Work log: Polymathic — April 1, 2026

What shipped today

The main work today was filling in missing tag metadata on 51 older published posts (GH-242). These posts — mostly imported content from pre-2026 — had no tags: field at all, which matters more now that every published post is automatically voiced and ingested into the podcast feed. Apple Podcasts and Spotify use episode tags for topic-based discovery, so untagged posts are essentially invisible to search. The fix was straightforward: read each post’s content, apply 3–5 relevant tags drawn from the existing tag vocabulary, and commit. The change touched a broad cross-section of the archive — everything from Rails helper posts to management essays to early ChatGPT commentary.

The work revealed a latent bug in the pre-commit hook that became impossible to ignore at this scale. The hook generates TTS audio for every staged markdown file in content/posts/, which is fine when you’re committing one or two new posts. But when 51 posts are staged with only frontmatter changes, the hook tries to call OpenAI TTS for all of them — burning time and API budget on posts that already have audio or that don’t need regeneration. The commit had to be forced through with SKIP_AUDIO=1 to avoid the runaway generation. That gap is now filed as GH-247: the hook needs to diff only the post body and skip audio generation when only metadata changed.

Completed

• GH-242 — Fill missing tags on 53 published posts (51 of 53 posts updated; 2 already had tags per the grep scope)

Carry-over

• GH-247 — Pre-commit hook triggers audio generation on frontmatter-only changes. Needs a body-diff check before dispatching TTS. This is ready-for-prep — needs an approach spec before dev.
• GH-246 — paulos blog idea command returning Authexis API 404. Low urgency but annoying; blocks quick idea capture from CLI.
• GH-243 — 70 legacy drafts in content/_drafts/ need triage/archive. Large but mechanical.
• GH-112 — Date display accuracy on /blog/ — still needs-clarification.
• GH-79 — Per-episode itunes:image in podcast feed — backlog.

Risks

The pre-commit audio hook is now a reliability problem at scale. Any bulk metadata edit to content/posts/ will hit the same issue. Until GH-247 is fixed, bulk frontmatter changes need SKIP_AUDIO=1. This is worth noting: the tag-filling work also means 51 posts now lack audio_url and will trigger TTS on the next individual commit that touches them — unless the hook correctly skips posts that already have audio. Verify the idempotency logic in the hook before committing anything that touches previously-voiced posts.

Flags and watch-outs

• SKIP_AUDIO=1 was used for today’s commit. Posts that were modified today may not have audio yet if they lacked it before. The hook should catch them on next commit.
• The two generated SASS resource files (resources/_gen/assets/sass/...) are dirty but ignored — these are Hugo build artifacts, not meaningful changes.

Next session

1. Prep GH-247 — Write the fix spec: detect frontmatter-only changes by diffing the post body (strip frontmatter block, compare hash), skip audio generation if body is unchanged. Assign to dev.
2. Check audio coverage — Run grep -L 'audio_url' content/posts/*.md after a test commit to verify the hook is picking up the newly-tagged posts correctly and not double-triggering on posts that already have audio.
3. Triage GH-243 — Start the content/_drafts/ cleanup. Even a quick pass to identify which files are worth keeping would help.