Real Madrid–Barcelona: a 3,000-match experiment with open records
Two fixed lineups, three venues, 3,000 simulations. Explore the results, exact squads, uncertainty and downloadable match records from our Clásico experiment.
What happens when the same Real Madrid and Barcelona squads play at three different venues? We ran DreamerBuild's match engine 3,000 times: 1,000 with Madrid at home, 1,000 on neutral ground and 1,000 with Barcelona at home. This article includes the starting XIs, the settings and a downloadable record of every score.
The results: 3,000 matches, three venues
Each card shows 1,000 matches. The percentage next to each team is the share of matches that team won; draws belong to neither team.
Real Madrid at home
1,000 simulated matches · Same squads and tactics
- Real Madrid won
- 44.3%
- Draw
- 24.1%
- Barcelona won
- 31.6%
443 out of 1,000 matches
241 out of 1,000 matches
316 out of 1,000 matches
For example: Real Madrid won 443 matches here, Barcelona won 316, and 241 ended in a draw.
Neutral ground
1,000 simulated matches · Same squads and tactics
- Real Madrid won
- 36.0%
- Draw
- 25.9%
- Barcelona won
- 38.1%
360 out of 1,000 matches
259 out of 1,000 matches
381 out of 1,000 matches
For example: Real Madrid won 360 matches here, Barcelona won 381, and 259 ended in a draw.
Barcelona at home
1,000 simulated matches · Same squads and tactics
- Real Madrid won
- 29.3%
- Draw
- 22.4%
- Barcelona won
- 48.3%
293 out of 1,000 matches
224 out of 1,000 matches
483 out of 1,000 matches
For example: Real Madrid won 293 matches here, Barcelona won 483, and 224 ended in a draw.
These are recorded simulation frequencies, not predictions for a real match.
On neutral ground Madrid won 360 times and Barcelona 381 times. A 2.1% gap is too small to confidently declare a superior team from this sample. The useful conclusion is that these two frozen lineups produce a close matchup under the chosen settings.
Madrid's win rate moves from 29.3% away to 44.3% at home, a 15.00 percentage-point difference. Barcelona wins 48.3% at home and 31.6% away. This measures the venue setting inside this engine; it does not estimate the real-world effect of a particular stadium or crowd.
What we kept fixed
- The exact same starting XIs, bench pools and 4-2-3-1 formations in all three scenarios.
- Both teams: balanced mentality, medium pressing and balanced tempo. Automatic pre-match tactics were disabled by explicitly supplying these choices.
- Clear weather, injuries enabled, 90 minutes, no extra time or shootout. The engine's automatic substitutions and in-match events remain enabled.
- A different, recorded pseudorandom seed for every match. These are independent seed streams across venues, not paired replays of the same match.
- Only the venue parameter changes between scenarios. 'Home' and 'away' in CSV column names are engine-side identifiers: Madrid stays home/first and Barcelona away/second even when the venue favors Barcelona.
Fixing pre-match tactics matters: the engine's automatic manager can choose a different plan depending on the venue. Leaving that enabled would mix a change of venue with a change of tactical selection. Here the initial plans are fixed, while injuries, cards and substitutions can still make individual matches diverge.
The exact lineups we used
We used the builder's default lineup selection on the recorded dataset. That selection uses available positions and market values; it is not a claim about either coach's preferred XI. The full team snapshots, bench players and calculated side strengths are in the input file. Player ages were not used to interpret these results.
Real Madrid · 4-2-3-1
| Slot | Player | Dataset role |
|---|---|---|
| GK | Thibaut Courtois | GK |
| LB | Marc Cucurella | LB |
| CB | Dean Huijsen | CB |
| CB | Ibrahima Konaté | CB |
| RB | Trent Alexander-Arnold | RB |
| CM | Federico Valverde | CM |
| CM | Eduardo Camavinga | CM |
| LW | Vinicius Junior | LW |
| AM | Jude Bellingham | AM |
| RW | Yan Diomande | RW |
| ST | Kylian Mbappé | ST |
Available substitutes: Arda Güler, Aurélien Tchouaméni, Álvaro Carreras, Rodrygo, Endrick, Brahim Díaz, Denzel Dumfries, Bernardo Silva, Carlos Espí, Raúl Asencio, Thiago Pitarch, Éder Militão, Andriy Lunin, Jorge Cestero, Antonio Rüdiger, Ferland Mendy.
Barcelona · 4-2-3-1
| Slot | Player | Dataset role |
|---|---|---|
| GK | Joan García | GK |
| LB | Alejandro Balde | LB |
| CB | Pau Cubarsí | CB |
| CB | Eric García | CB |
| RB | Jules Koundé | RB |
| CM | Pedri | CM |
| CM | Frenkie de Jong | CM |
| LW | Anthony Gordon | LW |
| AM | Fermín López | AM |
| RW | Lamine Yamal | RW |
| ST | Raphinha | ST |
Available substitutes: Dani Olmo, Rodri, Karim Adeyemi, Gerard Martín, Gavi, Marc Bernal, Gabriel Jesus, Roony Bardghji, Andreas Christensen, João Cancelo, Xavi Espart, Dominik Livakovic, Brian Fariñas, Hamza Abdelkarim, José Contreras, Jesse Bisiwu, Wojciech Szczesny.
Goals, chances and the most common scores
| Venue | Goals: Madrid / Barça | Model xG: Madrid / Barça | Shots: Madrid / Barça |
|---|---|---|---|
| Madrid at home | 1.53 / 1.26 | 1.56 / 1.24 | 11.87 / 11.17 |
| Neutral ground | 1.37 / 1.41 | 1.38 / 1.39 | 11.52 / 11.53 |
| Barcelona at home | 1.24 / 1.68 | 1.24 / 1.60 | 11.15 / 12.14 |
On neutral ground the mean score is 1.37–1.41. At Madrid's home venue it becomes 1.53–1.26; at Barcelona's home venue it is 1.24–1.68. The direction is consistent with the engine's venue advantage, but individual outcomes remain varied.
| Neutral score (Madrid first) | Count | Share |
|---|---|---|
| 1-1 | 126 | 12.6% |
| 2-1 | 90 | 9.0% |
| 0-1 | 86 | 8.6% |
| 1-0 | 84 | 8.4% |
| 1-2 | 77 | 7.7% |
The most frequent neutral score, 1-1, appears only 126 times. A popular scoreline is therefore not a dependable prediction for a single match.
What this sample cannot establish
For Madrid's neutral win rate, a 95% Wilson interval runs from 33.1% to 39.0%; Barcelona's runs from 35.1% to 41.2%. These intervals describe Monte Carlo sampling uncertainty under the fixed model. They do not include uncertainty about whether the model represents real football accurately.
- The engine uses its own market-value and position-based strength model. It does not know future form, real injuries or a coach's actual match plan.
- Changing a player, formation, tactical setting, dataset or engine version creates a new experiment. The exact percentages here should not be transferred to those conditions.
- A reproducible experiment can demonstrate what the software did; reproducibility alone does not validate its real-world predictive accuracy.
Who won each match? Explore the scores
You can read the individual scores right here. Choose a venue to see who won each recorded match.
Download data and technical notes
These files are for anyone who wants to analyze the data. You do not need them to understand the results above.
- All match data (CSV)
3,000 rows of scores and statistics. Open with a spreadsheet app such as Excel; the column names are in English.
- Experiment inputs (JSON)
The exact players, bench and settings, in a technical data format.
- Computed summary (JSON)
The numbers used in this article, in a machine-readable format.
- Reproduction notes (English)
Explains columns, random seeds and the reproduction command for developers.
- File verification values
Technical fingerprints for checking whether the data files changed.
Prefer to start from the squads? Use the Real Madrid lineup builder or the Barcelona lineup builder to pick your own starting XI and formation, then take it into the simulator.
Starting shapes to try on the pitch
Open this setup, then change the players and make it your own.
Open this setup, then change the players and make it your own.
Build your own Clásico scenario
Open the two teams, change the lineup or tactics and explore how the model responds. The interactive page uses the current dataset, so it is not an exact replay of this frozen experiment.
More articles
We simulated Beşiktaş v Galatasaray 1,000 times: who has the edge at Tüpraş?
Ahead of the 30 November derby we played Beşiktaş v Galatasaray a thousand times at each of three venues: win rates, common scores, top scorers and home advantage.
How much is home advantage worth? We played seven fixtures 21,000 times
The same two squads at three venues. Seven big fixtures simulated a thousand times each at home, on neutral ground and away: what home advantage is worth, and when it does nothing.
Rain, wind and snow: how weather changes a football match
We played the same Clásico a thousand times each in clear weather, rain, wind and snow. How goals, pass accuracy and shots changed, and how to pick a team for bad weather.