Background
Maria Ostrowski was born in Wrocław, Poland and moved to the United Kingdom at seventeen to read mathematics at the University of Warwick. She finished her undergraduate degree in 2013 and continued at the London School of Economics, where she completed an MSc in Statistics in 2014 and a PhD in Statistics in 2018. Her doctoral thesis was on hierarchical Bayesian methods for sparse panel data, with applications to estimating skill in noisy decision environments — a topic that turned out to be unusually relevant to everything she has done since.
From 2018 through 2022 she was a quantitative researcher at a London-based high-frequency trading firm, working primarily on European equity options market-making. The job was not glamorous: most of it was hardening the firm’s skew model against regime changes and writing internal tooling for the trading desk. The part she liked best was the work on counter-party identification — using anonymized order flow to estimate which counter-parties were informed and which were noise — which is, in retrospect, the closest analogue in traditional finance to what Poly Syncer does today.
In 2022 she left HFT for a research role at a regulated betting-exchange operator in London, where she spent three years building the quantitative framework that scored individual punters on long-horizon risk-adjusted return. The methodology she developed there — a hierarchical Bayesian estimator that shrinks individual scores toward a population prior in proportion to sample size — is the direct ancestor of the wallet-scoring composite she now ships at Poly Syncer. She joined Poly Syncer as a co-founder in November 2025.
What she works on at Poly Syncer
Maria owns quantitative research end to end. The most important artifact she ships is the wallet-scoring composite, which combines risk-adjusted return, calibration, position-sizing discipline, and category breadth into a single score that drives the public leaderboard. The composite is rebuilt nightly and a fresh model snapshot is deployed every two weeks; Maria personally signs off on every release.
The Sharpe adaptation is the second flagship piece of work. A naive Sharpe calculation does not work cleanly on prediction-market portfolios because the payoff distribution is bimodal — positions resolve to one or zero rather than drift on a Brownian path — and because most positions are held to resolution rather than rebalanced. Maria’s adaptation reweights the volatility term using realized variance in the time series of mark-to-market notional, which closely matches the intuition traders have for risk-adjusted return without breaking down on bimodal payoffs. The full derivation is in the whitepaper.
Beyond those two systems, Maria owns the position-sizing logic that the execution engine consults before each mirrored trade — the layer that turns research into automated Polymarket trading at production cadence. The defaults users see in the dashboard — the suggested allocation per wallet, the recommended max position size given account equity — come from a fractional-Kelly framework she calibrated against eighteen months of historical Polymarket fills. She wrote the public explainer of that framework in the Kelly criterion piece.
Perspective
Maria is, in her own words, a statistician who happens to work on prediction markets, rather than a prediction-market enthusiast who happens to know statistics. The distinction matters to her because the most common failure mode she sees in the broader space is people designing scoring systems that look correct on small samples and fall apart at scale.
“The hardest problem in evaluating Polymarket wallets is the one nobody wants to talk about: most wallets do not have enough trades for any individual statistic to be meaningful. A naive 30-day Sharpe on a wallet with twelve closed positions is mostly noise. The whole craft of doing this well is figuring out how much to shrink each individual estimate toward the population — and how to do that shrinkage in a way that updates appropriately as evidence accumulates. Get that wrong and you publish a leaderboard full of recently-lucky beginners.”
The technical answer to that problem, in Maria’s framing, is hierarchical Bayesian estimation with a category-specific prior. The category-specific part matters because the variance of returns differs sharply across Polymarket categories — a sports book has fundamentally different statistical properties from an earnings book or a long-tail political market — and a single global prior would systematically under-shrink wallets concentrated in low-variance categories and over-shrink wallets concentrated in high-variance ones. The Poly Syncer composite uses a hierarchical prior that learns category-level baseline performance from the full population and then partially pools individual wallets toward their category mean.
Maria has also been vocal that the most interesting research questions in the prediction-market space are not the obvious ones. The obvious question is whether top wallets generate alpha. The answer, empirically, is yes and it is not particularly close. The interesting questions are downstream: how persistent is wallet-level skill across categories, how quickly does skill decay when a wallet’s edge becomes widely known, and what is the appropriate Bayesian prior to put on a brand-new wallet that arrives with a track record from another venue. Those are the problems she spends most of her week on.
Recent writing
- Top Polymarket Traders: 9 Patterns We See Across the Best Wallets
- Mirror Polymarket Wallets: How to Allocate Across Multiple Leaders
- Wallet Scoring Methodology — How We Rank Traders
- Risk Management for Copy Trading on Polymarket
- Kelly Criterion for Polymarket Position Sizing
- Poly Syncer methodology — the wallet-scoring composite