In Search of Brilliancies in Chess

Few ideas in chess are as captivating as the brilliant move. Players search for them in their own games, spectators celebrate them in grandmaster battles, and modern platforms eagerly highlight them with glowing icons and celebratory effects. A single move can transform an otherwise ordinary game into something memorable. These moments linger. They are replayed, shared, and studied long after the result itself has faded. In many ways, brilliancies are the emotional currency of chess.

Yet for all their appeal, brilliant moves remain strangely ill defined. Everyone feels they know one when they see it, but attempts to formalise the concept often fall short. As chess has become increasingly engine driven, the gap between what feels brilliant to a human and what is rewarded by automated systems has grown wider. This tension is what led me to start thinking more seriously about what a brilliant move actually is, and whether it can be identified in any meaningful way by software.

As someone building a chess platform, this question quickly became more than philosophical. I wanted to reward moments of genuine insight rather than simply reinforce engine approved tactics. But to do that, I first needed to understand what I was actually searching for.

Brilliant Moves

A brilliant move in chess is a highly obscure, creative, and unintuitive idea, often an only move, that departs from established principles in order to achieve a decisive advantage or to preserve the balance in an otherwise lost position.

At first glance, such a move frequently appears wrong. It may weaken the king, give up material, concede structure, or violate long held heuristics about coordination and safety. What makes it remarkable is not recklessness, but the presence of a deeper idea that justifies these concessions. That justification is rarely immediate. Even strong players may struggle to understand why the move works until the position has been explored in depth.

Brilliance is not simply a function of calculation. Many strong moves are difficult to calculate, yet feel natural once the underlying plan is understood. A brilliant move resists that process. It does not follow familiar patterns or standard plans, and it is often overlooked precisely because it contradicts what experience suggests should be correct. The move may not involve obvious threats at all. Instead, it subtly changes the nature of the position, creating possibilities that did not previously exist.

This distinction helps explain why many engine approved sacrifices fail to feel brilliant. Leaving a piece hanging because there is a forced mate a few moves later is usually tactical and direct. The idea is concrete and often visible once attention is drawn to it. While such moves can be accurate and even elegant, they do not necessarily reflect deep creativity. They are executions of calculation rather than expressions of insight.

Truly brilliant moves tend to reshape the game in a more profound way. They reveal hidden connections between pieces, reframe strategic priorities, or uncover defensive resources in positions that appear beyond saving. Often, they are the only moves that work, yet they do not feel forced when played. They feel inspired, as though the player has seen something that others have missed.

This is also what makes brilliancies so addictive to study. They offer a glimpse into a deeper understanding of chess, one that goes beyond rules and heuristics. They suggest that even in a game governed by strict logic, there is still room for imagination and vision.

Searching for Brilliancies

With a working definition in place, the question naturally shifts from philosophy to practice. How do you actually search for something as elusive as a brilliant move. For me, this question emerged not from abstract curiosity, but from building software. As I integrated more analysis tools and engines into my platform, I began to wonder whether brilliance could be approached indirectly, not by trying to label it outright, but by triangulating it from different perspectives.

Traditional chess engines like Stockfish are extraordinarily good at one thing: evaluating positions with ruthless objectivity. Stockfish works by exploring vast search trees, pruning aggressively, and refining its evaluation through a combination of heuristics and brute force calculation. It does not care whether a move feels natural, elegant, or human. It cares only about whether the move improves the evaluation of the position under perfect play. Given enough depth, it will reliably identify the strongest continuation, the margin between alternatives, and the consequences of even small inaccuracies.

But this strength is also a limitation. Stockfish tells us what is best, not what is hard to find. A move that is obvious to a strong player and a move that borders on the absurd can receive identical evaluations. From the engine’s perspective, they are equal. From a human perspective, they could not be more different.

This is where my thinking began to shift toward pairing engines rather than relying on a single one. Alongside Stockfish, I had been experimenting with Maia, a neural network trained not to play perfectly, but to play like humans at specific rating levels. Unlike traditional engines, Maia is not optimising for objective strength. It is optimising for likelihood. Given a position, it predicts which moves a human player at a given level is most likely to choose.

This difference is crucial. Maia captures something that Stockfish fundamentally does not: expectation. It reflects habits, biases, and patterns of human decision making. It knows which moves feel natural, which moves align with common plans, and which ideas players tend to overlook entirely.

The idea that followed was simple, perhaps even naive, but intriguing. What if a brilliant move is one that lives at the intersection of these two systems. A move that Stockfish considers overwhelmingly superior to all alternatives, yet one that Maia believes humans are very unlikely to play.

Imagine a position where Stockfish identifies a single move that preserves the evaluation, while all other candidates lead to a significant collapse. The gap between the best move and the next best is large, not marginal. At the same time, Maia assigns that best move a very low probability, suggesting that most human players would never consider it. The move is not just strong. It is necessary, obscure, and unexpected.

Of course, this approach is far from perfect. There are many reasons a move might be unlikely for humans to play. It might be dull, defensive, or simply unfamiliar. There are also positions where human intuition aligns surprisingly well with engine truth, even in complex situations. A low probability alone does not imply creativity, just as a large evaluation gap does not imply insight.

But perhaps that is beside the point. The goal is not to produce a flawless detector of brilliance, but to move closer to capturing its essence. By combining objective necessity with human improbability, we begin to filter out moves that are merely strong and highlight those that are both critical and counterintuitive.

Applications

Thinking about brilliance in this way naturally leads to the question of what could be built from it. More than anything, I am curious about the results. I do not expect a perfect system, and I am not trying to solve a grand theoretical problem. What I find mentally stimulating is the idea of experimenting with this pairing of perspectives and seeing what emerges. To my knowledge, this specific combination of objective evaluation and human likelihood has not really been explored (I could be wrong) as a way to surface creative moments in chess, and that alone makes it worth pursuing.

A brilliancy detection tool built around this idea would not attempt to declare definitive truths. Instead, it would act as a lens. It would scan games for positions where the engine identifies a narrow and critical path forward, while a human trained model suggests that path is rarely chosen. The output would not be a badge of honour so much as an invitation to look closer. Why did this move matter so much. Why was it so easy to miss. What idea made it work. Even false positives could be interesting, because they would still highlight moments where human intuition and objective necessity diverge sharply.

What excites me about this approach is that it shifts the focus away from raw tactical fireworks and toward moments of genuine insight. It has the potential to surface quiet moves, defensive resources, and positional ideas that would otherwise be overlooked because they lack immediate spectacle. In that sense, it could change not just how brilliance is detected, but how it is understood.

I do not know whether this approach will work particularly well. It may produce noise, or highlight positions that feel unremarkable once examined closely. But even that outcome would be useful. It would tell us something about the limits of our definitions and the gap between what we value in chess and what we can measure. For me, that exploration is the point.

I hope you enjoyed reading this blog. I just wanted to put my thoughts onto paper and see what people think. Has this been done before. Am I being silly, naive, or is there something interesting here. As a chess novice, I would genuinely love to hear your thoughts.

Kind regards,
Toan (@HollowLeaf)