It seems that there are some informal levels of canon in Trek. Levels of canon in Star Wars are, of course, well-known. (For instance, I came across a line in the TPM novelization a few days ago about Qui-Gon's Jedi master having been in the order for 400 years. Dooku was only 80 in RoTS, and of course Yoda has been there for 900 years or so at that point, so it couldn't refer to either of them. So, it's simply an instance of higher (and later) canon overriding the older, lesser material.)
I don't think we've ever really had layers of canon suggested in Trek before.
Mike Sussman, producer of Enterprise and writer of the "In a Mirror, Darkly" episodes, disavowed some information he wrote that appears on a screen in the episode. He says that what he wrote (and some that did not appear on screen) was not "hard canon" in a TrekBBS post, noting in the episode commentary on StarTrek.com that he didn't get them cleared through the producers. On the other hand, he acknowledged in the episode commentary of IaMD's second part that if it's on-screen, it is canon, and that people would be freeze-framing the episode to see what was going on there.
This reminds me of the start of the Dominion War. In "A Call to Arms"[DSN5], the station is lost and the crew piles on to the Defiant and Rotarran. The two ships go to join the Federation/Klingon battlefleet. The producers were shocked, though, to see the episode air . . . the two ships join the fleet, and then the fleet is heading back toward Cardassian space as if to go kick immediate ass. The producers had to go back and change their plans for the start of the next season, showing that fleet trashed, because they'd never intended for the fleet to be headed back like that.
Or, there's the Defiant MSD. Ron Moore was astonished to learn that the Defiant had landing gear, but then when later asked if the Defiant would ever land he simply said it was a pricey effect and probably wouldn't be done.
But I digress. While the Sussman example specifically helps resolve the contradiction I mention in the latest site update (which contains a link to his comments), I'm wondering if it also applies to other information screens. These are commonly done by the art department, and sometimes contain in-jokes and whatnot that are commonly ignored anyway. Are these "soft canon", since they didn't go through the producers?
Of course, it's not too much of a leap from the info screens to say that dialog is considered more important than visuals overall in the Trek canon. Once the episode leaves the writers and producers . . . the guardians (or, frequently in Voyager's case, befoulers) of canon . . . it is left to the artists, VFX guys, and so on. While there are some people there like Okuda who are good about keeping things continuous (sometimes at the expense of the writers), there are other people like David Stipes (or the guys who Rob Bonchune mentions at HobbyTalk) who just don't know anything about Trek, or who change things up for dramatic reasons, or because it's 'kewl', and so on. There is oversight, of course, but it isn't like the writing team is there every step of the way.
It's something to ponder. But, then, it's tough . . . I mean, on the one hand you've got writers like Ron Moore, Manny Coto, Sussman, the "Bynars" Judith and Garfield Reeves-Stevens, and so on who can come out with some hella-cool stuff. On the other hand, you've got hacks like Braga or just weaker writers like Andre Bormanis. Do we really want their dialog overriding Okuda's screens or the work of the other Ron Moore, or Gary Hutzel? Imagine the Special Olympiad of a contradiction between Braga and Stipes!
For now, it seems clear that whatever is on the screen is canon, but that while the producers acknowledge this they focus primarily, but not exclusively, on the writers' works. Hence the changes to DS9's sixth season due to the surprise effects shot, and of course Sussman's comment in the IaMD2 commentary that he likes to leave spots open for the art department to go wild putting tech screens in and so on.
In other words, the dialog-over-visuals argument can be made, but I don't think it would ever be an end-all be-all position . . . there would be exceptions.