By XG Mind AI (xgmind.com)
A side grinds out a 1–0 victory, posts 0.7 xG to the opponent's 1.5 — and the comments immediately fill with "they got lucky." That instinct is useful. It just isn't finished thinking yet.
Maybe the opponent generated most of its xG after going behind and chasing the game. Maybe a penalty explains the whole gap. Maybe the two numbers came from providers that see the same shots very differently. The scoreline and the chance model are answering different questions, and an xG audit should surface those questions — not hand you a verdict.
Start by identifying the measurement
Expected goals assigns a scoring probability to a shot that was recorded. Typical features include shot location, angle, body part and the action before the shot. Some models also factor in goalkeeper and defender positions.
Hudl StatsBomb's xG explainer gives a telling example: a basic model might rate a shot at 0.30, while a richer model that knows the keeper is out of position rates the same shot at 0.65.
When two totals disagree, resist the urge to average them into manufactured agreement. First establish whether both providers counted the same shots and used the same definitions.
There's another boundary to respect: xG only covers moves that became shots. A missed cutback, a last-ditch tackle before the shot — those can tell a tactically important story without adding a single decimal to conventional shot-based xG.
Use a small audit table, not a verdict
The following is a fictional five-match example. The numbers illustrate interpretation; they are not any club's real results or an XG Mind performance claim.
Team A: goals scored 10, xG for 6.0, xG against 7.0, xG difference −1.0. Team B: goals scored 4, xG for 7.0, xG against 5.5, xG difference +1.5.
A scoreboard-only reading favours Team A's attack. The xG totals raise different questions: how did Team A convert those opportunities into ten goals, and why did Team B convert so few?
Finishing quality, goalkeeping, event-model limitations and plain variation are all on the table — and these totals alone can't separate them. Read the table as a prompt to investigate, not as a forecast that Team B is about to improve.
Regression is not a deadline
Regression to the mean is often sold like a debt collector: a team has scored "too many" goals and now must pay them back. Probability doesn't work that way.
An unusually hot stretch can contain both genuine signal and positive noise. When the observation was picked precisely because it was extreme, later observations tend to be less extreme. That does not mean the next match reverses the previous one.
There is no universal six-match threshold, no law that 135% of xG guarantees an imminent collapse, and no guarantee that a goals-to-xG ratio converges to exactly one.
Three checks before you interpret a gap
Penalties. Keep total xG and non-penalty xG separate when it matters. Don't mechanically subtract some universal constant. And winning penalties can reflect attacking behaviour — they're not automatically "unearned."
Game state. A team protecting a lead will happily concede territory and shots. Check when the chances happened, look for red cards and score changes.
Opposition and sample. Five fixtures against weak opponents are not interchangeable with five against strong ones. Competition, home/away split and time window all matter.
These are checks, not automatic adjustments. If you claim to have corrected for game state or opposition, explain the procedure instead of just placing "adjusted" in front of the metric.
What post-shot xG adds
Post-shot models incorporate information about the shot after it was struck — placement, sometimes velocity. They're often used to study goalkeeping. But the names PSxG and xGOT don't guarantee identical definitions across providers, so don't assume compatibility.
One spectacular match is worth describing as one spectacular match. It doesn't establish an inevitable reversal next week.
Keep the forecast and the audit apart
An audit after full-time has information that didn't exist before kickoff. It can explain recorded chance creation, but it can't retrospectively validate a forecast — unless the original forecast and its inputs were preserved.
The practical routine is short: identify the provider, inspect penalties and game state, check the fixture window, and read the shot pattern. Then ask whether the written conclusion is stronger than the evidence allows.
That's also why we keep our workings visible. Our methodology documents how our forecasts are produced, the performance ledger records outcomes publicly, and our missing-data guide covers what to do when a comparison lacks one side's metrics. An audit you can't reproduce is just an opinion with a spreadsheet attached.
Top comments (0)