This was nice to read :)

I haven't thought about this much, but wouldn't it be theoretically optimal in some sense for people to only vote for one option because there are no increasing marginal costs to subsequent votes to the same candidate? Which is solved by quadratic voting, but I am guessing you considered and rejected it, so quite likely there is something I am missing.

OK makes sense thanks! As I understand GiveWell is interested in/working on incorporating distributions rather than point estimates in general so hopefully discount rates fit in with that work (I don't have much context on what GiveWell's plans actually are).

I am a bit surprised you (seemingly) missed my post https://forum.effectivealtruism.org/posts/JvEHiKWWwtvT3YanA/givewell-misuses-discount-rates given the relevance is so clear from the title. I haven't thought about this more since writing it for the competition, but have no reason to think I was wrong then. The simple idea I propose in that piece is that GiveWell should use a probability distribution over possible discount rates, and that this will meaningfully change the cost-effectiveness of deworming charities.

Excellent, I hope this goes really well!

It would seem better and less confusing to me if the good folk at GWWC (@GraceAdams) add a field in the donation interface with a dropdown menu of which specific CE-incubated charity one wants to donate to (with the default being an agnostic donation for GWWC to distribute between CE charities as they see fit). 

Thanks for writing this! I agree making this decision very carefully seems well warranted. One minor thing - you generously offer for people to contact you to ask about out-of-scope things (delivery research). I can well imagine someone finding this post in several years' time and wanting to contact you, but I'm not sure if this anonymous account will exist/be monitored then; one possible solution is to link to a google form that would go to your email (without revealing what your name/email is).

In your conclusion with the casino analogy, I thought you were going to make an explore-exploit argument, eg:

Once we have played all the different games on offer for a while, we will both get better at each game, and work out which games are most profitable to play. Therefore, we should play lots of games to start with to maximise the information gained.

I think without this argument, the conclusion doesn't follow. If we are not planning to later narrow down on the best choices once we have learnt more, the case for spreading our resources now seems a lot less strong to me.

Ah good point that we cannot in general swap the order of the expectation operator and an inverse. For scenarios where the cost is fixed, taking the inverse would be fine, but if both the cost and the impact are variable, then yes it becomes harder, and less meaningful I think if the amount of impact could be 0.

Thanks, that was useful and interesting to read, and makes more sense now!

Great you are looking at more direct implementations for increased efficiency, I think my intuition is it would be less hard than you make out, but of course I haven't seen the codebase so your intuition is more reliable. For the different eras, this would make it a bit harder, but the pmf is piecewise continuous over time, so I think it should still be fine. Keen to see future versions of this! :)