicator. A high value signals a player who both wins individual duels and reads the game well enough to intercept passes before the duel occurs.

Defensive Score (Def Score)

A 0–100 composite index for defenders: TklW+Int/90 × 0.35 + Key Passes/90 × 0.25 + Passes into Final Third/90 × 0.20 + Discipline (fouls/90 inverted) × 0.20. Rewards defensive activity, passing, and clean play simultaneously.

PPDA — Passes Allowed per Defensive Action

The number of opposition passes allowed per defensive action (tackle, interception, foul) in the opposition half. Lower PPDA = higher pressing intensity. A team allowing 5 PPDA presses far more aggressively than one allowing 10.

Poisson Model

A statistical model that treats goals scored by each team as independent random events following a Poisson distribution, each with its own expected rate (λ). H/D/A probabilities are derived by convolving the two distributions over a 7×7 scoreline grid — summing the probability of all home-win scorelines, all draw scorelines, and all away-win scorelines.

Walk-Forward Validation

A leakage-free evaluation protocol where predictions are always generated using only data available before the match date. The model is re-fitted incrementally after each matchday — no future information ever leaks into earlier predictions. This ensures reported accuracy metrics reflect genuine out-of-sample performance.

Prior Blending

A technique that smoothly transitions a team's attack/defence rating from a prior estimate toward observed match data as more games are played. A tuned shrinkage constant controls how fast ratings update: a small constant means ratings converge to the data quickly; a larger constant keeps them anchored to the prior for longer. Constants are tuned independently for strength, home advantage, and other model components.

Confidence Score

The maximum probability across the three outcomes: max(P_H, P_D, P_A). Values below 42% are flagged as low confidence — the model sees a genuinely open contest and its prediction carries less informational value. Not a measure of model quality, only of outcome decisiveness.

xG Gap

Expected goals for the home team minus expected goals for the away team. Positive values favour the home side; negative values favour the away side. An independent directional signal that can confirm or contradict the probability output.

Style Label

One of four KMeans cluster assignments derived from six match-level features aggregated over the season: Possession/Dominant, High Pressing, Defensive/Low Block, Direct/Counter-Attack. Used in Match Detail and Prediction History to provide tactical context around model predictions.

Promoted Prior

Newly promoted teams start the season with ATT/DEF priors derived from the previous season's relegation zone, rather than the league-wide average. This reflects the expectation that promoted sides are typically below the division mean at the start of the campaign. Flagged with 🟡 in the predictions table.

LogLoss

The primary model evaluation metric. Measures the average negative log-probability assigned to the correct outcome; lower is better. A model assigning 70% to the correct outcome scores −log(0.70) ≈ 0.36 for that match. The production model achieves LogLoss = 0.9717 over the full walk-forward evaluation set.

U25 Value Target

Players under 25 years old who rank in the top 30th percentile within their league by composite score. Represents optimal value — sufficient current output to justify transfer spend, with meaningful development potential still ahead.

90s Played

Total minutes played divided by 90. Used as a sample size filter — players with fewer than 5–10 ninety-minute intervals are typically excluded or flagged as small-sample results. All per-90 metrics should be read alongside this figure.