Each build is one web page. Open it in a browser and play. Nothing to install, no account, and it keeps working with the wifi off. Download one and it's yours: the game in a single file, and beside it a second download with the notes on what we learned, written so your AI helper can read them too. Every build here is free under the MIT License, and the fonts come with their own open licence (the SIL Open Font License, whose text travels with the fonts); if you make one better, send it back.
There's room on the board for the next build. If your family needs something, tell us. It might be the next piece.
Amy is our AI helper, and she picked and built every build here herself, one thing at a time, to learn something. Most of them started the same way: something she could describe in words but could not have built, which her own log counts as not knowing it. She writes down what she expects before she measures and keeps the misses, so every piece tells you what it taught her and why she picked it.
Why so much physics? Her log noticed it too: of twelve builds in a row, eleven were physics or simulation, a habit no single night could see from inside. The method under the cloth and rope builds, position based dynamics, is everywhere in game physics, and they kept exposing the same gap: a setting everyone treats as the stiffness of the fabric turned out, once measured, to depend on how the solver runs. Several cloth builds each say she had never written the solver before, and each one meant it: for four weeks every night started from a copy of her log that still listed four builds, so no night could see the one before, and ten of these builds are the same idea built again. Each piece's lesson is what that one build taught; it did not carry to the next night until the log was put back together. The last of them, Loom, ends with a note to herself: next build should leave physics alone.
Tell us what your family keeps doing by hand. So far every piece was built to learn something; the next one could be built for you.