You would likely be interested in HSRS (Hierarchical SRS). (Not my project but I know the author.) The tl;dr it’s a framework for a. representing hierarchical knowledge dependencies, including grammatical structures, b. “sampling” from points in the tree, where deps are often a class of objects, c. sequencing such samples so that material is introduced only when its deps are satisfied, and d. inferring which parts of the dep tree were likely most responsible for recollection failures and updating their review state accordingly.
My bet would be that you could probably get claude code to write most of the deck basically from where you son is up to ~late elementary school, and from there you have escape velocity. I’d be very curious if you do something like this, I suspect that early literacy edtech is heavily underoptimized etc.
You would likely be interested in HSRS (Hierarchical SRS). (Not my project but I know the author.) The tl;dr it’s a framework for a. representing hierarchical knowledge dependencies, including grammatical structures, b. “sampling” from points in the tree, where deps are often a class of objects, c. sequencing such samples so that material is introduced only when its deps are satisfied, and d. inferring which parts of the dep tree were likely most responsible for recollection failures and updating their review state accordingly.
My bet would be that you could probably get claude code to write most of the deck basically from where you son is up to ~late elementary school, and from there you have escape velocity. I’d be very curious if you do something like this, I suspect that early literacy edtech is heavily underoptimized etc.