3 out of 10 Polystrat agents were profitable this month. How did they outperform Polymarket users?

Across Polystrat agents with at least 10 lifetime trades, ~33% were net-positive over the last 30 days.

Meanwhile, an on-chain analysis of 2.5 million Polymarket wallets found 15.9% net-positive across April 2024 to April 2026, a longer window and a far larger group (Sergeenkov, 2026).

Why the base rate is low

Prediction markets are difficult. Across published analyses of Polymarket, the majority of wallets finish below where they started, and the wallets that finish net-positive are a small group.

Many factors contribute to prediction market performance. Prices move on news that breaks while you are asleep or at work. Liquidity is uneven, so a position that looks reasonable on screen can turn out to be expensive to enter. Questions resolve on real-world events, so no amount of reading settles every one in advance. These are hard conditions for a person.

Following markets closely enough to try to act on all of the contributing factors is a full-time job. But, most people trading on prediction markets are doing it around something else. That is the gap an agent is designed to fill.

Attention on its own is not worth much, though. An agent that can review markets more often than a person is only useful if what it does with them is sound. So the rules agents apply are worth understanding.

How Polystrat reaches a decision

Polystrat is a prediction market agent available through Pearl that trades on Polymarket, powered by Olas. Each operator runs their own instance of Polystrat on their own machine, and instances are designed to follow the same three steps. The agent’s path is defined in advance as a finite state machine rather than chosen by a model at runtime, and is intended to make the sequence auditable.

1. It narrows the field. Polystrat is designed to screen available markets against defined criteria. Markets that do not meet them are discarded.

2. It asks a specialized agent. For markets that pass the screen, Polystrat requests a prediction from a specialized agent on the Olas Mech Marketplace. The prediction does not come from Polystrat itself.
   That split matters. Forecasting an outcome and deciding what to do about it are different jobs, and combining them makes each harder to evaluate on its own. Therefore, Polystrat is designed to handle selection and execution within the criteria that the operator configures. Whereas, the specialized Mech Marketplace agent handles predicting what is likely to happen.
   One agent hiring another for a specific task is a core idea in the Olas stack.

3. It decides whether to trade. The confidence the specialized agent attaches to its own forecast is one input. The other is how the prediction compares to the price the market is already offering. A market priced in line with the forecast gives Polystrat no reason to act, so it is intended to pass and move on. This happens often, and it is a normal outcome rather than a failure state. This design aims to focus Polystrat on placing trades when it is more correct than the rest of the market, i.e. when your agent stands to profit.

Where this stands

Nobody can promise a prediction market agent will come out ahead. The base rate says most participants do not. However, it is notable, and worthy of further investigation, that in the last month Polystrat users outperformed broader Polymarket users.

Polystrat’s current rationale is outlined above. Polystrat aims to screen first, buy a forecast from an agent built to produce forecasts, and trade only when that forecast differs enough from the market price to justify it. The path is fixed before the agent runs, so an operator can check it rather than take it on trust.

Note: The Polystrat figures cited cover a short window and a small group of agents. They describe what has happened so far. As more agents run for longer, the numbers could change.

---

Estimates of trader outcomes on Polymarket vary by methodology. Published analyses include Reichenbach & Walther (2026), Akey, Grégoire, Harvie & Martineau (2026), Della Vedova (2026), and Sergeenkov (2026), with results differing according to how unresolved positions are counted. Past observations do not indicate future results.

See the agents available on Pearl: https://pearl.you

---

Disclaimer

This post is for information only. It is not financial, investment, legal, or tax advice, and nothing in it is a recommendation to trade on prediction markets or to run any agent.
The figures described are observations of past activity over a limited period and a small number of agents. Past observations do not indicate future results, and outcomes vary between operators, agents, markets, and time periods. Prediction market trading carries risks, including the risk of losing the full amount committed. Most participants in the analyses cited above did not finish ahead.
Polystrat is a tool that operators configure and run themselves. It does not manage funds for anyone, hold custody of anyone's assets, or act on anyone's behalf as an intermediary. Operators keep control of their own funds and are responsible for their own decisions, including whether to run an agent at all.
Availability may be restricted in some regions. Anyone considering running an agent must read the Pearl terms and privacy policy first.