research

everything we have so far is open. one designer and an AI pair built the first game in about ten days in october 2026 and wrote down what mattered while it was fresh.

placeholder text. there is one document, and it is a field report, not a study. this page will grow as there is more to show.

the note

blending: notes from building a game with an AI pair is the running record: the design decisions, the deploy log with every prompt verbatim, what worked, what didn't, and what stayed unchecked. it is plain text on purpose.

• the text is released under cc by 4.0: share and adapt it, with credit.
• the code (the game, the server, the command-line player, the checks) is mit licensed, in github.com/smulz/blending.
• to cite it: bassoli, s. (2026). blending: notes from building a game with an AI pair. https://github.com/smulz/blending

what it says, in short

• the sound and pictures are the entrance. if they're off the beat they don't work as bait, so one clock drives both.
• small, atomic steps with immediate feedback are how the game was built and how it is meant to be played.
• an AI that holds the thread lets a mind that runs on tangents take them without losing the way back. the designer calls it correlating, not diverting.
• its limits, in its own words: one project, one designer, one model family, no control group. the claims are what we saw, not what we proved.

what we'd want to measure next

placeholders for questions we have not answered. if you study attention, learning, or games and want to help design any of these, get in touch.

• does the lure work on anyone besides the designer? who stays past the first round, and why?
• do kids who play notice the decision under the noise, and when?
• does moving to a beat help or hurt a kid with adhd hold a plan for three moves?
• when the tiles make the music, do players hear the counter melody they built, and can they name the interval?