After a data fix, re-derive every number quoted in prose

Once a pipeline fix shifts upstream counts (here one false candy match removed), rerun the pipeline and diff the gold outputs before touching the write-up. Recompute each number the blog quotes from the fresh export instead of patching only the ones you remember: counts per bowl, sample sizes such as 58 versus 59 candies, subgroup averages (one chocolate-candy average moved from 55.9% to 57.1%), and nearest-neighbor results, since a removed candy changes the swaps. Then grep the post for every old value before committing. Charts read the regenerated JSON and update themselves, but prose and summary tables do not.