check: the eval rows re-measured on the subjects left, and checked

readme: the 49/57 in the core sample table is OpenSPP's

narrow: one corpus, and it is Odoo

tidy: the counts the corpus outgrew, and a test CI would have failed

fieldtest: somebody else already decided, and a claim that was not true

recall: cluster the misses, and check every version claim

check: read the file, instead of only asking a model about it

eval: measure nexa check, and none of the 25 reviews passed

python: a fourth stack, so a plain .py file gets a real answer

eval: measure whether a rule can be retrieved, instead of inferring it

eval: measure verdict distribution over repeated runs