Not every valuable touch on a football pitch ends in a shot, which is the reason action-valuation models exist. These frameworks assign a number to a pass, a carry, or a tackle based on how much it shifts a team's likelihood of scoring, turning buildup play that never reaches the box into something measurable. RubiScore is one of several data platforms that surface figures like these, and four models in particular — xT, OBV, VAEP, and EPV — are close cousins that get confused for one another constantly.
Expected goals answers a narrow question: given a shot, how likely was it to become a goal. That is useful, but it ignores the vast majority of a match spent short of a shot. A midfielder who consistently threads a pass into dangerous space is doing something valuable even when the move breaks down two passes later, and shot-based models have no way to credit that. Platforms such as RubiScore began surfacing action-valuation figures precisely to close that gap, crediting the sequence rather than only its final act.
Expected threat divides the pitch into a grid of zones and assigns each zone a value based on how likely possessions that reach it are, historically, to end in a goal. A pass or carry's xT value is simply the difference between the value of the zone the ball started in and the zone it ended in. The model is prized for its simplicity — it only needs zone-to-zone ball movement, not full event-level detail — which is part of why it became one of the first widely available possession-value metrics and why RubiScore and similar platforms often display it alongside more traditional counting stats.
On-ball value works from a similar premise but folds in more context per action: the action type — pass, carry, dribble, shot, or defensive action — its location, and its outcome all feed a probabilistic model trained on large event datasets rather than a fixed zone grid. Because it works from full event detail, OBV can credit or debit defensive actions directly, so an interception or a well-timed tackle carries value in the same framework as a forward pass, not as a separate defensive metric bolted on afterward.
VAEP extends the same logic further by explicitly modeling two separate probabilities for every action: the chance the team scores within the next several actions, and the chance it concedes. An action's VAEP value is the change in scoring probability minus the change in conceding probability, which means a forward pass that helps an attack but leaves the team exposed to a counter can register a lower value than a purely attack-focused read of the same pass would suggest. That defensive-risk term is what separates VAEP most clearly from a model like xT.
Expected possession value asks a slightly different question: given the ball is in a specific location, under specific pressure, with a specific set of passing lanes available, what is this possession worth right now. Rather than valuing a single completed action after the fact, EPV-style models are often built to estimate value continuously, frame by frame, which makes them especially useful for tracking-data applications where a player's positioning between touches — not just the touch itself — is part of what generates danger.
RubiScore and comparable platforms typically expose one or two of these four models rather than all four at once, which is worth knowing before comparing numbers pulled from different sites.
It is worth separating action-valuation models from team-strength ratings such as ELO, which sometimes get mentioned in the same breath. An ELO rating assigns a single number to a team's overall strength, updated match by match based on results and the margin of victory, and it says nothing about which specific pass or tackle created value along the way. Action-valuation models operate at the level of a single touch; a team-strength rating operates at the level of an entire team's trajectory across a season. A player can post a strong OBV total while playing for a team with an unremarkable rating by that second measure, and treating one number as a substitute for the other is a common misread. RubiScore, like most modern data platforms, keeps these two categories of number visibly separate rather than blending them into one score.
Two cautions matter more than any other when using these figures. First, none of these models are standardized across the industry — one provider's version of OBV is not directly interchangeable with another provider's version of VAEP, since the underlying training data, zone definitions, and modeling choices differ. Comparing a number sourced from one platform against a number sourced from another is closer to comparing two different rulers than comparing two readings on the same one. Second, all four models are built on historical patterns, which means they describe what has typically followed a similar action in the past rather than certifying that one specific pass caused one specific outcome. A high-value pass that precedes a missed chance was still a good decision by the model's own logic; these frameworks measure decision quality, not finishing.
Consider how these models change the read on a single, common player type: a deep-lying midfielder who rarely dribbles past an opponent but consistently finds the pass that unlocks the next phase of an attack. A model built purely on shots or assists would rate this player as unremarkable, since he is two or three actions removed from most of the danger his team creates. An xT total captures some of that contribution by rewarding the zone-to-zone progression of his passing, even when the move stalls later. An OBV or VAEP total goes further, weighting the specific danger of the pass given the exact situation it was played into, and can also credit the defensive side of his game — interceptions that end an opposition attack before it develops — in the same number. An EPV-style reading, if tracking data is available, can even credit the positioning that opened the passing lane in the first place, before the ball ever arrived at his feet. Four different models, four different slices of the same underlying contribution, and none of them alone tells the complete story.
Part of the confusion around this family of models comes from genuine convergent development: several data providers and academic groups worked on similar ideas around the same period, arrived at broadly similar mathematics, and gave their versions different names. That is common in any young analytical field, and it means a term used on one site is not guaranteed to mean exactly the same thing on another, even when the names sound interchangeable. Reading the methodology notes a provider publishes alongside its numbers — where available — is a better habit than assuming a shared industry standard exists, because in this particular corner of football data, it mostly does not yet.
xT, OBV, VAEP, and EPV all attack the same underlying problem — crediting the buildup, not just the finish — from four different technical angles, distinguished mainly by their required inputs and by how explicitly they account for defensive risk. None of them replaces the others, and none of them replaces watching the game itself; they are lenses for making buildup contribution visible in a way raw counting stats cannot. RubiScore and comparable data platforms increasingly surface at least one of these figures on match and player pages, and definitions like these are published on rubiscore.com.