Draw y=x2 on tracing paper and slide the paper 3 units up. Every point rises by 3, so the new graph is y=x2+3. Slide it 4 units right instead. A point that sat above x=0 now sits above x=4, so to find the new height at x you look up the old height at x−4, and the formula is y=(x−4)2. That is why a minus sign inside the brackets moves the graph to the right.
Stretches follow the same logic. y=3x2 triples every height, making the parabola look narrower, and y=−x2 flips it upside down. Inside the function, y=(2x)2 reaches each height with half the input, so the graph is squeezed toward the y-axis by a factor of 21. Outside changes act on outputs and do what they look like; inside changes act on inputs and do the reverse.
A transformation keeps the parent's basic shape: a parabola stays a parabola and a square-root curve stays a square-root curve. Rules such as x2+x3 or (x2)2=x4 change the shape itself, so they are not transformations of x2. Feeding a linear rule into the parent, as in (x+1)2, is a transformation (a shift left 1), but feeding in a curved rule is not.
Think of it like this. Transformations work like the controls in a photo editor: move, resize, stretch, flip. The picture's content never changes, only its position and proportions. The horizontal controls are wired backwards, though: typing x−4 moves the picture right, and multiplying x by 2 squeezes it.
Core idea. A transformation shifts, reflects, stretches, or compresses a known parent graph by editing its formula.