Most phone cameras don’t take a photo all at once — they read the sensor one row at a time, top to bottom. If the subject moves during that sweep, every row catches it at a slightly later instant, so a straight object comes out skewed and spinning props turn to jelly. The camera is telling the truth about time; it just can’t hold the whole scene in one moment. Slide the speed and watch the bar lean.
A rolling shutter exposes row r at time t₀ + r·Δt (line delay Δt). A subject moving horizontally at speed v is displaced by v·Δt more on each lower row, so a vertical edge is sheared into a diagonal of slope v·Δt per row — the total top-to-bottom skew is v·Δt·(rows). Rotating blades exceed this and smear into curves (the ‘jello effect’). A fail-loud self-check throws unless a moving subject produces a nonzero shear (and a still one produces none). The single-instant scene is the thing lost — the frame is a diagonal slice of spacetime, not a slice of time. ◆ real sensor physics, node-verified.
A linear shear models horizontal motion exactly; true rotating-subject ‘jello’ is the same per-row delay applied to a rotation (curved, shown in the 3D corner). Global-shutter sensors avoid this — at a cost, honestly the tradeoff, not a defect to fix in software alone.