Full-time Transductions
What if you show the transductions all the time? Don’t make the user click to see transductions. If transductions can be shown instantaneously, then just show them whether the user wants them or not. Make the computer work, not the user.

If users don’t want to look at the transductions, then they don’t look at them. Even if users only need to see the transductions 30% of the time, it still saves 0.3 clicks per interaction, as well as removing the confusion with Generate Debug.
On the other hand, maybe:
The user needs to see the parsing in the context of whatever is behind your window (i.e., the content of the parent window). Expanding to show transductions makes that harder.
Some users are confused or distracted by transductions, so it’s best to make it an item of progressive disclosure to insulate certain users.
Visual Weight
If both transductions and debug information should be under progressive disclosure, then the problem you have is that, as a consequence of visual design and layout, the Generate Debug has more visual prominence than Show Transactions.
The lower right of a dialog is where users tend to be looking after studying the parsing, and, by habit, the lower right is where users expect the “do this next” control to be (like in a wizard).
With its border and shading, a command button appears larger and has more visual contrast to its background than a check box, so it stands out more.
So swap it around. For example, debug can be a lightweight link, which it is, in a way, in the sense that it links to a new window. I’d label it “Debug Info” rather than “Generate Debug” to emphasize this, and make it consistent with what the user wants to do (which is see the debug, not generate it per se). Transductions can be a progressive disclosure button (with the “>>” symbology), which makes its effect on the window more predictable than a check box does.

In (Weak) Defense of Confirmation Dialog
Regarding the general problem of preventing accidental activations, the confirmation dialog is indeed the GUI equivalent of physically covered button or switch. It has the advantage over a physical cover that you can provide user with text explaining exactly why the command may be a bad idea. For example, the confirmation could say “Generating debug may take several minutes.”
On the other hand, users almost always think they know what they’re doing (and, believe it or not, they’re usually right about that), so they’ve gotten into the habit of not reading confirmations and going ahead anyway, so all you get with a confirmation dialog is a clunky annoyance.
Safety Covers
If you’re only worried about slip-of-the-mouse accidental clicking (“I meant to click “Insert” but instead I clicked “Nuke”), not misunderstanding the UI (“I thought ‘nuke’ was like ‘warm up the data’”), then a better alternative to a confirmation dialog is to use a single-item menu button:

It’s not a conventional use of menu buttons, but it’s faster than confirmation dialogs because there’s less mouse movement and no mental re-orientation to a whole new dialog. So there’s less self-documentation, but also less clunk.
Show transductions
"button" (it is a checkbox BTW) effect the user or the process? How does that feature manifest once triggered in this way? Same goes for theGenerate Debug
, what happens? We are all guessing and it makes a mess of answers.