Sorry I don't have the specific position to post, but this has come up a few times and I wanted to see if anyone else had experienced something similar.

Endgame. I have all but five checkers off, which are stacked on ace point. GNU has one man back, on my five point. Score is 2-2, match to 7. GNU on roll. GNU offers a single game resignation, which I decline. GNU rolls 5-5 immediately, bears off a checker, single game won.

It seems that the AI "knows" the outcome of the roll before the action is offered, correct?

"Gnever trust GNU" is my rule of thumb, as we've had a lot of GNU errors described here over the years.

In the position you're describing, the player with the single checker is a 57-43 favorite on roll. So the real question is "Why is GNU offering a resignation when he's a favorite?"

Even for GNU this is pretty bad, so I have to ask if you've described the position correctly.

Sorry, I didn't describe the position clearly, obviously.
GNU has yet to bear off a checker, and has all checkers safely in its home board, save for the single checker on my 5-point.
Hope that is clearer.

I experience the same thing with XG Mobile (the free version).

It looks like the engine has already rolled when offering the resignation. (Or at least knows the outcome of its next roll like you mentioned.)

At first I was finding that doubly insulting. I was like:
1) Why are you offering me 1 point when I can still mathematically win a gammon!
2) Furthermore, you roll a lucky double to escape the gammon!

Exactly. By itself, it's no big deal, as long as those instances are confined to resignation offers. But, of course, it makes one wonder in certain circumstances, other types of pre-roll decisions , like offers, takes, and drops, could also be dependent upon the AI's "knowledge" of the upcoming roll.

I practice with XG every day and have never noticed anything suspicious about its cube action. I doubles in positions that are doubles, as can be checked by performing rollouts on the position.

Just for the record, I wasn't insinuating that there was something wrong with XG. I just think that when it offers some kind of resignation that isn't confirmed mathematically, it's surely because it has already rolled (although we don't see the dice because they're hidden behind the pane asking for the said resignation). At least on XG Mobile on my phone.

My data set is too small to evaluate this claim

You didn't say what level the cube is at. If it's at 8, that's qualitatively different than if it's at 1, 2 or 4. But assuming it's at 1, 2 or 4, that's a bug IMO.

It seems like what Gnu's doing is:

1. Roll
2. Calculate equities for possible moves
3. Resign if appropriate, wait for response
4. Display dice
5. Move

When it should display the dice right after rolling. Although actually this wouldn't explain everything -- if Gnu needs some time to calculate the equities, it does in fact display the dice before it pauses to think. Maybe it's doing a quick 0-ply eval before deciding to resign and then displaying the dice and doing a full analysis if it chooses not to resign? I don't know, you'd have to ask a Gnu developer.

That's a wildly different situation. The resign thing could easily be a bug where it's just displaying the dice at a weird time. There's no reasonable way this taking and dropping thing could be a bug. It would be malicious coding in an open source code base. And the maliciousness would be easy enough to detect and prove by any user as Gnu would make different cube actions in the same position depending on what roll was coming.

In the latest instance I can remember clearly, cube was still neutral, as I'd maintained a "too good to double" position.

I just thought it very odd that GNU would offer the resignation for a single point, cube neutral, in a 2-2 match to 7, in a situation where a gammon was still a mathematical possibility, (I'm closer to even odds of gammoning there, actually) then roll the double which allowed it to escape the gammon.

I haven't noticed the strange drops/takes either, and have done some rollouts since, so I'm with you...probably just a bug for resignations.