Skip to content

How we grade traders, score signals, and where we fall short.

Every score, grade, and metric traces back to the pipeline described on this page.

31,822
Traders tracked
12,756,432
Market positions analyzed
40+
Quant metrics per trader per day
20,547
Days of time series data

Cross-platform analytics

Prediction market data is fragmented across platforms with different architectures. 0xinsider normalizes all of it into a single analytics layer so you can compare traders, track large trade activity, and analyze performance regardless of where the trades happen.

Polymarket
On-chain trade data from the blockchain. Full position reconstruction, P&L computation, and trader profiling. Our most complete dataset with the longest history.
Kalshi
Exchange API data from the CFTC-regulated platform. Large trade tracking and market monitoring. Individual trader identities are not available due to exchange anonymity.

From raw trades to scored trader profiles

Raw trade data doesn't tell you which traders have real edge and which got lucky. Turning 12,756,432+ scattered positions into graded trader profiles requires multiple layers of processing. Each layer feeds the next.

01
Ingest multi-platform data
Pull trade data from Polymarket (on-chain) and Kalshi (exchange API). Normalize across different data formats into a unified schema.
02
Reconstruct complete positions
Raw fills aren't positions. We match entries to exits, track splits and merges, handle market resolutions, and compute mark-to-market daily P&L for every trader.
03
Compute verified P&L
Calculate per-market profit and loss from reconstructed positions. Formula: realized P&L = SELL + MERGE + REDEEM - BUY - SPLIT. Every P&L number maps to verifiable on-chain data.
04
Extract behavioral features
Derive dozens of features from each trader's activity — holding duration, category concentration, position sizing patterns, timing, maker/taker ratios, and how these relate to outcomes.
05
Run ML models
Machine learning models trained on each trader's behavioral patterns. The models predict outcomes, detect strategies, classify trader types, and flag anomalies.
06
Score, grade, and classify
Compute 40+ daily quant metrics, assign S-through-F grades via Bayesian scoring, classify each trader into one of 10 strategy types, and detect large trades with signal scoring.
07
Categorize markets
Every market and large trade is classified into one of 12 categories — crypto, politics, NBA, NFL, soccer, MLB, hockey, MMA, tennis, pop culture, science & tech, and geopolitics — using keyword matching, metadata lookup, and ticker pattern analysis.

The grading system

Every trader with sufficient history receives a letter grade from S (exceptional) to F (severely underperforming). The grade is a composite of six normalized dimensions, each passed through Bayesian shrinkage to prevent thin histories from receiving inflated scores.

01
Profit Magnitude
Raw P&L scaled to a benchmark. Measures how much absolute profit a trader has generated. Higher is better, but this alone doesn't tell you about risk.
02
Sharpe Ratio
Risk-adjusted return — how much profit per unit of volatility. A high Sharpe means consistent returns, not lucky streaks. The best proxy for separating skill from luck.
03
Capital Efficiency
Profit relative to total volume deployed. A capital-efficient trader extracts more return per dollar risked. Particularly important in prediction markets where capital is locked until resolution.
04
Profit Factor
Gross gains divided by gross losses. A profit factor above 1.0 means the trader makes more than they lose. S-grade wallets average above 30 — for every dollar lost, thirty dollars gained.
05
Max Drawdown
The deepest peak-to-trough decline in equity. Penalizes traders who experience severe portfolio drops. S-grade wallets average 4.2% max drawdown. F-grade wallets: 78.4%.
06
Consistency
Rewards traders who maintain edge across varying market conditions and time periods. A trader who performs well in one month and terribly in the next scores poorly here, even if the average looks good.

Grade distribution

S
Top 3%. Exceptional across all dimensions.
A
Strong. Proven track record, reliable edge.
B
Above average. Solid fundamentals.
C
Mixed. Some wins, but inconsistent.
D
Below average on multiple dimensions.
F
Severely underperforming. Systematic losses.

Bayesian shrinkage pulls thin histories toward the population average. A wallet with 6 winning trades doesn't automatically receive an S-grade — the system requires statistical confidence before rewarding extreme scores. Grades update daily as new markets resolve. Full methodology: How Bayesian Confidence Scoring Works.

Smart money trade detection

We monitor real-time trade activity across both platforms. Large trades are flagged and enriched with contextual data to help you assess their significance.

Signal scoring
Each large trade is scored from 0 to 100 on size, trader grade, market liquidity, timing, and contrarian positioning. Higher scores indicate more potentially informative trades.
Spread analysis
Top-of-book bid-ask spread measured in basis points at trade time. Tight spreads (under 25bps) indicate liquid markets with active price discovery. Wide spreads (over 1000bps) flag illiquid orderbooks where any single trade can move the market.
Category tagging
Every large trade is classified into one of 12 market categories by checking title keywords, market metadata, asset tags, and ticker patterns.
Trader enrichment
Each large trade is linked to the trader's profile — their grade, win rate, P&L, and strategy type. A $50K trade from an S-grade market maker means something different than the same trade from an ungraded wallet.

Insider Radar

The Insider Radar analyzes trading patterns for signs of information asymmetry — trades that look like someone knew the outcome before the market did. Flagged trades receive a suspicion score from 0 to 100.

Timing analysis
Trades placed within hours of market resolution with high conviction (buying near 0 or selling near 100) are statistically anomalous. The radar measures the gap between trade time and resolution, weighted by price distance from the outcome.
Edge magnitude
How far was the trader's entry price from the final resolution? Buying at $0.15 on a market that resolves Yes ($1.00) represents enormous edge.
Coordination detection
Multiple wallets acting in concert — funded from the same source, trading the same markets within minutes, with similar sizing — suggest coordinated activity. The radar identifies wallet clusters through on-chain fund flow analysis.
Volume spike detection
A sudden surge in trading volume on an otherwise quiet market, immediately before a resolution-relevant event, is a classic insider pattern. The radar compares trade-hour volume against the market's historical baseline.

Strategy classification

Each trader is classified into one of 10 strategy types based on observed behavioral patterns. Classification factors in trading frequency, holding duration, category concentration, maker/taker ratios, position sizing, and actual performance. Profiles that show strategy-like patterns but no real edge are reclassified as Speculators.

Directional
Accumulator
Arbitrageur
Market Maker
Scalper
Swing Trader
Event-Driven
Momentum
Algo Trader
Speculator

Strategy labels are behavioral, not self-reported. A trader labeled “Market Maker” consistently provides liquidity on both sides of a market. A trader labeled “Speculator” shows no consistent strategy pattern despite active trading. Explore all strategy types at /trading-strategies.

Risk-adjusted metrics applied to prediction markets

Standard risk-adjusted metrics, computed daily for every tracked trader. Win rate alone tells you nothing. These metrics tell you whether a trader has genuine, repeatable edge.

Risk-adjusted returns
Sharpe ratio, Sortino ratio, Information ratio — how much return per unit of risk
Drawdown analysis
Max drawdown, average drawdown, drawdown duration, time underwater — worst-case loss exposure
Edge metrics
Profit factor, expectancy, Kelly fraction — expected profit per dollar risked
Risk measures
Value at Risk, Conditional VaR, P&L volatility — potential loss in a bad day
Consistency
Win rate, equity smoothness, cumulative profit, daily P&L — repeatable returns vs lucky streaks
Confidence intervals
Statistical confidence bounds on key metrics — how much to trust a sample of 20 trades vs 2,000

Data freshness

Not all data is equally fresh. We prioritize sync frequency based on trader activity and importance to minimize stale data where it matters most.

Hot
Every 5 minutes
Active in the last 24 hours, or S-grade traders. The traders you're most likely to be watching right now.
Warm
Every hour
Active in the last 30 days, or followed by someone. Relevant traders who haven't traded today.
Cold
On demand
Inactive for 30+ days. Synced when someone visits their profile or an admin triggers a refresh.

Current coverage

31,822
Traders tracked
12,756,432
Market positions analyzed
4,356
Traders with daily quant metrics
20,547
Days of time series data
Jan 1970
Metrics tracking since
Daily
Metric refresh frequency

Where we fall short

Here's what our system does not do well.

Survivorship bias
We track traders selected for their activity level and performance history. Traders who lost everything and stopped trading are underrepresented. This skews aggregate metrics upward — the real prediction market success rate is lower than what our data shows.
Kalshi trader anonymity
Kalshi does not expose individual trader identities. We track large trades and market activity, but cannot build individual trader profiles or grades for Kalshi participants. Polymarket remains the primary source for trader-level analytics.
Model accuracy versioning in progress
Models are retrained periodically, but we don't yet version them or publish accuracy trends over time. We're building a proper validation pipeline. Until it's done, we won't publish accuracy numbers we can't fully verify.
Resolution-dependent P&L
Losing shares on Polymarket never receive REDEEM events. Our P&L computation handles this by zeroing the losing side when the winning outcome is known, but markets with delayed or disputed resolutions may show temporarily inaccurate P&L.
Backward-looking only
Every metric, grade, and score is based on what already happened. Market regimes change, strategies decay, and past patterns are not guaranteed to repeat. We surface what a trader did — not what they'll do next.
Curated, not random
The traders we track were selected for having meaningful trading history. This is not a random sample of all prediction market participants. Conclusions drawn from this set describe the active, established traders — not all participants.
Insider Radar false positives
The Insider Radar flags statistical anomalies, not confirmed insider trading. A high suspicion score means the trading pattern is unusual — not that the trader definitively had inside information. Some flagged trades are simply well-timed bets based on public analysis.

What changes next

We're building versioned model tracking with proper validation and published accuracy metrics that are statistically rigorous. When we can demonstrate real predictive skill against a naive baseline, the numbers will appear on this page.

We're also expanding platform coverage — deeper Kalshi integration and exploring additional prediction market platforms as the ecosystem grows.

This page describes what our models produce and wherethe data comes from — not how accurate they are. We publish methodology now. Accuracy numbers come when we can verify them.

You've seen the method.

Now see what the data reveals about the traders you follow.

Get Insider access →About 0xinsider

7-day money-back guarantee · Free tier available