A pointer to the study that you're probably thinking of is research done by Paul Nation and Robert Waring (vocabulary researchers). They cite their own 1985 study and a 1989 study with the following quote: "With a vocabulary size of 2,000 words, a learner knows 80% of the words in a text which means that 1 word in every 5 (approximately 2 words in every line) are unknown. Research by Liu Na and Nation (1985) has shown that this ratio of unknown to known words is not sufficient to allow reasonably successful guessing of the meaning of the unknown words. At least 95% coverage is needed for that. Research by Laufer (1989) suggests that 95% coverage is sufficient to allow reasonable comprehension of a text. A larger vocabulary size is clearly better." [1]
1. http://www.fltr.ucl.ac.be/fltr/germ/etan/bibs/vocab/cup.html
However, I have no doubt that the paper I was reading, references this one. Great find. Thanks!
Perhaps you're thinking of glosses? I'm not familiar with graded readers, but I am rather familiar with mediæval manuscripts, wherein it's common to see copies of Latin works with little annotations near certain (sometimes all) words. When the gloss has been added above (or below) the word(s) they annotate, they're said to be interlinear, but marginal glosses aren't unheard of.
The glosses were sometimes written by the same scribe that made the copy, but often they appear to have been added later, perhaps by the owner of the book—sometimes in a comically small hand so as to fit in narrow spaces :)
Thus, I found myself to be more successful with a notebook.
There is another reason though. In addition to the word I write the Japanese definition of the word as well (from a children's dictionary). I mostly avoid using/memorizing English translations. This is just my personal preference, and certainly not the speediest or most effective way depending on your goals.
Every day, I'd do that, and also do my SRS reviews. Every week or so, I'd reread the whole week's text. If it was a particularly difficult book, then I might do the same thing after a month (ie. read the whole month's work).
Almost always, by the time I got to half way through the book, I didn't need the SRS software anymore. Authors tend to use the same vocab and grammar over and over again. So I could read several books written by that author with almost no difficulty (although it depends on the level).
Anyway, what I found was that this was the fastest way to bootstrap me to free reading. If I didn't memorise vocab as I went, then I'd need to look up stuff for the entire book (and beyond). But if I did memorise vocab, then I could easily get to a point where I'm just reading casually without any desire to look stuff up. Studies (which I don't have pointers to at the moment, unfortunately) have shown that you need 95% comprehension (on average) to learn new vocabulary and grammar from context. So the key is to get yourself there (in the context of what you are reading) as quickly as possible. It makes a big difference.