David Pearce’s idea
If part of the information sensitivity you care about is feeling worse cardinally in proportion to the increasing badness of your situation, if you just keep chaining events that make your life worse, you’ll either go below 0 (using any constant ratio of badness-pain increase) or can’t stay proportional, if you decrease in happiness exponentially given linear welfare reductions (e.g. moving a set fraction of the way towards 0 for each unit of welfare reduction, like halving)
Without foreknowing your lifetime budget of misfortune, you can’t budget your happiness properly
Could a function like softmax (or some sequence of repeatedly softmaxing the set of hedonic bliss for grades of goodness/badness) preserve the cardinal and or ordinal relations between options? Maybe just look into RL policy rewards to see what the upsides or downsides of using actual negative reinforcement vs different amounts of positive reinforcement are.