The three, briefly
The grid is slow, cheap and educational. You draw a grid on the reference and a matching one on the canvas, then copy square by square. It takes the longest to set up and it teaches you to see relationships, which is why drawing tutors still insist on it.
Transfer paper is the least glamorous and the most reliable at small sizes. Print the drawing, put graphite paper underneath, trace the lines. No distortion, no maths, no equipment, and it stops being practical somewhere around A3 because you need a printed original at final size.
A projector is the fastest by a wide margin and the only one that scales to a wall. It is also the only one with a failure mode that is invisible while it is happening.
The failure nobody mentions
A projector produces an undistorted rectangle only when its lens is exactly square to the surface, at the same height as the centre of the image, and level. Any other position throws a trapezoid.
Standing slightly to one side makes one edge of the picture longer than the other. Standing lower than the centre makes the top wider than the bottom. Both happen by default, because the projector is on a table and the canvas is on an easel, and neither is where the geometry wants them.
The reason this catches people is that the distorted image looks completely convincing. There is nothing to compare it against on a blank canvas, so the trapezoid is traced faithfully and the error is only discovered later, when a face is subtly wider on one side and nobody can say why.
This is also why a projected image that fits the canvas is not evidence of anything. A trapezoid can fit a rectangle perfectly well by overhanging on one side and falling short on the other.
Four corners and a scale bar
The correction is to stop trusting the projection and start measuring it. Pacyra puts a marker at each corner of the projected guide and lets you drag each one onto the corresponding real corner of the canvas. The image warps to match, which undoes whatever the angle was doing.
That takes about a minute and removes the entire class of error. It is manual on purpose. The room is not something software can see, so a check you perform is worth more than a calibration a program claims to have done.
Then measure. A scale bar is thrown alongside the guide, and holding a ruler against it tells you whether 10 centimetres on the canvas is 10 centimetres in the plan. If it is not, the size is wrong and everything traced from it will be wrong by the same proportion. Check before you draw anything, not after.
Choosing between them
Under A3 and working from a printed guide, use transfer paper. It is faster than setting up a projector and it cannot distort.
Above A3, or on a wall, use a projector with corner correction and check the scale bar. Nothing else scales.
Use the grid when the point of the exercise is to improve your drawing rather than to arrive at a finished picture. It is genuinely better at that, and it is the only one of the three that leaves you more capable than it found you.
None of them is cheating. Painters have traced, gridded and projected for centuries, and the finished painting is not made better or worse by how the first lines arrived. What matters is that the lines are in the right place, which is the only thing this whole comparison is really about.