Scoreboard Data vs AI Match Analysis

Two different questions: "what happened" vs "what does the data point to"

Live-score and statistics sites are the foundation of following football. They show a match's score, lineups, possession, shot counts, even xG, accurately and in real time. These sites work like a data scoreboard: they present what is happening quickly, neatly and reliably.

11Stat operates on a different layer. It takes the same kind of raw data (form, goal history, odds) as input, but its output is not a table — it is a probability distribution assigned to every outcome. The question changes: a stats site answers "how many goals did this team score in the last five matches"; 11Stat tries to answer "given this form, this data quality and this model agreement, what does the match's probability structure look like".

These two approaches are not rivals but complementary. Good analysis is fed by a good data base. 11Stat's job is not to replace stats sites but to place a measurable reading of probability and uncertainty on top of that data.

Raw statistics and modeled probability are not the same

A statistic is a backward-looking record: it summarises what happened. A probability is a forward-looking estimate: it states how likely something that has not happened yet is. Confusing the two is a common mistake.

For example, "the home side won its last six home games" is a strong statistic, but on its own it is not a probability. Weighing that run against opponent strength, squad absences, rest days and the market odds, then converting it into a calibrated probability, is the model's job. When the model says "70% probability", it also tracks whether similar past situations really resolved that way about 70% of the time — this is called calibration.

The layer 11Stat adds: model agreement, data quality, risk

A stats site can give you xG, but it will not tell you how different models interpret that xG or how strongly they agree. This is exactly where 11Stat's added value begins:

• Model agreement: do independent engines, Poisson/Dixon-Coles, Elo/form and Monte Carlo, point the same way? A read like "3/3 engines aligned" states plainly how consistent the signal is.
• Data quality: when data is thin (a new season, lower divisions, a sparse fixture list), 11Stat lowers its confidence score and deliberately suppresses inflated signals. A raw stats table usually gives no such warning.
• Risk level: every read is labelled Low / Medium / High; this guarantees nothing, it only communicates uncertainty honestly.
• Market sanity: a check on whether the model's probability lines up with a reasonable odds range.

For detailed definitions of these layers, see the match analysis and how it works pages.

An xG example: same number, different reading

xG (expected goals) appears both on stats sites and in 11Stat, but how it is used differs. On a stats site, xG is a historical measure summarising the quality of chances created in a finished match: "this team generated 1.8 xG but scored 0".

In 11Stat, xG is an input. The models turn each team's expected goal volume into a score-probability matrix, and from that single matrix the 1X2, over/under and both-teams-to-score (BTTS) probabilities are derived at once. As a result, xG and the result probability do not contradict each other; they form a consistent picture. Moreover, 11Stat does not limit this read to a single match: it evaluates it together with multiple form windows (5/10/20 matches) and home/away splits.

In short: a stats site shows you xG; model analysis converts xG, together with other signals, into a probability and flags how reliable that probability is.

How to use both together

The healthiest approach is to layer the two sources rather than substitute one for the other:

1. A stats site for raw facts. For the score, lineups, injuries, live match data and form runs, live-score sites are fast and reliable.
2. Model analysis for the probability frame. For how those facts combine to shape the match's probability structure, how much the models agree and how reliable the data is, read 11Stat.
3. Take uncertainty seriously. High risk or low model agreement tells you to read even strong-looking statistics from either source with caution.
4. Treat no output as a guarantee. Neither raw statistics nor a probability model guarantees a result; football is a high-variance sport.

For daily multi-match summaries use the bulletin page, and for term definitions the glossary. 11Stat accepts no bets, sells no coupons and guarantees no result; every output is a probability model, simulation or historical backtest, and is not betting advice.