There is a bug in the mooclet API code where priors are being counted twice before being sampled in ts_configurable experiment. We have not played around with priors much until recently, when this was discovered. The effect would be a thumb on the scale for any priors higher than the default.
The fix is to remove a couple lines where this is mistakenly happening in the mooclet code, and to write some unit tests around this stuff to prove that it's working correctly.
There is a bug in the mooclet API code where priors are being counted twice before being sampled in ts_configurable experiment. We have not played around with priors much until recently, when this was discovered. The effect would be a thumb on the scale for any priors higher than the default.
The fix is to remove a couple lines where this is mistakenly happening in the mooclet code, and to write some unit tests around this stuff to prove that it's working correctly.