Three-layer organization: primary topic (one per article, for ranking and
brief balance) + grouping tags (1-4 per article from a controlled vocabulary,
the organic "wandering" axis) + tonal flavor.
- taxonomy: add technology + learning topics; 4 calm tag families
(Discovery & Wonder, People & Kindness, Solutions & Progress, Mind & Craft)
defined in code, not the DB; ALLOWED_TAGS union + coerce_tags validation.
- db: article_tags(article_id, tag) join table + tag index.
- llm: tags added to the classifier json_schema (enum-constrained, maxItems 4)
and system prompt; normalize_scores coerces tags; upsert_article_score
replaces a row's tags atomically on every (re)classification.
- queries: feed gains a tag filter and exposes tags via group_concat; tag_counts.
- api: Article.tags, feed tag param, and /api/families with per-tag counts.
- tests: coerce/normalize/upsert/tag-filter/reclassify-replace/tag_counts +
/api/families. 99 passing.
Corpus reclassify (re-tag + new primary topics) runs separately against the
local LLM. Frontend (B2) pairs with this; the live site is unchanged until then.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The grid stays typographic; the hero is the one intentional visual slot. At
brief-build time we fetch a hero-quality image for the daily five that lack one:
- enrich.py reads ONLY a page's <head> og:image/twitter:image and stores just
the URL (never the body).
- SSRF-guarded: http(s) only, 6s timeout, 300KB cap, <=3 manual redirects each
re-validated, and hosts rejected if any resolved address is private, loopback,
link-local, multicast, reserved, or unspecified.
- image_checked_at column caches success AND failure, so an article is never
retried forever.
- Wired into build-brief and cycle (brief items only, only if image missing and
unchecked). Everything else stays metadata-only.
- Verified live: today's five all carry images (feed + enriched).
Tests: og:image parser, head-only scope, IP guard across internal ranges, and
enrich success + failure-caching (85 total).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Add source health columns (last_success_at, last_error_at, last_error,
consecutive_failures, review_flag, review_reason) via SCHEMA + migration.
- poll_source maintains them: success resets the failure streak and records the
success time; failure increments it and stores the latest error.
- review_sources() flags active sources that are stale, repeatedly failing,
low-acceptance, duplicate-heavy, or doom-skewed (high cortisol/ragebait) over
a recent window. It is purely advisory: it sets review_flag/review_reason and
never changes the active column (human stays in the loop), clearing the flag
when a source recovers.
- CLI review-sources; cycle runs it as a final step (--no-review to skip);
source-report shows a review line for flagged feeds.
- Tests: healthy/failing/stale/low-acceptance/recovery and never-deactivates.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- New source_candidates staging table (status suggested/quarantined/rejected/
promoted, preview_json snapshot) so untrusted/suggested feeds stay out of the
real ingestion path until reviewed.
- sources.py: save_candidate (re-preview never revives a curator's rejection),
list_candidates, reject_candidate, promote_candidate (copies into sources,
inactive by default — active on approval; never automatic).
- CLI: suggest-source / list-candidates / promote-candidate / reject-candidate.
- API: read-only GET /api/candidates (writes stay CLI-only — no unauthenticated
public write surface yet).
- Fix deprecated ElementTree truth-value test in _parse_rss.
- Tests: candidate lifecycle (save/list/promote/reject, status preservation,
name derivation) — 51 total.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- LocalModelClient.embed() calls the OpenAI-compatible /embeddings endpoint
(local nomic model); base_url shared with chat, model via GOODNEWS_EMBED_MODEL.
- New article_embeddings table and articles.duplicate_of column (+ migration).
- dedup module: embeds missing articles, clusters near-identical stories within
a date window by cosine similarity (pure-stdlib, vectors normalised once), and
marks all but the highest-ranked member of each cluster as a duplicate.
- 'dedup' CLI command; cycle now runs poll -> classify -> dedup -> brief.
- Feed and brief queries hide duplicates, so a story carried by multiple
outlets shows once.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- New taxonomy module: single source of truth for 6 topics x 5 flavors,
shared by the LLM response schema (enum-constrained) and validation.
- Classifier now assigns one topic + one flavor per article; json_schema
enums force valid values, with coercion as a safety net.
- article_scores gains topic/flavor columns via an idempotent migration.
- New 'list-category' command to browse by topic and/or flavor, ranked by
composite score.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
thejayman77