How a flywheel holds a cube upright on one corner. Linearised plant, optimal control by Riccati recursion, and the torque budget that decides whether a build is possible before you buy anything.
Real work that has not been written up here yet. No link, because there is nothing behind it — which is the point of listing them separately.
Marauder build on an S3, custom target, display and touch.
OpenSCAD models, print-in-place constraints, shell count as the real gate.
Streaming audio, plugin engines, wake-word gating.
Variants as files, cases as rows, measured instead of guessed.
Server-authoritative combat, client-side animation that costs nothing.
Why every project has the same six sections. Coming back to a build a year later, the question is never “where did I put this” — it is “what did I buy, what did I run it on, and what did I measure”. Fixing the shape means a new project is a folder of six answers rather than a blank page.