Expected Goals
Event dataCore equation
What it sees
What it misses
Best use
Credit note
0.14Example value
ShotPrimary unit
MediumData demand
Interactive framework atlas
Every model answers a different question.
Tap a framework. The right panel updates the definition, formula, data required, blind spot, and a clean pitch graphic. Computational descriptors are separated into cards so nothing is bunched together.
Model lab
Watch value move as the game state changes.
Adjust the sliders. The pitch recalculates a simplified teaching model showing how separation, pressure, velocity, and zone danger reshape player value. The formulas are instructional, not a replacement for provider-specific proprietary models.
Dynamic possession value simulator
The same pass can be valuable, neutral, or harmful depending on where it starts, where it ends, how much pressure exists, and whether the receiver controls space.
xT teaching deltaxT(end) − xT(start) = +0.080
Net valueΔP(score) − ΔP(concede) = +0.056
Pitch controlcontrol(receiver zone) = 69%
Visual intelligence theater
See the hidden work that touch statistics miss.
Three live graphics separate off-ball gravity, pressing disruption, and value stacking. The motion is intentionally label-light so the shapes teach before the words explain.
Off-ball gravity
A forward can create value by pulling defenders away from the highest-value lane before the pass is played.
Decoy runLane opensReceiver gains space
Pressure field
Defensive value can appear as narrowed passing angles, delayed decisions, and lower future possession value.
Pressing shadowRisk risesOptions shrink
Value stack
A complete evaluation layers shot value, progression, receiving, defending, transition protection, and role context.
xGxTPitch controlRole fit
Position-agnostic evaluation
The objective stack is layered, not single-stat.
No public metric fully evaluates every player regardless of position. A serious evaluation stack combines goal-value math, spatial context, role adjustment, and uncertainty.
1
Net probability value
Convert actions into scoring-minus-conceding probability so attackers, midfielders, and defenders share a common unit.
2
Spatial influence
Use tracking, 360 context, pitch control, off-ball opportunity, receiving separation, and lane creation.
3
Role adjustment
Compare players against tactical peers using minutes, possession share, league strength, team strength, and job profile.
4
Reliability layer
Separate signal from noise with sample size, opponent strength, volatility, repeatability, and model confidence.
Clean interpretation rules
What great analysts refuse to ignore.
These rules prevent overreading touch-based statistics and protect the evaluation from misleading rankings.
Event data is powerful, not complete.
It captures discrete actions well, but basic versions struggle with decoy runs, pressing shadows, defensive shape, and space creation away from the ball.
Tracking data adds context, not magic.
Player location and velocity reveal pitch control and off-ball value, but every tracking model still depends on assumptions and calibration.
Role matters.
A center back, ball-winning midfielder, winger, and striker should be compared by shared value units and role-specific expectations.
Attribution timeline
The lineage behind the models.
The timeline gives credit without pretending that broad industry metrics have one universal inventor.
1990s+
Expected-goals foundations
Shot-quality modeling grew through academic and industry work over many years. Modern provider models vary by features and calibration.
2011
Possession-value predecessors
Sarah Rudd’s public Markov-chain work helped shape later zone-transition approaches to valuing possession.
2018
Expected Threat and Pitch Control
Karun Singh popularized the public xT framework. William Spearman presented influential pitch-control and off-ball opportunity work.
2019
VAEP and Possession Value
Decroos, Bransen, Van Haaren, and Davis formalized VAEP. Stats Perform launched Possession Value for short-window scoring probability.
2021
OBV, EPV, and richer spatial modeling
StatsBomb introduced On-Ball Value publicly. Fernández, Bornn, and Cervone published a soccer EPV framework for observed and potential actions.
Framework comparison
Use the right model for the right question.
This table is intentionally compact and scrollable on mobile rather than compressed into unreadable text.
| Framework | Question answered | Core math | Best for | Main blind spot |
|---|---|---|---|---|
| xG / npxG | How likely was the shot to become a goal? | P(goal | shot features) | Shot quality and finishing context | Non-shooting contribution |
| xA | What shot value did the pass create? | Σ xG from assisted shots | Chance creation | Passes not shot, off-ball movement |
| xT | How much did ball movement improve zone threat? | V(end) − V(start) | Progression and territory value | Basic grids miss pressure and velocity |
| VAEP / OBV / PV | How did an action change scoring and conceding probability? | ΔP(score) − ΔP(concede) | On-ball action value | Depends on event context and model design |
| g+ | How did the touch affect scoring across current and next possession? | current + next possession value | MLS-style public player contribution | On-ball framework; tracking value still separate |
| Pitch Control | Who controls space at a coordinate? | P(team controls x,y,t) | Receiving lanes and defensive shape | Requires tracking and assumptions |
| EPV | What is possession worth at an instant? | E(next goal outcome | state) | Observed and potential actions | Complex and data-heavy |
| Packing | How many opponents were bypassed? | Σ bypassed opponents | Line-breaking passes and carries | Raw counts need threat weighting |
| Possession-adjusted defense | How much defending happened per opportunity? | events × 50 / opp poss% | Cross-team defensive comparisons | Opportunity-adjusted, not quality-adjusted |