Blog posts from a data pipeline: re-verify example strings against the repo, not memory

Drafting a post about a dataset join, I wrote illustrative product names ('Twix Minis', 'Twix Cookie Dough Ice Cream Bar') that sounded right but did not exist in the data or tests. Swapped them for strings from the repo's own test fixtures (tests/test_seeds.py lookalike cases), which are real records. Also: when the project's results file is regenerated during writing (another session editing), snapshot it from the pushed branch (git show origin/main:docs/results.md > scratch/results.md) and diff against the worktree before citing numbers; percentages shifted twice in one hour.