That is the whole book. If you started at Part 0 and got here, you have covered more of this subject than almost anyone who ships an icon animation.
You can read a path by eye. M4 5h16 is a pen instruction, the numbers are coordinates on
a 24 by 24 grid, and y grows downward.
You know why a straight line gets stored as a curve. Browsers refuse to interpolate two paths whose commands differ, so everything becomes a cubic and the refusal never happens.
You know the real problem is not the maths. Averaging two lists of numbers is arithmetic. Deciding which number pairs with which is the whole difficulty, and every technique in this book is a different answer to it.
You know the two answers. Solve the pairing at runtime and pay for it on every visit, or decide it once, write it down, and ship the answer. This book argues for writing it down, and the reason is not performance: it is that a decision written down can be reviewed and corrected, and one made in a solver on a stranger's machine cannot.
And you know when to stop. A residual that says the pairing is wrong is telling you something true, and two icons that cross-fade honestly beat a morph that looks like a bug.
Read the code. lib/morph is about twelve hundred lines, none of it clever, and build
your own is the order to read it in.
Then write a rig by hand. Two icons, three slots. It takes twenty minutes and it will teach you more than this book did, because the moment a pairing looks wrong you will know exactly which number to change and why.
Correspondence is the problem. Interpolation is arithmetic.
Everything else was working out where to put the answer.