I have no idea what my reasoning was with this comment, but it makes no sense as a justification, since either way the probability of choosing other branches of the select is non-zero.

[zenhack]

          I have no idea what my reasoning was with this comment, but it makes no sense as a justification, since either way the probability of choosing other branches of the select is non-zero.

I want to stare at this a bit more; I wonder if there's some bookkeeping that I knew we needed to be doing here, but apparently failed at actually ensuring we did? I'll have to have a look when we revisit fixing the estimation problems.

Originally posted by @zenhack in #446 (comment)