How to analyze a prediction market trader's track record

Correction (July 16, 2026): this guide previously presented undated wallet and cohort figures as real evidence without a committed query or provider artifact. Those figures, the named wallet examples and their paired charts are withdrawn. The URL remains stable so earlier citations reach this correction. The charts on this page now come from committed studies, each with its sample and run date in its caption.

Separate the sources first. A provider-level portfolio total answers how the account stands overall. Resolved-market rows answer narrower questions about signs, distributions and per-market outcomes. Open positions, deposits, redemptions, rebates and other portfolio effects must not be forced into a market-row sum.

Then check the boundary of the evidence: retrieval date, market-status scope, sample depth, null handling, and whether one event dominates the result. A live profile changes. A durable comparison needs a dated artifact and a reproducible calculation.

Win rate counts positive outcomes, not their size. A trader can be right often and lose money when the occasional loss is much larger than each gain. Another can be right less often and stay profitable when the winners are larger than the losses.

Win rate also follows the price paid, as the chart above shows. Buying contracts near $1 produces many small wins with little upside, and one loss can erase them.

Treat a win-rate comparison as incomplete until it names the resolved denominator, the payout basis and the time boundary. Read it beside gross gains, gross losses, resolved-market count and position size.

Profit factor is gross positive result divided by the absolute gross negative result on one declared row scope. Above 1, gains exceeded losses on that basis. Below 1, the opposite.

The denominator matters. Zero-loss samples need an explicit policy, unresolved markets must not silently enter the calculation, and a thin sample makes the ratio unstable. Compare two wallets only when both use the same status, accounting and time rules, and read a trend only when every point is computed the same way.

A cumulative P&L chart shows path, not cause. Check whether it uses portfolio accounting or a market-row projection, whether open positions are included, and when each point becomes final.

Look for concentration. One vertical jump can dominate an otherwise flat history. Repeated gains across independent resolutions are a different kind of evidence from one windfall, and neither shape proves what information or process produced it.

Use the live profile for current values, and keep a dated capture before citing a curve publicly.

Category breadth and category skill are different questions. A concentrated trader may understand one domain deeply. A diversified trader may depend less on one event cycle. Neither pattern is automatically better.

Compare categories only after normalizing their taxonomy, resolved scope, capital basis and observation window. A category with a handful of markets should not be read like one with a deep history.

A grade is a product summary. Open the scorecard to inspect profit, resolved-market depth, calibration, risk and consistency, including which metrics are unavailable or uncertain.

Read the letter beside realized profit, resolved-market depth and the risk metrics. A short history leaves more uncertainty than a deep one, even when the headline results look similar.

Cohort-relative values move when the wallet changes or the comparison population changes. Record the retrieval date before using a grade in an article or a decision.

A short perfect record is still a short record. The chart above shows how far a measured win rate can sit from the truth at each sample size. Check how many independent markets resolved, whether one payout dominates, and how the result changes when open positions are excluded.

Maximum streaks are sensitive to sample length, so compare them with the full distribution of wins and losses.

Profit beside extreme drawdown or concentration describes a fragile path. Compare position size with liquidity and with the rest of the wallet before reading anything into an absolute dollar amount.

A benchmark is only as sound as its denominator. Record which wallets qualified, how nulls were handled, which date the cohort represents, and whether every wallet used the same accounting scope. The chart above is one dated example: the graded wallets on September 16, 2026, by grade.

Percentiles are relative to the selected population. They set no absolute threshold for skill and they move as the population changes.

Use the current leaderboard filters for exploration. Before publishing a cohort claim, capture the exact read-only query and its output beside the article.

Profit alone does not identify its cause. Look for evidence that entries beat the market price, that results persist across independent opportunities, that risk stays survivable, and that the method works at the observed position size.

A markout compares the entry price with a later market price. Positive movement can support the idea that an entry was well timed, but the horizon, side normalization, missing prices and eligible-entry count must be explicit. A decaying positive markout and a reversing negative one are useful conceptual shapes, not proof until reproduced from dated observations.

Execution matters too. Maker versus taker behavior, fees, liquidity and price impact can explain why a strategy that looks attractive in summary is hard to reproduce.

Two equal portfolio totals can hide different downside. Compare typical and maximum position size, drawdown, concentration, capital deployed, and time spent below the previous high.

Average and median position size answer different questions. A wide gap between them exposes occasional outsized bets.

Risk labels, automated classifications and anomaly summaries are prompts for investigation. Inspect their inputs and dates, and return to the underlying positions before turning a label into a public claim. Hourly and category concentration describe when and where a wallet operates. They do not prove the schedule or category caused the result.

Aggregate results hide where gains and losses originate. The chart above shows how concentrated they are across sports wallets, by percentile group. For one wallet, break results down by a declared category taxonomy, entry-price range, position-size range and market status.

Each bucket needs a denominator and a capital basis. Sparse buckets look extreme, and moving the bucket boundaries can change the story. Treat the current profile breakdown as mutable exploration, not as a dated citation.

A profitable history is not automatically copyable. Latency, liquidity, fees, capital, holding period and automation can make the observed execution unavailable to another trader.

Concentration measures can show whether many positions behave like a few underlying bets. Read them with market context, not as a universal cutoff.

Study the method and its constraints instead of mirroring trades. Use a live profile to form questions. It cannot guarantee that another account can reproduce the result.

To keep a wallet in view while you study it, the wallet tracker guide at 0xinsider.com/learn/how-to-track-a-polymarket-wallet covers finding its address, reading its open positions beside its settled P&L, and following it.