You're not a Platonist, are you?
Regardless, it would be plausible to imagine a character conceived of by an author, who has curious behaviors in early stories, the motivation for which is only revealed in a much later story. This would be an example of a process of "discovery" about a character, but this discovery is happening in the form of revelations in the text rather than by the act of the author.
If you wanted to make the analogy more precise, you could imagine a character created with peculiar traits, and the author later realizes a single explanation for those traits which they had not originally conceived, which they then write into a story. In these two cases, the effect on the reader might be identical, but in one essential information about the character was held back and in one it was developed later.
All of this is a little academic, though, since none of the information about Holmes & Watson in the post-1923 stories has a material effect on their characters, as ACD's estate attempted to argue.