Analysis Methodology

These methods power the Analysis section on trader profiles. Every number has a documented formula — no black boxes.

Market Signals#

Market Signals answers: "Which official market contexts, keywords, and short phrases does this trader tend to trade best versus other profiled traders?" It is profile-first, all-time, and built from resolved market outcomes plus official Polymarket metadata.

Signal Generation#

Each market can contribute multiple candidate signals drawn from the official question, event title, market slug, event slug, group label, and official event tags. Polyanna keeps meaningful single words and short phrases, then prefers official labels over noisier free-text variants when they collapse to the same cleaned signal.

Edge Score#

The headline edge score is:

edge = trader_signal_win_rate - signal_baseline_win_rate

A positive value means the trader historically won more often than the profiled baseline for that signal. A negative value means they underperformed it. Signals are ranked primarily by edge, with attempts and total PnL used as confidence and tie-break context.

Confidence#

Confidence is driven by sample size. Low means only a few qualifying markets, medium indicates a modest sample, and high is reserved for larger signal histories. Signals with too little trader history or too little baseline history are hidden entirely.

Strategy Classification#

A rule-based system that classifies each trader into one of six strategy types based on observable on-chain behavior. No AI or machine learning — just transparent thresholds.

Strategy Types#

• Market Maker — High maker ratio and high frequency. Posts limit orders on both sides, profits from the spread.
• Sniper — Few markets with high win rate. Picks spots carefully and bets with conviction.
• Specialist — Activity concentrated in a few market topics. Deep expertise in specific domains.
• Diversified — Active across many categories. Spreads risk broadly rather than concentrating bets.
• High Frequency — High trade count per market with many total trades. Actively manages positions with rapid execution.
• General — No dominant pattern detected. May use a mix of approaches or have insufficient data for classification.

Evidence & Confidence#

Each classification includes an evidence list showing the exact data points that led to the label. Confidence is derived from how far the data exceeds the threshold — the stronger the signal, the higher the confidence.

Bot Behavior Breakdown#

Decomposes the bot probability score into its four constituent signals. Only shown when bot probability exceeds 80%.

Component Scores#

• Frequency (40% weight) — Normalized fill rate per active day. High values indicate trading pace beyond typical human capability.
• Maker Ratio (30% weight) — Percentage of fills from passive limit orders vs aggressive market orders. High maker ratio suggests automated market-making.
• Activity Entropy (25% weight) — Shannon entropy of hourly trade distribution (UTC). Maximum entropy means uniform 24-hour activity — no sleep pattern.
• Size Regularity (5% weight) — Inverse coefficient of variation of trade sizes. High regularity means consistent sizing typical of algorithmic execution.

Bot Type Classification#

From the dominant component signals, bots are classified into types:

• Automated Market Maker — High maker ratio + high frequency
• Automated Taker / Momentum Bot — Low maker ratio + high frequency
• Passive Liquidity Provider — High maker ratio + moderate frequency
• Mixed Automation — No single dominant pattern

Risk Deep Dive#

Answers the question: "If I copy this trader's strategy, what losses should I expect?" All metrics are computed from the trader's PnL time series and enriched trade behavior.

Max Drawdown & Recovery#

The deepest peak-to-trough decline in cumulative PnL, plus how many days it took to recover back to the previous high. A trader with +$100K PnL but -$50K drawdown is fundamentally different from one with -$5K drawdown. See also Max Drawdown metric docs.

Profit Factor#

The ratio of total winning trade PnL to total losing trade PnL:

profit_factor = sum(winning_pnl) / |sum(losing_pnl)|

A value above 1.0 means the trader makes more than they lose. Above 1.5 is strong. Above 2.0 is exceptional. Below 1.0 means net negative per trade.

Average Win vs Average Loss#

The ratio of average winning trade size to average losing trade size. A trader with a 40% win rate but a 3:1 win/loss ratio can still be highly profitable. This metric separates win rate from trade quality.

Longest Recovery#

The maximum number of days between two all-time-high PnL points. Shows the longest period a copy-trader would have gone without seeing new highs — a measure of patience required.

Risk Summary#

A plain-language assessment based on the combined risk metrics:

• Profit factor >1.5 with quick recovery → "Moderate risk — strong recovery pattern"
• Profit factor <1.0 → "Currently net negative per trade — high risk to copy"
• Large drawdown → "Significant drawdown history — prepare for large swings"