In-progress continuous operations should freeze rather than zero or cancel when tracking is lost

What it is

If tracking dies while a timeline is being dragged, a volume twisted, or a model rotated, the in-flight continuous operation should freeze at the last valid value: the picture holds, the value does not run. Zeroing snaps the value to a default or to the start; cancel discards the whole stroke. Freeze admits “the channel broke; the person may not have meant to stop.” Zero and cancel write a sensor fault as a user decision. This is independent of engagement grace: even if the session is still alive, a continuous value must not keep integrating while the hand cannot be seen.

Why it happens

A continuous mapping integrates joint position into a parameter. After loss, using the last noisy frame or a predictor to extrapolate lets the value drift in the blind, then jump on recovery. Zeroing looks “safe,” but for a slider already at 70% it is vandalism: the user has to walk that 70% again. Cancel may fire gesture semantics such as “drag off to undo,” turning an occlusion into a deliberate undo. Freeze latches output on the last stable sample and stops integrating. To avoid treating a noisy frame as that last stable sample, confirm loss over a short window before entering freeze; do not latch on a one-frame shake.

Studying it

Have participants drag a continuous control to mid-range, then insert an invisible occlusion. Compare freeze, zero, cancel, and blind extrapolation. Dependent measures include value error after recovery, reports of “it ran by itself / it vanished,” redo length, and accidental undo. Vary loss duration from short to long to see whether freeze remains acceptable after a long loss, or whether people start wanting cancel.

Where it stops holding

Safety-interlocked continuous quantities (hook height, tool feed) may have to enter a safe state rather than freeze in mid-air; freeze is not a universal safety policy. A discrete tap has no in-flight value to freeze. Predictive extrapolation can be smoother than freeze for a one- or two-frame loss; once the blind interval is perceptible, stop. When several people share one continuous object, freezing A's drag may block B; that needs an object-level lock.

Applying it

• Declare a loss policy per continuous parameter: freeze by default, and draw the frozen control as paused, not idle.
• Do not keep writing the noisy position from lost frames into the parameter; confirm over a short window before freeze.
• Acceptance: drag to about two thirds, cover the lens; the value should stay nearby, must not return to 0, must not close the panel. Recovery policy is a separate test; this one only checks the instant of loss.

Related

• Same group: C4.23.1 Tracking loss must be distinguished from ordinary action gaps so it is not treated as disengagement · C4.23.3 After tracking recovers, the system must decide whether to resume the frozen state or require a restart · C4.23.4 Prolonged tracking loss should prompt the user to adjust position or lighting
• Adjacent: C4.19 Re-entry grace · C4.25 Predefined fallback paths
• Search: freeze · tracking loss · continuous control