My impression is that the adversarial mode is significantly closer to deliberation than the anti-offensive model; there's still the underlying goal of "find out what's the case,
I'm not sure. The adversarial mode is much more about convincing someone that something is the case rather than actually trying to find out if that is actually the case.
The anti-offensive mode is generally about burying what is the case; the adversarial mode lets it at least have a day in court.
If you're interested in finding out what is the case, I would argue that the anti-offensive mode doesn't work, the adversarial mode works as a system, but not individually, and the deliberative mode works both as a system and individually. There are situations where the adversarial mode works better; in particular, specialization of labor in research.
A blog post by Alistair Roberts, as curated by Steve Sailer. (Steve's version is shorter and more targeted; the original blog post is the fourth in a series on triggering and suffers for its reliance on the particular issue.)
It seems like a very useful dichotomy, and strongly reminds me of Ask and Guess.