How Similar is TOO Similar in the Great Agent Hunt?

Research agents for even a short while and you're almost sure to come up with two competing bits of wisdom: 

LOOK FOR AGENTS WHO REP THE BOOKS MOST LIKE YOUR OWN!

vs

AGENTS WON'T TAKE ON WORK TOO SIMILAR TO WHAT THEY ALREADY REP.

Guess what? BOTH these contradictory statements are true! ....Yayyy??

Of course you want to pick an agent who does the kind of books you do, and hopefully reps some authors you admire. . . but yep, that agent will likely decline if the books are too similar. I wrote a post way back in 2011 about WHY agents can't take on work that competes with what they already rep. It's all still true, so I won't rehash it here. We know the WHY. But how can you tell if your book falls into this problematic area?

A quick way to decide if your book might be too close to what an agent already reps: If you break your book and the comparable book(s) down into general CATEGORY, TONE and THEME - TWO of these can match. But if all three overlap, it's probably too close.

In other words: 

I could rep two funny picture books ... but not two funny picture books about Ninjas. I could rep two picture books about Ninjas, if one was funny, while the other was non-fiction/factual. I could rep two funny Ninja books... if one was a picture book and one was a middle grade. (That isn't to say that there isn't room IN THE WORLD for multiple funny picture books about Ninjas, btw... just that I personally would feel uncomfortable repping all of them!)

In the case of something like "heartfelt middle grade fiction about girls growing up" - where there are certainly lots of great books that seem to overlap... the differences might be more subtle. I rep both Linda Urban and Kate Messner, for example - two great authors, both sometimes writing in a similar space - but you wouldn't confuse CROOKED KIND OF PERFECT with BRILLIANT FALL OF GIANNA Z. You just wouldn't. On the surface there are similarities, but there's a difference at the bone. 

So if you're researching an agent who reps what you write... and you've thought about the chart and you see the surface similarities but you still think YOUR difference is different enough... you might as well try querying the agent... why not, right? Nothing to lose. Nobody is going to be mad at you - the worst that can happen is, you get a rejection, and that isn't anything to lose sleep over.

Does this make sense? Helpful, or have I muddied the waters even further?