Wallet Profiler

Deep behavioral profiling and auto-profit strategies for Polymarket wallets. Analyze timing, market selection, order flow, and edge, then auto-execute.

Pro Feature. Requires a Pro or Ultra subscription. Get started at api.mathematicalcompany.com

Wallet Profiler

How it relates: Wallet Profiler builds on score_wallet() scoring and extends it with 4-dimensional behavioral analysis and automatic strategy generation. For raw trade data, see Wallet Analytics. To execute, see Copy-Trading.

Goes beyond basic scoring to build a complete behavioral profile of any Polymarket wallet. Analyzes how a wallet trades across four dimensions and generates concrete strategies to profit from its patterns.

Behavioral Profiling

Temporal patterns, market selection, order flow, and edge detection across 20+ metrics.

Strategy Generation

Up to 10 rule-based strategies with triggers, sizing, and execution params.

Auto-Execution

One-call hunt() or pipeline hunter() to profile and trade automatically.

MCP Integration

Available as an MCP tool for AI agent workflows.

---

Quick Start

import horizon as hz

# Full profile
profile = hz.profile_wallet("0x1234...")
print(profile.best_strategy)

# Auto-execute best strategy (standalone)
hz.hunt("0x1234...")

# Pipeline mode (inside hz.run)
hz.run(
    markets=["some-market"],
    pipeline=[hz.hunter("0x1234...")],
    interval=30.0,
)

---

Analysis Dimensions

Temporal Analysis

Answers when does the wallet trade?

Metric	Description
Hour distribution	Trade count per hour (0-23 UTC) with peak hour and Shannon entropy
Day distribution	Day-of-week distribution with weekend ratio
Session	Dominant session: US (13-21 UTC), Asia (0-8), Europe (7-16), or round-the-clock
Burst analysis	Burstiness metric B = (sigma - mu) / (sigma + mu) of inter-trade intervals
Active days	Percentage of days in the analysis window with at least one trade

Market Selection

Answers what does the wallet trade?

Metric	Description
Categories	Inferred from market titles: politics, crypto, sports, economics, entertainment, science
Win rate per category	Performance breakdown by category (requires position data)
Herfindahl index	Market concentration (0 = diversified, 1 = single market)
Early entrant score	Fraction of buys near opening prices (0.45-0.55)

Order Flow

Answers how does the wallet enter and exit?

Metric	Description
Building style	single_entry / scaling_in / dca (based on avg buys per market)
Exit style	single_exit / scaling_out
Hold time	Average hours from first buy to first sell per market
Size trend	increasing / decreasing / stable (linear regression on chronological sizes)
Dip buyer score	Fraction of buys below market average price (0-1)
Rip seller score	Fraction of sells above market average price (0-1)

Edge Detection

Answers is the wallet good? Includes the composite score from score_wallet() and extends it with deeper metrics:

Metric	Description
Timing quality	Average (current - entry) / entry across positions
Win rate by size	Win rates for small / medium / large trades (33rd/66th percentile buckets)
Streak type	streaky / mean_reverting / random (lag-1 autocorrelation)
Edge in bps	Estimated edge derived from average PnL percentage
Confidence	0.1 to 0.9 based on sample size (10+ to 200+ positions)

---

Strategy Generation

The profiler generates up to 10 rule-based strategies, each with a concrete trigger condition, implementation type, and execution params:

Strategy	Trigger	Type
fade_loser	composite < -0.3	fade
copy_winner	composite > 0.3 + info score > 0.15	copy
session_pre_position	Bursty + dominant session > 60%	session_trade
front_run_timing	Timing CV < 0.3	front_run
stop_hunt	Stop cluster > 30%	counter_trade
fade_direction	Directional bias + negative edge	fade
copy_best_category	Category win rate > 60%	copy
concentration_exploit	HHI > 0.3 + negative score	fade
copy_dip_buyer	Dip score > 0.6 + positive edge	copy
timing_exploit	Hour entropy < 2.0	session_trade

Each ProfitStrategy contains:

Field	Description
name	Strategy identifier
description	Human-readable explanation
mechanism	How the strategy works
expected_edge	"high" / "medium" / "low"
risk_level	"high" / "medium" / "low"
confidence	0-1
impl_type	"copy" / "fade" / "front_run" / "counter_trade" / "session_trade"
params	Execution parameters (inverse, size_scale, poll_interval, etc.)

---

Auto-Execution

hunt() and hunter() use Copy-Trading under the hood - they profile the wallet, pick the best strategy, then delegate to copy_trades() / copy_trader() with the appropriate params.

hunt(wallet): Standalone

Profiles the wallet, picks the best strategy, and auto-executes via copy_trades() with the appropriate params.

hz.hunt(
    "0x1234...",
    size_scale=0.5,
    max_position=1000.0,
    exchange="paper",
    dry_run=True,
)

hunter(wallet): Pipeline

Creates a pipeline function that profiles on first call and delegates to copy_trader() each subsequent cycle.

hz.run(
    markets=["some-market"],
    pipeline=[hz.hunter("0x1234...", size_scale=0.5, dry_run=True)],
    interval=30.0,
)

The profile is stored in ctx.params["wallet_profile"] for inspection.

---

MCP Tool

The profiler is available as an MCP tool:

profile_wallet(address="0x1234...", trade_limit=500)

Returns a flattened dict with all metrics and strategies.

---

Data Types

Type	Fields
FullWalletProfile	wallet, score, analysis, temporal, market_selection, order_flow, edge, strategies, best_strategy
ProfitStrategy	name, description, mechanism, expected_edge, risk_level, confidence, impl_type, params
TemporalProfile	hour/day distributions, session, burst, active_days
MarketSelectionProfile	categories, concentration, diversity, early entrant
OrderFlowProfile	building/exit styles, hold time, size trend, dip/rip scores
EdgeDetection	information score, timing quality, win rate buckets, streaks, edge bps

---

Algorithms

All analysis is pure Python (no ML dependencies):

Algorithm	Purpose
Shannon entropy	Distribution uniformity
Coefficient of variation	Regularity detection
Linear regression slope	Size trend detection
Lag-1 autocorrelation	Streak analysis
Herfindahl index	Market concentration
Burstiness metric	B = (sigma - mu) / (sigma + mu)
tanh normalization	Bounding scores to [-1, +1]
Keyword matching	Market category inference

Wallet profiling uses Polymarket’s public Data API. Results are based on historical behavior and may not predict future performance. Use responsibly and ethically.