IN DEFENCE OF ENDRICK

Why The Brazilian Wonder-Kid Was Right To Take That Shot

Its the fifth minute of injury time in Real Madrid’s opening Champions League group match against VfB Stuttgart.

Real lead 2–1. The Germans have a late free kick. Its the last play of the game. Stuttgart load the box and toss the ball into the mixer. The ball breaks to eighteen-year-old Brazilian star-boy Endrick on the edge of his own box and the kid from Sao Paulo is off and running.

As Endrick blasts up the field, Kylian Mbappe and Vinicius Jr smell blood. The Ballon d’Or nominees sprint to take up flanking positions. The three Real superstars are baring down on two desperately retreating Stuttgart defenders. Just behind the front three, Güler and Valverde are also in support — its 5v2.

Within seconds Endrick has arrived roughly thirty yards from the Stuttgart goal. He has Mbappe to his left and Vincius to his right. Mittelstadt and Rieder are doing their best to block Endrick’s straight line to goal. Mittlestadt is also poised to press Vinicius, Rieder is blocking the passing lane to Mbappe.

These fast-break scenarios with attacking overloads are trained over and over again at all levels of football. Coaches want to prime players to be able to exploit the various types of advantages afforded by the situation:

• numerical (3/4/5v2)
• qualitative (fast attackers are often better at beating defenders than the defenders are at stopping the attacker)
• dynamic (retreating defenders who are effectively running backwards while attempting to maintain engagement with advancing attackers while also tying to protect a large space are significantly handicapped in terms of possible actions)

The aim is to leverage these advantages to create a high probability scoring chance. The most common process is the for the ball-carrier to continue dribbling until one of the defenders is forced to engage them to stop a central shot from a dangerously close distance. When the defender stops retreating and engages, the ball carrier can slip a pass to the side for the flanking attacker to receive with a direct line to goal.

From here, the wide attacker can either shoot themselves or commit the goalkeeper/defender into a blocking action which frees up another teammate for an easy chance.

What is an ‘easy chance’? The difficulty of any chance in measured by its xG value. xG is a quantity determined by cross-referencing a suite of variables from relevant historical shot data such as:

• shot location (distance/angle)
• height of ball when shot is taken
• direction of assisting pass
• number of defenders between shot location and goal
• degree of pressure being applied to shot taker
• positioning of goalkeeper
• etc etc

Its not rocket science. The idea is that an attacking situation as favourable as a 5v2 should always end up with the generation of a shot which has a high percentage chance of being scored. Anything less would mean a bad decision has been made by a player somewhere during the passage of play.

Endrick didn’t wait until a defender engaged him. Endrick didn’t pass to Mbappe or Vinicius. Endrick took a shot from the thirty yards. According to SofaScore the shot had an xG value of 0.05. That’s a 1 in 20 chance of scoring. Not very high.

So, not only did Endrick stick two fingers up to any notions of galactico pecking order, he also said a big fuck you to the stats. Sure, Endrick scored, but minor details like that don’t stop criticism of his decision. ‘He got lucky this time’ the data-bods would say. The kid needs to cut that sort of thing out if he wants to make it at the highest level, his ‘decision making’ needs to improve.

Its inaccurate to blame the criticism of Endrick’s decision solely on the ‘data-nerds’. TNT Sports ‘analysts’ Joe Cole and Steve Mcmanaman’s (two guys who are hardly sticklers for statistics) post match commentary went like this:

COLE: He actually does everything wrong here

McMANAMAN: He does everything wrong doesn’t he!!??

(Laughter)

COLE: He takes the shot too early, he should’ve passed it….

McMANAMAN: It makes a mockery of football doesn’t it, because as Joe said, everything he did there was wrong. If you were his teammate you’d be shouting and screaming just to put Vinicius in onto his right but he ends up with the goal.

Here’s where it gets interesting. What if we flew in the face of the overwhelming consensus and took seriously the proposal that Endrick was actually correct to take that shot? How might we justify such a claim in the face of all rational, objective and statistical (and TNT’s) analysis?

For this claim to carry any weight we would have to tell a story about why the xG value of the shot was actually significantly higher then 0.05. Its true that different xG models will give different values. This is the basis on which these models compete in the market. Whose models come closest to reconstructing the reality of the event?

All ‘expected-anything’ models are just quantified interpretations of someone’s assessment of the difficulty of completing an action. Ultimately its the human data-scientist who decides when to stop including variables into the model’s calculations. So, what if we built a model that included a bunch of variables that the existing xG models had failed to consider?

The most obvious place to start would be to factor in how good Endrick is at shooting. Its well known that individual player quality will effect the xG value if its factored into the equation, and generally speaking its appreciated as relevant ‘context’ when evaluating such data.

Maybe we can argue that Endrick’s traction-engine left foot power-striking might raise our xG value a little. But lets be realistic, it isn't gonna be nearly enough to convince the more stringent of probability based decision judgers.

We have to go deeper. The main failing of xG models (and any predictive model of this nature) is that, because they need numerical values to work with, they cannot account for the infinitely complex array of non-numerical factors which are influencing the outcome of the event.

These non-numerical factors or qualia concern the subjective experiences of the people involved in the event rather than the objectively verifiable aspects as such as shot-distance or whether the shot was a header or a footer.

That’s all fine, but talking about subjective aspects like ‘confidence’ etc doesn’t really help us. We’re moving perilously close to citing vibes as a decisive factor.

We’re trying to build a serious model here and its not as if the professional model builders haven’t already thought about this stuff. Our proposal that Endrick was correct to shoot is still on decidedly shaky ground.

Lets look closer at the action. Lets really try to put ourselves in Endrick’s boots and, as much as is possible, feel what he was experiencing as his reality unfolded in the moments leading up to the shot.

The first thing we might notice is that Mittelstadt (#7 closest defender to Endrick) seems to be expecting the pass to Vinicius. This is probably because of all the training sessions he’s taken part in and that the direct passing line to Mbappe is blocked by Rieder. Not great positioning from Mbappe so Vinicius looks like the obvious choice.

Probably Mittelstadt feels that encouraging Endrick to shoot from thirty yards while maintaining a covering position on Vinicius is the lesser of two evils when compared to closing the shot and committing himself early.

This means that Mittelstadt is till retreating and not expecting the shot when Endrick pulls the trigger. Mittelstadt’s backwards motion makes it pretty much impossible to extend out a leg — or any other body part — to make a effective blocking action.

When the ball leaves Endrick’s left foot Mittelstadt is upright, twisting with both feet off the ground as his momentum carries him backwards. He is in no state to influence Endrick’s shot. But thats still kind of OK. The chances of the shot going in are, well, about 1 in 20.

At this point we need to assume an unusually high level of perceptual sensitivity on the part of Endrick. And its reasonable to do so based on past evidence. Earlier this year Endrick executed a quite remarkable piece of skill while playing for Palmeiras.

As the ball bounced off the turf at the edge of the opponent’s box, Endrick shapes to shoot on the half-volley with his left leg in a backswing motion. Two Botafogo defenders fully commit to blocking the half-volley. But rather than make the strike, Endrick uses his swinging left-knee to cushion the ball as it rises, perfectly setting it to the side for a left foot finish into the bottom corner.

Its a staggering piece of perceptual synthesis. Exactly the type of moment you watch football for. Endrick is able to account for mind-bogglingly complex array of dynamic variables. What factors might you have to account for when trying to construct an expected-success model for this piece of deceptive ball control?

• the bounce trajectory of the ball (spin, speed, direction etc)
• the movement of the two defenders
• the surface area of the knee required to contact the ball
• given that the knee has a relatively hard surface, what strength of contact is required to cushion the ball to the side to enable a shot with the next touch?
• maintaining his own balance
• etc etc etc

None of these are easy (or even possible) to quantify.

Another of Endrick’s actions in Brazil further helps us to see why he chose to shoot against Stuttgart.

In a 2023 Brasileirao game against America MG we are given a vital glimpse into the logic of Endrick’s shot-decision making process.

Endrick picks the ball up just outside the box and sets himself for a left foot shot. At the moment he takes the shot Endrick is immediately faced by three blocking defenders. There are also three further defenders between Endrick’s shot location and the goal.

Not only is Endrick able to calibrate a shot trajectory through the blocking bodies but, crucially, he seems able to account for the goalkeeper’s motion and perception (or lack of it) at the moment of the shot.

From behind the goal we can see how the Goalkeeper’s view of Endrick is almost entirely obscured by the blocking defenders. This means the goalkeeper cannot read the cues of Endrick’s body shape to anticipate the type/direction/power of the shot.

Not only is the goalkeeper unsighted at the moment of the shot. Crucially, they are also caught in suspended animation. Endrick shoots at precisely the moment when the goalkeeper makes a two-footed hop off the ground to set himself for when he expects the shot to come.

Because neither of his feet are in contact with the turf at the moment he needs to push off to dive, and coupled with lack of a clear sight-line, the goalkeeper is unable to get across the to Endrick’s powerful low shot.

Is it reasonable to suggest all of these high-resolution aspects somehow show up across Endrick’s perceptual field? I would say yes. In fact I would say that its precisely this type of perceptual sensitivity that makes great players great. Keeping perception and action tightly coupled. Adaptive behaviours emerging in concert with relevant real-time environmental information. This is what makes footballers more or less skilful.

Lets return to the Bernabaeu. As Endrick shoots, Mittelstadt is precisely in the goalkeeper Alexander Nübel’s eyeline, Nübel is unsighted. Like Mittelstadt and Rieder, Nübel was also retreating as Endrick advanced. The vertical angle replay shows Nübel back-peddling, and at the moment of impact he has just hopped back and landed two-footed with the weight on his right side.

Because of these bodily dynamics, Nübel’s first movement after the shot is hit is actually a small jump to his right — the opposite direction to where he needs to get to. This subtle but pivotal correction is required to get the purchase needed to make the dive to his left.

As usual Endrick’s shot is low and powerful. It bounces and skids of the greasy surface just before it reaches the flailing Nübel adding a further layer of complexity to an already nightmarish scenario. Nübel gets hands to the ball but can’t keep it out.

By constructing a model which allows for the bizarre perceptual capacities of Endrick’s footballing vision it seems at least feasible to argue that shooting was a good call. Maybe you can’t back it up with numbers and hard data, but those quantifiable aspects aren’t the only information sources available when analysing the game of football.

Is it really so outlandish to suggest that the world’s best players simply see things that we mortals don’t?

At a moment when trends in football’s tactical and coaching development are informed and influenced by data models more than ever, events like Endrick’s goal remind us that no model can ever fully capture reality’s infinite complexity.