Researching additional high-value data types for labeling beyond the ones already analyzed. Expanding the scope of what The Constraint will produce as a public dataset.

Debugging and structuring the first challenges for The Constraint. Refining format, instructions, and evaluation criteria to ensure they are clear and reproducible for participants.

Evaluating the infrastructure needed to support experiment uploads on notlabel. Exploring Marimo as the notebook layer — reactive, Python-native, and well-suited for reproducible research workflows.

Postponed onboarding session with a researcher group. Rescheduling pending.

Onboarded a new research team to notlabel. They are now running their own research on the platform.

Configuring inference pipeline and Hugging Face integration for The Constraint challenges. Testing first challenges to validate they meet the quality bar for the first edition.

Selected the first challenges to publish based on deep research findings. Curated for maximum dataset value around small LLM usage in domain-specific tasks.

Set up The Constraint social media presence. Accounts ready to start building the community ahead of the events.

Shipped the challenge platform for The Constraint — live and open to the public.

Running deep research on the most valuable model constraints for the first edition. Goal: identify constraint combinations that generate a genuinely useful public dataset — high research value for the small LLM and domain-specific use case community.