i wonder why this very simple solution doesn't get nearly as much "best practice" than the other one above there >^.^<

I think that +v ?? v would be more accurate.

?

Wow! Great job!

Nice

What's the extra space at the beginning for?

Nice, I forgot that NaN is a falsy value

clever

Absolute Genius.
Comparing that with my solution, especially knowing that this was a 7kyu kata, makes me seriously doubt my own rank :'D
Made me learn about \b though, i guess i would have come up with something prettier if i knew about it ..?

Cached version of matrix.length and thus faster.

In my opinion
-1 for i
-1 for forEach rather than map ( the behavior of map is more verbose than modifying an object and expecting it to be modifed in the original array )

however, i am sure the forEach is more performant, if only marginally.

Why use width instead of just using i < matrix.length?

Update: I downloaded Firefox 29 on OS X, typed +"0" in the console and it returned 0
To double chech I copied the whole function and run it like this: sortDict({1:5,3:10,b:2,6:3,0:8})
The "0" was returned as a string in the array.
So now I see that if we define "0" in a variable zero and then compare +zero || zero it returns "0" and that is the case even in the newest browsers and on Node 5.1
As pointed in another answer, the best way to do it is >> isNaN(k) ? k : +k

The way I tested this on Node 5.1 and Chrome 48 it worked fine. I tried +"0" and it returned 0
Do you think this because the V8 engine in the newest version of Node and Chrome use ECMAScript6?