Performance Posture

Speed comes from modeling choices, not heroic optimization. Standard drag tables are resampled once onto a small set of smooth spline knots, so each evaluation is a handful of branch-light arithmetic ops rather than a table walk. The curve view caches its segment index, which pays because Mach falls monotonically through a flight. The hot path has no heap, no exceptions, and no virtual calls.

No universal speed claim is made here: the benchmark tree reproduces the comparisons locally, and per-trajectory measurement beats any published figure. Instruction counts from the benchmark suite are tracked in CI, so a change that costs the hot path shows up as a regression before it merges.