How Game Review works 🔬

Our game review feature analyses your moves, highlights your best moments, and suggests improvements.

Under-the-hood it uses our best bot, the Panda, to analyse your game. The Panda plays the game flawlessly, so you can be confident you can trust its judgements on whether moves are good.

How can you use the game review?

After you finish a game, click the Review Game button. This will take you to the review for your game, where you will be presented with a wealth of information about how your game went. What does it all mean? Let's unravel it.

First, look at the top-right on desktop, or the bottom of your screen on mobile. There, the review introduces the players, their accuracies, and a graph charting the light player's chance of victory throughout the game. The graph falling means dark has a higher chance of winning, and the crown is given to the player who won. 👑

Next, the Panda summarises your game and offers you some key metrics about your moves. Your moves are categorised from brilliant down to unfortunate blunders. I recommend focusing on the extremes: cherish your excellent and brilliant moves and strive to avoid blunders.

Once you are done with the summary of your game, click the prominent Review Key Moves button. Clicking this button will carry you to a specific move to review in more detail. You will be shown moves that the game review thinks are the most relevant for you to learn from.

You will be presented with detailed insights about one of your moves. In the example shown below, a mistake is highlighted to learn from. The Panda suggests that taking the central rosette would have been better than introducing a piece. The board shows this suggestion, with your move highlighted, and the best move marked with a star.

After reviewing your move, you can click the button to progress to the next move that the Panda has selected for you to review. The Panda will usually pick out 6-8 moves for you to review in each game.

How are game reviews made?

We look at every move you and your opponent made, and analyse each of them using the Panda bot. We compare each move you made to the move that the Panda would have made. If you made a worse move, we can then tell you exactly how much your winning chance fell because of your choice.

Once we have used the Panda to identify how good your move is, we then generate a description of why your move was best, and about why you should have made a different move instead. These descriptions are generated by matching what the Panda thinks with English descriptions of moves.

Move Categories

We categorise each of your moves as one of the following:

• Brilliant. The rarest categorisation. The brilliant category is reserved for moves that are very hard to spot, and that have a big impact on your game. Brilliant moves are a special type of best move.
• Excellent. The excellent category is used for moves that are hard to see. Excellent moves are a special type of best move.
• Best. The best category is used when you made the move that leads to your highest chance of winning.
• Great. The great category is used for moves that are very close to being the best move. Your chance of winning went down, but only by a small amount.
• Good. The good category is used for moves that could have been better, but that don't hurt your chance of winning too badly.
• Mistake. The mistake category is used for moves that you should try to avoid making. These moves have a significant impact on your chance of winning the game.
• Blunder. The blunder category is used for mistakes that are particularly hurtful to your chance of winning. Players should try to avoid blunders at all costs.

Limitations of the Panda

The Panda plays the game flawlessly. However, that doesn't mean that its moves are always the right move to make. Sometimes you may want to deviate from what the Panda would play.

The Panda always assumes that your opponent will play flawlessly from that point onwards. However, if your opponent makes mistakes in their play, the Panda will not take advantage of that! In these cases, you may want to play differently to the Panda, so you can win more against a specific opponent.

For example, you might notice that your opponent refuses to leave the central rosette, even to capture. In this case, you can worry less about moving your pieces that are just after the central rosette. This could free you up to play better against your opponent in other areas of the board. This might be the best play against this specific opponent. However, the Panda may still identify this as being bad play.

Hopefully you now understand the Panda's weakness. However, don't let that make you think that you can use this to beat the Panda! If a player notices that you leave your pieces in front of the central rosette longer, they might also use that knowledge to play better against you. The Panda has no flaws like this for other players to take advantage of.

That is why you should still strive to learn to play like the Panda! Playing like the Panda will stop your opponents being able to exploit your mistakes.

Want to dive a bit deeper?

One of our long-time community members, Raph, has developed a website where you can explore what the Panda would do from any position in the game. It only supports the Finkel ruleset right now, but it is great for exploring positions in more detail. You can try it out on Raph's website here.

We have also open-sourced the source code of the Panda! If you'd like to experiment with the solved game, and you have a technical background, you might find the following resources useful:

• The final trained models are available on HuggingFace.
• Our Java library, RoyalUr-Java, can read and train these models.
• Our Python library, RoyalUr-Python, can read these models.
• Jeroen's Julia implementation can really quickly train models that use a different file format.