Docs / Technical
Technical documentation.
The reference layer. Everything here is verified against the engine source and uses real identifiers (function names, table columns, constants), so a claim on these pages can be checked rather than taken on trust.
Two pages, at two scopes: the first maps what may write each object of a course; the second goes down one level into the single most consequential generation, the one that decides the exercises you are given.
-
Course modification.
The complete map: every course object, which means can write it (direct edit, AI generation, deterministic algorithm), and the exact context each generation reads. Plus the three data-flow diagrams.
Read
-
Session engine.
Exercise generation at source level: the six stages of the grid, the data structures, every calibration constant, the prompt assembly block by block, the validation loop, and a table of known gaps.
Read
How to read these pages.
- Known gaps are published. Both pages name mechanisms that are built but not fully wired, with their exact location. A reference that documents only what works is a brochure, and a gap on a public list is a gap that cannot quietly persist.
- Deterministic where it counts. Scheduling, review selection, the exercise grid and most grading are pure computation. The AI authors content into slots the engine has already chosen, and cannot escape them. That boundary is enforced in code, not requested in a prompt.
- Narrative first, if you prefer. Everything here has a plain-language counterpart in how it works.