Finite WORLD rigid-frame velocities over a positive interval: translation secant in metres/second and shortest-arc rotation secant in radians/second. Both frames must describe the same rigid identity in WORLD coordinates. The caller owns deterministic actual sampling and selects one continuous branch; these two frames cannot reveal intervening turns, constraints or impacts. This is not an instantaneous derivative or a momentum estimate.
Rotation admission uses the engine's existing 1e-6 unit-norm rule. Admitted rotations are normalized into private copies for the logarithm. A computed relative quaternion with exactly zero scalar component refuses because its half-turn has two equally short axes. Nonpositive/nonfinite time, invalid components or unrepresentable arithmetic throw RangeError without changing either input. Scale is outside this rigid frame observation; a deforming point needs its own actual point samples.
Finite WORLD rigid-frame velocities over a positive interval: translation secant in metres/second and shortest-arc rotation secant in radians/second. Both frames must describe the same rigid identity in WORLD coordinates. The caller owns deterministic actual sampling and selects one continuous branch; these two frames cannot reveal intervening turns, constraints or impacts. This is not an instantaneous derivative or a momentum estimate.
Rotation admission uses the engine's existing 1e-6 unit-norm rule. Admitted rotations are normalized into private copies for the logarithm. A computed relative quaternion with exactly zero scalar component refuses because its half-turn has two equally short axes. Nonpositive/nonfinite time, invalid components or unrepresentable arithmetic throw RangeError without changing either input. Scale is outside this rigid frame observation; a deforming point needs its own actual point samples.