Encirclement of Polymarket Bots

Author: Liu Kaiwen, EeeVee, BlockBeats

On November 22, 2025, a silent showdown was taking place on a prediction market called Polymarket.

On one side of the confrontation was a mysterious trader named @totofdn. On the other side was an automated arbitrage bot named sunshines.

It all began with a trivial order. @totofdn placed a tiny sell order: 5 shares of No at $0.34. This action instantly compressed the market's bid-ask spread to less than $0.04—this was the magic number that triggered the platform's "order book rewards."

Almost in the same second, sunshines reacted. A massive sell order was slammed into the order book: 100 shares of No at $0.34. The bot arrived, strictly following the code's instructions, to earn the platform's liquidity rewards.

But it didn't know that this was exactly the signal @totofdn had been waiting for.

@totofdn unhesitatingly consumed all of the bot's sell orders, acquiring 100 shares of No at an average price of $0.34. Meanwhile, the bot was forced to take on 100 shares of Yes at an average price of $0.66, completely unaware that it had fallen into a trap.

This was just the beginning. Over the next four hours, this "fake order, real eat" combination was repeated over and over. Sunshines acted like an out-of-control ATM, spitting out cash time and again. In four hours, dozens of repeated operations resulted in over $1,500. The bot's account was precisely emptied, while @totofdn left quietly, unscathed.

This was a meticulously planned "cognitive encirclement" targeting automated scripts. It revealed a truth about on-chain arbitrage: here, automation does not equal intelligence; AI can enhance efficiency but may also lead to greater losses.

The "Optimal Solution" and "Fatal Flaw" of Platform Incentives

To understand the intricacies of this battle, we must first return to the rules of Polymarket itself. As a decentralized prediction market, liquidity is its lifeline. To incentivize users to provide depth to the market, Polymarket designed a mechanism called the "Order Book Rewards Program."

The core idea of this mechanism is simple: those who provide liquidity to the market can earn rewards. Specifically, as long as users place their limit orders within the blue maximum spread line (the so-called "spread") in designated markets and meet certain share requirements, they can proportionally share in the reward pool provided by the platform. Rewards are automatically distributed at midnight every day, straightforward and direct. This spread is usually the current midpoint price ±3ct to 4ct, with the specific width set in real-time by Polymarket.

Once any rule is quantified, it inevitably gives rise to specialized "score-farming" strategies targeting that rule. Polymarket's order book rewards quickly attracted a group of special "miners." They were not concerned with the outcomes of the predicted events; they only cared about how to efficiently earn rewards. Thus, automated arbitrage bots like sunshines came into being.

The code logic of these bots is as follows:

• Scan the market: Continuously monitor all markets that meet the reward conditions.
• Assess the spread: Check if the current market's bid-ask spread is less than a certain threshold (e.g., $0.04).
• Trigger orders: Once a market's spread meets the liquidity reward requirements, immediately place an order within the spread range that complies with the reward rules.
• Claim rewards: Wait for the midnight reward distribution.

From a coding perspective, this logic is flawless; it perfectly utilizes the rules. The bots tirelessly "fill the spread" across various markets, contributing liquidity data to the platform in exchange for rewards. They are the "optimal solution" to the rules, the "model citizens" in Polymarket's eyes.

However, the problem is that these bots only analyze spreads, shares, and rewards; they do not understand market sentiment, opponent analysis, and have no risk control. They cannot discern whether that suddenly appearing tiny order that compresses the spread to trigger conditions is a genuine trading demand or a carefully laid trap.

When @totofdn placed that sell order of 5 shares of No at $0.34, sunshines' code told it: "Opportunity has arrived! The spread has been compressed to 1¢, quickly place an order to earn rewards!" It had no idea that this $0.01 spread was fake, artificially created. It only saw the "optimal solution" of the rules but failed to recognize the "fatal flaw" behind this solution.

Ultimately, this bot, born for rewards, became the prey of a more advanced hunter due to its mindless pursuit of rewards.

From Physical Warfare to Cognitive Warfare

From MEV to Jito, and now to the "bot hunting" at Polymarket, the silent war of on-chain arbitrage is undergoing a profound evolution.

If the early MEV (Maximum Extractable Value) wars were a "physical war" centered around gas fees and block space, today's on-chain games increasingly resemble a "cognitive war" that tests strategy and psychology.

In the primitive era of MEV, victory belonged to those with the fastest networks, toughest hardware, and highest priority packing rights—"scientists." They were like a group of trucks barreling down a highway, using absolute power and speed to front-run, sandwich, and liquidate, extracting value from ordinary users' transactions. It was a simple and brutal era, where the competition was about whose "muscle" was more developed.

Subsequently, MEV solutions represented by Jito emerged, attempting to establish a new order for this chaotic physical war. By auctioning block space, Jito redistributed MEV profits, allowing validators and stakers to share in the spoils. This somewhat alleviated network congestion but also made MEV acquisition more "legitimized" and "industrialized." The war moved from the shadows to the light, from individual heroism to an arms race between professional institutions.

The incident at Polymarket reflects that on-chain games have entered a new stage. What determines victory is no longer millisecond-level latency or exorbitant gas fees, but rather an understanding of the rules, insights into market players, and the application of strategy.

@totofdn did not use any advanced hacking techniques or mobilize massive computational resources. His only weapon was a profound understanding of Polymarket's reward mechanism and an accurate prediction of the behavioral patterns of automated scripts like sunshines. He won a war of information asymmetry and, more importantly, a war of cognitive dimensions.

The laws of the on-chain dark forest are changing. Purely automated scripts, if lacking the ability to dynamically adapt to the environment and the awareness of opponent strategies, will find it increasingly difficult to survive. They are like species that have not fully evolved; while highly efficient in specific ecological niches (like farming rewards), they become powerless when the environment changes or when faced with more advanced predators.

From the physical war of MEV to the order war of Jito, and now to the cognitive war of Polymarket, on-chain arbitrage is evolving from a game of "engineers" to a game of "strategists" and "psychologists." In this increasingly complex dark forest, only those participants who can continuously evolve and enhance their cognitive dimensions will ultimately survive.

Recommended Reading:

Rewriting the 18-Year Script, U.S. Government Shutdown Ends = Bitcoin Price Will Soar?

$1 Billion Stablecoin Vanishes, What’s the Truth Behind the DeFi Chain Reaction?

MMT Short Squeeze Event Review: A Carefully Designed Money-Making Game

Click to Learn About Job Openings at ChainCatcher