By JTT Experiments
Research and prototyping with software and AI.
This is where By JTT tests ideas before deciding what deserves more work. Experiments can be product prototypes, game mechanics, developer tools, AI workflows or research exercises. Each entry records what was tried, what was used, what was tested and what was learned.
Latest experiment
Falling Blocks: Banked Clears
A from-scratch falling-block game testing a Banked Clears mechanic: completed rows persist until the player banks them for a burst multiplier. Built through the Experiments ByJTT Agent and reviewed with the Gauntlet.
Open experiment →Updates
View on XHow experiments work
Research first
Understand the idea, existing work, technical constraints, audience, market and licensing before choosing what to build.
Keep the question small
An experiment should test a specific assumption or mechanic. It does not need a full product roadmap to be useful.
Test the claims
Automated checks, execution, independent review and real-device testing are used where they match the question being asked.
Record what happened
Models, tools, sources, licences, results, failures and changes are documented when they are relevant to understanding the experiment.
Reuse good existing work
Established open-source tools and already-solved problems are preferred to unnecessary custom code, with attribution and licensing kept explicit.
Continue only when earned
A useful experiment can graduate into a standalone product or game. Others can remain useful as research without pretending to be more complete than they are.
AI is part of the method, not the result. It can accelerate research, code, testing and production workflows, but the experiment still has to answer a real question. See the full timeline, read how the lab works, or inspect the ByJTT Gauntlet used for evidence-based review.