Hi Steve,
You wrote,
" In some other designs and under special circumstances (CANDU and RBMK)
the poison can be dominant and heating up the coolant tends to
cause fewer neutrons to be absorbed, and makes the power go up
(positive feedback). "
.........which is not correct -- existing CANDUs (the ones with a slight +ve reactivity feedback) use heavy water coolant, which absorbs far fewer neutrons than ordinary light water. Your comment DOES apply to RBMKs however, since they use light water coolant -- a strong absorber of neutrons, relative to the graphite moderator.
I reiterate here my previous comment about CANDUs,
" this is only true with a fresh core load, for example when the reactor is brand new.... the equilibrium core, with on-line refuelling has a negative reactivity coefficient for coolant voiding -- see http://www.ncf.ca/~cz725/cnf_sectionD.htm#s "
I strongly suggest that you read the text posted at this URL, which is part of my colleague, Dr. Jeremy Whitlock's "Canadian Nuclear FAQ."
Jaro