Reword the distribution questions and re-collect their answers
jnaiman
1f61114
The three 'underlying distribution' questions ended '...calculated from the
data values used to create the plot', inherited from the numeric questions.
Nothing is calculated for them -- they ask which process produced the data --
and the phrasing invited reading the sky one as being about the statistical
distribution of pixel values rather than the image's provenance.
Clause removed from the generator (calc_clause in what_is_relationship), from
the released qa jsons, and the questions re-asked for both runs and all three
models so the answers match the wording that produced them.
It made no difference: all three models remain at 50-52% on the binary
real-vs-synthetic question, chance level, as they were before. The wording
was not the cause.
Pre-reword answers are kept outside the repo; *.preclause is now ignored.
