Learning to weld is learning to see through a mask. The welding arc is blindingly bright; the protective visor darkens everything except the small puddle of molten metal. The novice must simultaneously control torch angle, travel speed, arc length, and filler rod feed while interpreting the puddle's color, shape, and motion through a darkened lens. Traditional instruction: watch the expert, then try. The expert's critical adjustments — the subtle wrist rotation when the puddle widens, the pause when it narrows — happen too fast and too small to see from a student's vantage point.
Augmented reality changes the channel (arXiv:2603.07959). WeldAR overlays guidance directly onto the welder's view during the live task. Not a separate screen. Not a video to watch first. The instruction appears where the work is happening — a projected path for the torch to follow, a color-coded indicator for speed, a warning when the angle drifts.
The key finding is not that AR-guided novices weld better during guidance — that is expected. It is that AR-guided novices weld better after guidance is removed. Skills learned with real-time AR overlay transfer to unassisted independent performance more effectively than skills learned from traditional video instruction. The guidance is not a crutch; it is a scaffold that, once removed, leaves the motor pattern intact.
This makes sense biomechanically. The novice with AR guidance practices the correct motion from the first attempt — correct angle, correct speed, correct distance. The neural pathways encode the right pattern immediately. The novice with video instruction practices incorrectly first, then corrects. The neural pathways encode the wrong pattern first, then must be overwritten. Unlearning is harder than learning.
The tool teaches best when it disappears. The weld that matters is the one made without help.