Participation Eligibility

• Participants must represent a USA-based company with more than 20 employees*.
• The Challenge is open to individuals or teams of up to 6 people. Teams will designate a Team Leader, who will serve as the primary point of contact.
• Participants must be able to commit 10-15 hours across 1-2 weeks in February and March 2026 to the Challenge.
• Final eligibility shall be determined after application evaluation, at the discretion of the Organizer.

*Practitioners at companies based outside of the US may also register; we will consider their participation in the Challenge on a case-by-case basis

Use Case & Data Provision Requirements

• Participants can select any binary supervised machine learning (ML) use case and model on tabular data that has been in production for at least thirty 30 days.
• Data tables must reside in one of Databricks, Snowflake, or BigQuery platforms.
• Participants must provide access to normalized data related to the use case or a functional equivalent. All data should adhere to general relational database model requirements.
• A data dictionary should be provided if available.
• Participants are required to provide multiple linkable data tables (at least 4 and up to 20 tables).
• Views are acceptable if used to remove Personally Identifiable Information (PII) or for data cleaning.
• Acceptable table types include Dimension tables, Slowly Changing Dimension (SCD) tables, Event tables, Time Series tables, and Snapshot tables.
• Tables should only include structured data—no data containing images, PDFs, or nested JSON files.
• Data tables containing time-based records (SCD, Event, Time Series, Snapshot) must include a time column with a consistent format and a primary key column.
• Dimension tables must contain a unique primary key column.
• Teams must be available to answer questions on their data to help FeatureByte develop a basic understanding of the data, schema, and relationships.

Data Access Requirements

• FeatureByte must receive read access to all the tables in the dataset.
• Write access must be provided to a dedicated schema in the data platform for storing FeatureByte-generated files and tables.
• A service account must be created in the data platform for FeatureByte to read and write data.
• Appropriately sized cluster to handle the data volume on which  computations will be performed. Minimum requirements are: 
  • Snowflake Warehouse: X-Small
  • Databricks SQL Warehouse: Small
  • BigQuery: NA

Evaluation Criteria

• Model evaluation is based on predicted probabilities for the test dataset.
• Model performance improvement is considered meaningful if the AUC of the Participant’s model is at least 0.01 higher than the model created using FeatureByte. I.e. (AUC_Participant-AUC_FeatureByte) > 0.01

Evaluation Rules

• Participant's business-production model must have been in production for at least 30 days.
• Participant must provide a training observation table containing the target, ideally the same or very similar to the one used for training their model
• Participants must provide the partition schema with targets that was used to train and test their model (including test and holdout tables, if a holdout is available) and their model's AUC for the test/holdout data.
• At the evaluation stage, participants must submit predicted probabilities generated by their model, as well as target values for the test dataset.
• To claim the prize, participants must provide a list of features used in their model and indicate feature importance.
• Final determination of prize eligibility will be made by a panel of independent judges.

Prizes

• Winner (First Place): US$10,000.00
• First Runner-Up (Second Place): US$5,000.00
• Second Runner-Up (Third Place): US$2,500.00