FinTech · Market Microstructure · Execution Quality

Testing executable edge in Polymarket BTC short-horizon markets.

A public research portfolio project studying whether apparent prediction-market pricing edge can survive spread, fill probability, latency, order failure, model gates, and settlement outcomes.

Main takeaway: the public sample does not support a profitability claim. The interesting result is the gap between positive tick replay and weaker live-like execution outcomes.

Open Live Dashboard View GitHub Repo Read Reports
Signal funnel showing execution stages
Execution funnel: where theoretical edge is lost before filled exposure.

Key findings

What the project shows

01

Simulation-to-live gap

Pure tick replay can show positive edge, but live-like ledger records become weaker after latency, quote staleness, order failure, fill probability, model gates, and settlement outcomes are included.

02

Execution funnel dominates

The bottleneck is not a single cost such as spread. The harder question is whether a signal survives acceptance, fill probability, ML EV filtering, order submission, and settlement.

03

Extreme probabilities are fragile

Very low and very high probability buckets are less stable in the public sample, especially around the final resolution window where small price changes can cause large binary-outcome errors.

04

ML is an execution gate

The public workflow exports safe scalar ML EV and fill-probability diagnostics, but it does not ship private model artifacts, raw feature JSON, or production execution logic.

Visual summary

From signal edge to executable exposure

The dashboard and reports focus on the gap between apparent pricing edge and realized execution quality. The figures below summarize the public-sample execution funnel and observed status breakdown.

Signal funnel chart
Signal funnel
Execution status breakdown chart
Execution status breakdown

Research artifacts

Explore the public demo

Live Dashboard Interactive Streamlit dashboard built only on public sample artifacts. Execution Quality Report Signal funnel, rejection reasons, edge decay, settlement PnL, and author takeaways. Probability Calibration Report Fair probability vs market-implied calibration and tail-bucket instability. ML Filter Report ML EV gate, fill-probability diagnostics, walk-forward validation, and caveats. Risk Simulation Report Monte Carlo terminal PnL, drawdown, losing streak, and path dependency. Six Research Notebooks Sample data, execution quality, reference price, calibration, ML diagnostics, and risk.

Methodology

Public-safe research workflow

The project uses anonymized and downsampled public sample data to separate theoretical edge from executable edge. It keeps private raw ledgers, wallets, order IDs, model artifacts, and live execution systems out of the public repo.

Public sample data
Fair probability analysis
Edge and execution diagnostics
ML EV / fill-probability gates
Reports, notebooks, dashboard
Monte Carlo risk simulation

Tech stack

Tools used

Scope and disclaimer

Research demo only

This is not financial advice, trading advice, investment advice, or a recommendation to participate in any market. The repository does not claim to provide a profitable strategy, does not claim to predict BTC, and does not provide a live trading or betting system.