The hardest part of an open source poll would be anonymization. Demographic information in some areas could be enough to come pretty close to uniquely identifying an individual.

The rest of it: Likely Voter Models, Demographic Weighting, etc would be pretty easy to have open source and reproducible once you've solved the issues.

Yes anonymization is the problem I was thinking of.

Open source would not be nearly as impressive as reproducible.

Yes, although is reproducibility even possible in polls? People's opinions change on a whim. In fact people's opinions can change simply because they were asked a question so the answer might be different if asked the question again,

I guess you could conduct the same poll with multiple sets of people and see how the results compare, but in a way that's what poll aggregators do.

Isn't the reasonable definition of reproducible for a poll having overlapping margins of error? A useful poll should do that.

The result is qualified, comparisons of results shouldn't ignore that.

OK, we've added "Aggregators" to the story title to clarify.