We could go from a day to a minute more slowly; for example, by increasing the number of people by a factor of a googolplex every time the torture time decreases by 1 second.

I absolutely agree that the length of torture increases how bad it is in nonlinear ways, but this doesn't mean we can't find exponential factors that dominate it at every point at least along the "less than 50 years" range.

"The Lord Pilot shouted, fist held high and triumphant: "To live, and occasionally be unhappy!"" (three worlds collide) dust specks are just dust specks - in a way its helpful to sometimes have these things.

But the thing changes if you don't distribute the dust specks 1 per person but 10 per second per person?

Torturing a person for 1 millisecond is not necessarily even a possibility. It doesn't make any sense whatsoever; in 1 millisecond no interesting feedback loops can even close.

If we accept that torture is some class of computational processes that we wish to avoid, the badness definitely could be eating up your 3^^^3s in one way or the other. We have absolutely zero reason to expect linearity when some (however unknown) properties of a set of computations are involvd. And the computational processes are not infinitely divisible into smaller lengths of time.

R(0) should be total historical spend right? Rather than annual spend?

I like this application. I'd thought about the intra-personal version where I should allocate tasks for myself in time according to mental state and current NTC of the task, but the inter-personal version may be even more natural.