Other has 5, 5, 7, 5 x 8, 9, 10, 10, 11. I'll try and do an average provision cost for bribery later today, by assuming that higher provision cost = better choice.
I got sidetracked. As nobody posted their deck that bribery screwed over, I'll just go with the two extremes I posted. Spoiler tags don't seem to work, so I'll leave out the detailed working.
One extreme is an enslave deck with minimal units. Unique units are 2 x 5, 7, 5 x 8, 9, 2 x 10, 11. For this unit composition, the median highest provision cost is 10. There's a 62% (816 out of 1320 unique draws) chance of having at least one 10+ option among the 3 cards. The average highest provision cost among the 3 choices is 9.61
Other extreme was the NR deck with almost all units, half of them 4-5. Unique units are 2 x 4, 7 x 5, 6, 2 x 7, 8, 2 x 9, 2 x 11, 12.
This time, the median is 9. 44% (2166 out of 4896 unique draws) chance of having one of the 11+ options available. The average highest provision cost is 9.18
Most deck compositions look like they'll see bribery pull a card with a 9-10 provision on average. Bribery is probably too cheap currently, as you're using an 8 provision card to consistently play a 10+ provision card, and the likely worst case is a 7 provision that's not super helpful for the board state/your deck. It probably should be at least a 9 (if you make it a little less than the average it pulls due to the risk of unhelpful abilities) or a 10-11 (if you make it a little higher because of the benefit of it being a tactic & easier to tutor/repeat and the utility of getting 3 potential options). I'd say it should be at least a 10. It certainly seems an auto-include in every single NG deck at its current cost.
Of course, this assumes the bribery draw is randomly from the pool of unique units. I haven't seen evidence it isn't, my own use of the card in arena suggests it really is random, and selective memory & anecdotal data based on the times our opponent got a really good draw isn't good data. I'd still be curious to see the results of trying bribery a few hundred times and recording the options, but not enough to do the testing myself.