Worked example: phase as an outcome

Suppose every trial contains an interpretable inspection landmark.

registration = register_to_landmarks(
    gaze,
    observed_landmarks=inspection_times,
    reference_landmarks=np.array([1.0]),
)

Analyze timing deformation

phase_fit = fit_phase_fpca(
    registration,
    n_components=2,
)

The phase FPCs describe dominant ways in which traversal timing departs from the reference.

Preserve raw landmark latency

landmark_frame = phase_landmark_frame(registration)

The phase function and the observed landmark times answer related but not identical questions.

Check what registration changed

sensitivity = compare_registered_unregistered_fpca(
    registration,
    n_components=3,
    scaling="dimension_sd",
)

registration_sensitivity_frame(sensitivity)

If matched component similarity or score correlation falls substantially, registration has changed the covariance structure rather than merely producing a cleaner plot.