The fact that kill(all) doesn't remove stuff from the facts database feels counterintuitive to me, as well. But perhaps it is logical as you can do a assume(U>0); without ever defining U (and therefore put U into the scope of the kill(all);.
A more stable rss method for wrstcse
wrstcse: Let two AIs review the package
rtest_wrstcse.mac: Added a missing delimiter
wrstcse: The results of an extensive code review.
wrstcse: Update the wc_distrib2rssparams test
wrstcse: Avoid recalculating the typical value
share/contrib/wrstcse.mac: fix standard deviation calculation in wc_distrib2rssparams