Follow the changes that shape each edition: source coverage, clustering behavior, ranking signals, methodology, and important browsing improvements.
Evidence
Added deterministic evidence dossiers
Story pages now show why a cluster ranked, how coverage developed, and what kind of sources support it without relying on model-written explanations.
- Each story detail page now includes why-ranked reasons, a coverage timeline, source mix, and first/last article timestamps.
- The evidence panel is generated from article metadata and ranking signals, so it still works when LLM enrichment is unavailable.
- Lead and standard story cards can show compact ranking evidence before a reader opens the full source file.
- Article age now follows article publish time first, so older articles discovered later do not look newly published.
Signal quality
Stabilized topics and feed health
The live edition now uses a curated topic taxonomy and clearer feed-health behavior so categories and status signals stay predictable.
- Topic categories no longer drift when optional enrichment suggests new runtime slugs.
- Finance joined the curated taxonomy so market-moving AI infrastructure stories are not forced into unrelated sections.
- High-confidence policy, security, finance, and chips/compute stories get deterministic category guardrails before centroid classification.
- Filtered Hacker News feeds can be empty without counting as source failures, while real fetch errors still surface in health state.
Browsing
Added a newest-first Latest view
Readers can now switch from the ranked front page to a chronological view of the freshest clustered technology stories.
- Use Latest when you want to see what just published, not what has gathered the strongest signal.
- Keep Today's News focused on ranked importance while Latest stays chronological.
- Fresh single-source stories are easier to spot before they gather wider coverage.
Transparency
Published the methodology
SoT now explains how each edition is assembled, how ranking works, and what the system can and cannot reliably infer.
- See why a story appears high in the edition before deciding what to read.
- Understand how source corroboration, freshness, and velocity shape the feed.
- Know the current limits of the product, including source coverage and live cluster volatility.
Sources
Expanded source coverage
More credible publishers and primary sources now contribute to each edition, giving readers stronger corroboration signal across stories.
- Important AI and technology stories are more likely to appear with multiple independent sources.
- Primary-source updates from AI labs and major publishers help reduce blind spots.
- Noisy or unreliable feeds are kept out so the edition stays focused on real news.
Clustering
Improved story grouping and ranking
The event model became more conservative about false merges while improving same-event recall for articles that describe the same incident differently.
- Related articles are more likely to appear under one story instead of scattered duplicates.
- Separate events from the same company are less likely to be merged into one confusing cluster.
- Stories confirmed by multiple sources stand out more clearly from early single-source reports.
- The feed is easier to trust because grouping and ranking behavior now has stronger regression coverage.
Discovery
Made the site easier to discover and share
Stable reader pages became easier for search engines, social previews, and answer engines to understand.
- Shared links now have clearer titles, descriptions, and preview images.
- Search engines see stable reader surfaces instead of short-lived story URLs.
- The site identity is more consistent across browser tabs, previews, and public metadata.
Browsing
Improved navigation, briefings, and digest reliability
The reader experience became easier to scan when topic counts grow and when briefing digests contain more supporting bullets.
- Topics remain reachable even when the active section list gets long.
- Daily and weekly briefings can show more supporting context without overwhelming the front page.
- Older briefing bullets stay readable even when their live story links have expired.
- Headlines and summaries render cleaner when publishers include encoded punctuation.
Briefings
Added daily and weekly briefing digests
The front page gained compact daily and weekly summaries generated from already-computed story clusters.
- Start with a compact summary before scanning the full story ledger.
- Compare today's important stories with a broader weekly view.
- Briefings continue to render even when AI-generated summaries are unavailable.
Product foundation
Launched the first event-first technology news reader
The initial SoT build connected public feeds, local embeddings, a scheduled cache, and a public reader interface around events instead of duplicate articles.
- Read one story per event instead of scanning repeated coverage from every source.
- See which stories are confirmed by multiple outlets and which ones are accelerating.
- Browse by front page, trending stories, topics, and full source lists.
- Use a dense newspaper-style interface built for reading, not decoration.