Skip to content

Local-first prediction-market infrastructure

OddsFox Pipeline

Build inspectable local prediction-market warehouses with Dagster, dlt, DuckDB, dbt, and Python.

Hypertrial-owned MIT software. No hosted service or bundled production data. Licence scope · Operator responsibilities.

Get started Query the warehouse

Pipeline

Start in the repository

uv sync --group dev

Start with a task

Analyze the data

Open a local DuckDB warehouse, pick a documented mart, and start from tested SQL and Python examples.

Analysts hub

Operate the pipeline

Install the project, keep schedules disabled, complete a validated scope run, then keep the warehouse healthy.

Operators hub

Contribute code

Change adapters, dbt marts, orchestration, or docs with the right quality gate.

Contributors hub

Integrate downstream

Consume documented marts without treating pipeline output as execution. Strategy consumers use private wc2026.v1 separately.

Integrators hub

Recent releases

Highlights from 0.2.0:

  • Golden mart polymarket_wc2026_market_hourly_odds selects a primary CLOB token for every admitted market and publishes primary_outcome_label.
  • Polymarket hourly odds sync fetches CLOB price history via batched POST /batch-prices-history.
  • make export-marts-parquet exports present *_marts relations to Parquet.
  • Kalshi candlestick sync and series-scope registry refresh surface partial failure counts without changing fail-open partial-success behavior.

Full changelog.

Supported local scopes

Version 0.2.x supports Polymarket FIFA World Cup 2026 (golden mart: polymarket_wc2026_market_hourly_odds) and Kalshi WC2026 stage and group-winner markets. FIFA fixture/results ingestion runs in the Kalshi full pipeline and isolated match-minute path, not the default Polymarket quickstart.

An isolated finalized Polygon WC2026 settlement-history mart with internal-audit and operator-local technical-export paths is optional and advanced; it is not required for ordinary odds analysis.

This site is software documentation and does not host datasets.

Choose a scope, read the FAQ, or review the architecture before extending the pipeline.