← Back to /admin

2026-04-24-blog-tags.md

audits/2026-04-24-blog-tags.md · 2,808 bytes

Audit — Blog Topic Tags — 2026-04-24

Target. 15 random tagged blog posts from blog/metadata.json.

Questions. Do topic tags reflect argumentative content (per L29), or are they surface-keyword matches?

Verdict. 8/13 valid posts honestly tagged. 3 S7-style surface-match drifts found. 2 empty stubs are still carrying tags — tagging pipeline bug.

Summary

Verdict Count
honest 8
mis-tagged (includes surface-match) 3
under-tagged 2
over-tagged 0
data-issue (empty stub) 2

Flagged posts

Patterns

1. bitcoin-protocol catches the keyword "blockchain". Any post that mentions blockchain once in passing gets tagged. S7's info-theory failure repeated in a different pillar.

2. identity-history catches the word "history". Both geographical-aspects-cybercrime and arguably blake-and-milton drift via this path.

3. Empty stubs carry tags. Pipeline bug: posts with word_count < 200 should not receive tags (they have no structural content to classify).

4. philosophy is under-applied. Posts about fallacies, meaning, metaphysics go uncategorised as philosophy.

Recommendations (each a scripts/* edit)

Overall pass rate (valid posts): 8/13 (62%). The classifier has repeatable surface-match failure modes. Re-run quarterly. Fix pipeline before next re-tag.