BTC 15-Min Microstructure KDE Pipeline

Real-time microstructure analysis of Polymarket BTC 15-minute prediction markets. Records L2 order book data via WebSocket, computes rolling statistical features in Rust, and visualizes their distributions as dark-themed KDE plots segmented by time phase.

Architecture

Rust core (src/lib.rs) — PyO3 native module:

• Connects to Polymarket CLOB WebSocket, maintains a live L2 order book
• Computes 6 rolling features over a 500-tick window per book update
• Writes feature snapshots to Parquet (throttled to 1 per 200ms)

Python orchestrator (main.py):

• Discovers the next active BTC 15-min market via Gamma API
• Waits for the 15-minute clock boundary, then triggers Rust recording
• Segments data into 3 time phases and generates KDE plots

Features Computed

Feature	Description
Tick Intensity	Order book updates per second (activity proxy)
Sample Entropy	Predictability of mid-price returns (m=2, r=0.2σ)
Permutation Entropy	Ordinal pattern complexity of raw prices (order=3, delay=5)
Contract Price Entropy	Shannon entropy of L2 volume distribution
Hurst Exponent	Trend persistence via R/S analysis (>0.5 = trending)
Microprice Divergence	Top-10 level VWAP minus mid-price (directional pressure)

Time Segments

Each 15-minute window is split into three phases for comparative analysis:

• First 12 Min (cyan) — baseline regime
• Last 3 Min excl 30s (magenta) — transition zone
• Last 30 Sec (yellow) — settlement convergence

Setup

Prerequisites

• Python >= 3.9
• Rust toolchain (for building the PyO3 module)
• maturin >= 1.0

Install

# Create venv and install Python deps
python -m venv .venv
source .venv/bin/activate
pip install polars seaborn matplotlib numpy requests maturin

# Build the Rust extension
maturin develop --release

Usage

Live recording (full pipeline)

# Wait for the next 15-min boundary, record 900s, then plot
python main.py

# Skip the wait (start immediately from current boundary)
python main.py --skip-wait

# Custom recording duration
python main.py --duration 600

Plot from existing data

python main.py --plot-only data_btc-updown-15m-1771688700.parquet
python main.py --plot-only data.parquet --boundary 1771688700

Tech Stack

• Rust — async WebSocket client (tokio + tungstenite), order book engine, feature math, Parquet I/O (polars)
• PyO3 / maturin — Rust → Python bridge
• Python — market discovery (requests), data segmentation (polars), KDE visualization (seaborn + matplotlib)

License

MIT