Project Underdog · October 2026

Football beliefs.
Tested.

Twenty things coaches say. Three held up. The rest looked like tactics until the data asked a harder question: was the moment already different before the decision?

The short version

Most football advice describes good moments. It does not create them.

Project Underdog tested common coaching beliefs against 3.6 million real on-ball decisions and 30 matches with player tracking. The result is not a new tactical system. It is a smaller, more useful list of things worth trying.

20beliefs tested
3.6Mon-ball decisions
30tracked matches
3mechanisms that held
The constraint: every idea had to repeat across men's international, men's club, and women's football, survive like-for-like comparisons, and pass a check for what the moment looked like before the decision.
What survived

Three small edges were more durable than the big claims.

One more where the ball lands

On long balls, the useful number is not the target player. It is the players around the landing spot. Teams kept the ball 84-89% of the time when they had one more player there, versus 63-72% when numbers were level.

Set pieces narrow the quality gap

Team quality mattered less at set pieces than in open play. In mismatches, underdogs won 37-38% of set-piece chances compared with about 33% of open-play chances.

Use one side to open the other

Keep the ball on one side and the defensive block slides. Tracking data showed the far-side winger gained about 1.3 metres. The space is real; turning it into chances is still the coach's job.

What failed

The scoreboard is mostly a graveyard of confident advice.

Sixteen popular beliefs showed no reliable edge in the tested data. That does not mean the ideas are impossible. It means the evidence did not justify treating them as rules.

"Shoot more"The extra shots were worth about +0.3 goals per 100. Most of the apparent gain existed before the shot.
Press on classic triggersBack passes and bad touches did not make pressing pay more once the situation was described properly.
Attack the weak sideWeaknesses did not repeat reliably from match to match, so targeting them created no dependable edge.
More runners and faster playOnce defenders and context were counted, extra runners and urgency added no measurable benefit.
The honest limit

Observational data can kill a myth. It cannot finish the experiment.

The paper is built from open and public datasets: StatsBomb, Wyscout, Metrica, SkillCorner, DFL tracking data, and football odds. It is a serious attempt to separate decisions from the moments that produce them, but it is not a controlled trial. A new attacking tactic still has to survive on a pitch.

That is also why the paper tests AI tactics tools with caution. Simulators produced tactics with apparent 95% win rates, but the winning tactics were bugs. In the study, simulators could overreact by 10-50 times. A model can make a football answer look precise without making it true.

What a coach could use on Monday

Give long-ball targets one extra teammate at the landing spot. Treat set pieces as a real underdog lever. Use possession on one side to move the block before switching. And before trusting any rule, ask the question that removed most of the folklore: was the moment already better before the decision?

Read the full 8-page whitepaper →