Oftentimes as a dictatorial regime enters its waning days you face a choice as to whether or not to offer the leaders guarantees if they agree to give up power. The downside is that this seems to create bad incentives -- the bad actors get away scott-free when it would be better for evil to be punished. The upside is that with guarantees they may actually give up power, whereas regime leaders who know that if they give up power they'll be treated harshly will probably hold on to power with the utmost brutality. Timothy Burke, thinking of Zimbabwe, says this kind of thinking is bunk:
So even if we understand people like Mugabe and his inner circle as calculating, incentive-evaluating, rational deciders, I think there is every reason for them to laugh behind closed doors at the hubris of the experts and activists, whatever the latest policy nostrum on tribunals, interventions, sanctions, golden parachutes or so on might be. Because what anyone outside of the rarified settings where generic 12-point plans for peacemaking and incentivizing prosecutions for genocide are composed knows is that every such action is and will be sui generis. The sand castles that the experts build today around one case will be washed away by the tides of history in short order. What happened in the end to Charles Taylor or Auguste Pinochet or Saddam Hussein or Slobodan Milosevic has little implication for tomorrow’s dictator and mass murderer. Because the people who play with constructing the machinery of incentive aspire to a kind of reliable managerial authority that they will never have, they are writing blank checks that no one will ever cash. Whether or not someone like Robert Mugabe dies peacefully in his bed, lives out his last years far from his home country, ends up in a pleasant prison while the United Nations dithers for a decade over his fate, is shot by an up-and-coming rival, or ends up torn to shreds by a mob is a matter of particular circumstance. That’s probably something most authoritarians know already, having ridden the vissitudes of history as far as they have.
I guess that seems plausible enough, though I find it pretty unsatisfying as a conclusion and though perhaps it's just technocratic hubris, I'd like to try to see some data.