What is Cravensworth?

Cravensworth is a experimentation framework for Django inspired by Indeed's experimentation tool, Proctor.

Why do we need another feature flipper?

Alternatives like django-waffle, django-flags, or gargoyle have been around for a long time. They provide similar functionality and are battle tested. So, why do we need yet another alternative?

Cravensworth is designed to address common limitations of these libraries. Cravensworth is more than just a feature flipper; it's built for experimentation and flexible control over your applications.

Here's how Cravensworth stands apart:

Simplified and Non-Intrusive Design

Many Django feature flag libraries tie switches directly to database models, leading to:

• Mandatory Database Migrations: Cravensworth requires no migrations out of the box. Database models are¹ used exclusively for data storage and retrieval, ensuring database changes are intentional and controlled.

• Overloaded Data Models: Other libraries can encourage placing business logic or caching within data models, which complicates maintainability. Cravensworth promotes clean separation of concerns.

• Clunky APIs: Instead of forcing you to memorize complex Domain Specific Languages (DSLs) for conditional logic, Cravensworth lets you to use simple Python expressions.

• Fewer Flag Types: Some libraries distinguish between different types of flags (e.g., switch, sample, flag). This complicates. Cravensworth provides only one type: experiment. Although the concept of a "switch" exists in Cravensworth, it is only for convenience; switches are experiments.

Less Intrusive Overrides

Cravensworth provides a robust override mechanism that avoids the pollution of your APIs and offers the ability to restrict override capabilites to known audiences.

Advanced Experimentation Capabilities

Cravensworth is built for experimentation:

• Multivariant Support: Run multivariate tests with multiple experiment variants.

• Data Export: Cravensworth allows the export of experiment state so you can integrate with whatever logging or data collection tools you use.

• Reliable Variant Assignment: Other libraries may rely on unreliable techniques like cookie-based variant pinning, which can skew results and doesn't work well across multiple devices. Cravensworth uses a hashing strategy to ensure stable variant assignments based on identities, even when a request object isn't available (e.g., in background jobs), leading to more accurate and reliable experiment results.

Why doesn't it support {feature}?

We are yet in baby times, here. Not all planned features are implementated, and it's possible not all needed features are known.

If you think something important is missing, check if it's in the Roadmap. If it's not there, start a discussion by creating an issue on GitHub.

Getting started

It's time to make your monster. Happy experimenting!