9 min · 2026-08-18
Why xG is not enough
xG is a shot model. It misses threatening possessions with no shot. It is not a verdict on the result.
New here? This page uses terms like xG, PPDA, and progressive passes. The free Foundations lesson explains them in a few minutes. Continue on the Learn path →
What xG is actually scoring
A shot from close range usually carries more xG than a hopeful shot from distance. xG estimates that difference from historical shots. The unit is the attempt. No shot means no xG. That is useful for chance quality. It is limited when pressure, territory, or progression never becomes an attempt.
A team can lead xG and still lose. That is not proof xG failed. It is proof one match has noise. Calling xG a deserved-to-win table adds a claim the model did not make.
Shot value is not action value
A pass into the box may raise threat before any shot exists. xG does not score that pass. xT, VAEP, OBV, and related models ask whether an action made the state more dangerous. That is a different question. It is also easy to overstate.
StarkPitch should not turn a homemade action-value number into a standard model. Keep the distinction clear. xG can describe the shot process well. It can still miss part of the match story.
Two open-data checks
Spain against Morocco in 2022 shows the first tension. Spain moved the ball forward often, but the xG Timeline stayed modest. Possession was real. Chance quality stayed limited. Quote both, or the story is incomplete.
Germany against Japan shows the second tension. Germany led xG, but Japan won with a low-xG goal included. That is finishing and match-state variance. It does not discard xG. It also does not erase Japan’s second-half shot burst.
Claims that survive
Safer: “They occupied territory and progressed often, but the shot process stayed modest.” Safer: “They led on xG and still lost; single-match conversion is noisy.” Riskier: “xG proves they deserved it.” Riskier: “possession value proves they were better.”
Use the Match Analysis xG Timeline and half splits for practice. When layers disagree, write down the disagreement.
Key points
- xG scores shots. It does not score every threat before a shot.
- Action-value models ask a different question. They are not upgraded xG or proof of superiority.
- High progression with flat xG is a useful warning sign.
- High xG with a loss shows single-match noise. Write process claims, not verdicts.