Skip to main content
A Crunch Node is the set of services you build and host to run your Crunch. It contains your competition logic — how data flows in, how models are called, how predictions are scored, and how results are reported.

Crunch Node Introduction (5:48)

Your Crunch Node relies on Model Nodes to execute participant models remotely. You never run or manage models directly — you call them over the network via the Model Runner Client.

​
What a Crunch Node does

In practice, your node handles four responsibilities:
  1. Data ingestion — collect or host the data your competition needs (market feeds, datasets, APIs)
  2. Model orchestration — send data to connected models, trigger training and inference, and collect predictions
  3. Scoring — evaluate predictions against ground truth using your scoring function, aggregate results, and build a leaderboard
  4. Reporting — expose results via API so the dashboard, participants, and external systems can consume them
Diagram of a Crunch Node with workers for data ingestion, prediction, scoring, and reporting, connected to the Model Orchestrator

Typical Crunch Node setup showing the four responsibilities

​
Infrastructure requirements

You are responsible for hosting your Crunch Node and everything around it: which data to use, when to trigger calls, how to compute scores, and how to produce results. A simple cloud server or on-premises machine is usually enough. The required capacity depends primarily on what you do around predictions — ingestion, scoring, aggregation, and storage. You do not need to size infrastructure for model execution. Models are deployed, scaled, and managed by Crunch Labs through the Model Nodes.
The crunch-node package provides a production-ready engine with all four responsibilities built in. You configure behavior through environment variables and a CrunchConfig. See the Getting started guide.

Next: Model Nodes

Understand the managed infrastructure that executes participant models securely.