Skip to main content
Sensible Stats / Established research

Completion is a modest ambition

A pass can reach its destination without improving anyone’s afternoon. A foundational action-value paper helps explain what completion leaves out—and how a coach might review the next decision.

By Elias Vale · Published · 6 min read

An ivory football rests between a broad teal forward sweep and a curling vermilion return ribbon, balanced among navy semicircles on textured warm paper.
A pass and its competing consequences. AI-generated conceptual editorial illustration by Sensible Stats; not a match reconstruction, measured route or model output.
In this article

Liverpool’s attempt to build from a goal-kick helped Arsenal build the winning goal instead. In the official account of the 1 November game at Anfield, Arsenal recover possession, Declan Rice overlaps Noni Madueke down the right and his low delivery finds Kai Havertz. The finish arrives in the 16th minute. The match ends 0–1. A restart is a small administrative ceremony; this one changed departments rather abruptly.

That report is our football reference, not footage we have independently reviewed or a complete action dataset. It raises a useful coaching decision: how should a build-up be assessed before the next instruction is to keep playing out or send everything long? Pass completion is part of the answer. Treating it as the whole answer is how a statistic acquires responsibilities above its pay grade.

References:[2] Liverpool FC: Liverpool narrowly beaten by Arsenal at Anfield — official match report

Put something on both sides of the ledger

In their 2019 paper, Decroos, Bransen, Van Haaren and Davis introduce VAEP: Valuing Actions by Estimating Probabilities. It values recorded on-ball actions through changes in scoring and conceding probabilities. Its implementation looks ten actions ahead, using the latest three actions plus game context. Increased scoring probability adds value; increased conceding probability subtracts it.

The distinction is useful even before anyone opens a laptop. Imagine a completed pass that gives its receiver room to turn. Now imagine one that arrives at the same patch of grass with an opponent already closing it. The completion column can give both a tick. A coach still has to decide whether the team has acquired an opportunity or merely transferred an emergency.

These are hypothetical situations, not descriptions of an unobserved Liverpool pass. Nor does the framework’s name give an event record a panoramic view of the press. A number can organise a question while still lacking part of the scene.

References:[1] KDD 2019 · ACM conference proceedings

A probability is not a certificate of good judgement

The study covers 11,565 league matches across seven European top divisions from 2012/13 to 2017/18. Training through 2015/16 precedes evaluation in 2016/17; training through 2016/17 precedes evaluation in 2017/18. Table 3 reports scoring Brier errors of 0.01376 with all features versus 0.01562 using location alone; conceding errors are 0.00547 versus 0.00560. Brier error squares the gap between a forecast probability and the eventual yes-or-no outcome; lower is better.

Those are probability benchmarks, not measured coaching effects. They support taking context seriously, without establishing a final exchange rate between every pass and every footballer. The authors note the absence of objective ground-truth action values, omitted off-ball contributions and difficulties comparing players across teams and leagues.

For a staff deciding how to restart, the distinction is practical. A rating of what happened is not automatically a ranking of every option that could have happened. An apparently available teammate may have been covered outside the camera frame. A fine-looking alternative on pause may require a pass nobody could realistically deliver. Retrospective confidence has exceptionally good positioning.

References:[1] KDD 2019 · ACM conference proceedings

The goal cannot review the whole policy

At Anfield, the reported recovery, overlap and finish make the build-up question worth asking. They do not establish that Liverpool should abandon the approach. We have neither a complete ledger of their restarts nor a VAEP score for the sequence. A goal is a clear outcome. It is a very small committee for deciding a policy.

The conversation point is to ask what the first action leaves the next player able to do. Then ask what danger remains if possession changes hands. Those questions help distinguish an instruction to be brave from an instruction that explains where the help is meant to come from.

The opponent belongs in that discussion. Arsenal’s recovery and Rice’s supporting run are part of the official account, not incidental scenery around a Liverpool mistake. But the account cannot supply every starting position, passing lane or rejected choice. Any explanation that needs those details must wait for usable evidence rather than decorate the gap.

References:[2] Liverpool FC: Liverpool narrowly beaten by Arsenal at Anfield — official match report[1] KDD 2019 · ACM conference proceedings

More context is still a research problem

This is an established paper, not a newly arrived discovery. Research continues to add context: Tamimi and Tran’s 16 October 2026 journal article describes combining event data, StatsBomb 360 freeze-frames and possession sequences in an opponent-aware reinforcement-learning framework. Only the publisher’s abstract is accessible here. It identifies a research direction; it does not let us independently endorse the reported gains or prescribe a Premier League tactic.

Our earlier pitch-control review asked what reachable space enables. Tuesday’s City–Brighton study separated an intervention from the opportunity that follows. Action valuation adds another question to that growing notebook: what changed for each side? The connection guides our interpretation. It does not turn the Anfield report into a test of any of these models.

References:[3] Tamimi and Tran (2026): Support-aware dual-agent reinforcement learning for soccer possession and action valuation with spatial–temporal modeling — publisher abstract only[4] Sensible Stats: Useful space and the time it takes to reach it — Research Review[5] Sensible Stats: The save is not the end of the shift — Weekend Study

Something to change in the review room

For a coach and analyst, here is our untested staff-review comparison. Before choosing the next build-up exercise, take a consecutive block of goal-kick build-ups from authorised full footage. Keep the ordinary exits, failed attempts and opposing recoveries. A highlights reel cannot supply that comparison because its editor has already selected the interesting endings.

First, record the first-pass outcome alongside the receiver’s next visible options, pressure and how the phase ends; mark obscured information unknown. Next, review the same phases twice: once using your usual completion-led summary, then with those contextual notes. Compare which situations each review would prioritise. Finally, choose one specific restart-and-support situation to rehearse, and record the cue, instruction and subsequent outcomes when you revisit it.

Use that comparison to decide what needs practice, not to invent decimal action values from handwritten notes. If the usual review already includes this context, retain it as the baseline. Keep opponent and score attached, and retain cases that contradict the initial diagnosis. This routine has not been tested as a coaching improvement; its immediate payoff is a more explicit decision about what to work on.

A completed pass deserves its tick. The next player deserves something useful to do with it.

References:[1] KDD 2019 · ACM conference proceedings[2] Liverpool FC: Liverpool narrowly beaten by Arsenal at Anfield — official match report

Tags:VAEPAction valuationPassingLiverpoolArsenalBuild-up play

Pass it on

Know someone who enjoys the football behind the numbers? Send them this piece.

From research to forecasts

These findings are not automatic inputs to our model. Explore the accuracy record and methodology or the Premier League forecasts.

Found a mistake? Send a correction.

Established research · · 5 min read

The clock on the bench

Noah Okafor needed only a few minutes to decide Leeds–Tottenham. A study of Championship substitutes explains why a coach might start the clock again when a player comes on.

Source: PLOS ONE · 14(1), e0211563 · Peer reviewed

Read the article →