Locked decisions

No vehicles: the contestant sprints, vaults, climbs, balances, swings and slides. The fixed-step model ships throttle, brake, lean/balance, hop and the traversal assist for each obstacle class.

Forward motion is on Z only. Left/right is lean and balance correction, which keeps the side-cinematic camera honest and keeps Mode A input counting deterministic.

One accumulator, one integrator. Input polling is edge-captured and consumed inside the same fixed step, so hop counts can never drift from physics.

Gate items are adapted from SAT practice-exam item styles and indexed by answer integer. MC rounds use gameshow cadence. Shipping as typed data now; the query surface is already a function call, so swapping to a Cloud-backed bank is a one-file change.

Each rider carries a seen set; the query falls back to the full pool only when a bucket is exhausted, and to pure gameplay when no question matches the obstacle parameter.

One authoritative world object stepped outside both canvases; each canvas is just a camera onto it. Profiles, level progress and trivia stats already use the exact row shape a Cloud table will hold.