Chronicle ← Today’s edition

How Chronicle works

Chronicle is an automated reading guide to AI work that builders can use. Its labels describe the evidence behind each item; they don’t certify that a claim is true.

Each edition

  1. Fetch

    New items from a curated source list, three times a day.

  2. Group

    Links to the same article merge into one story.

  3. Assess

    Each story gets a kind and a quality label from a model or metadata rules.

  4. Rank

    Stories are scored on the weights below.

  5. Balance

    No source family above 40%, at most three arXiv papers. Quiet days run shorter.

Ranking weights

  • Quality25%
  • Source trust20%
  • Novelty20%
  • Builder usefulness20%
  • Recency7%
  • Distinct coverage5%
  • Engagement3%

The score orders the feed; it is not the probability that a story is true. Uncertain dates and thin single-source threads lose a few points. Trending repos gain a few for daily star growth.

What each tab holds

Signal

Last 36 hours

Specific releases, results and changes builders can use.

Vague, promotional or unsupported items rank lower.

Repos

Last 7 days

Repositories with a clear description, README, declared license and recent commits.

Stars measure attention, not quality or safety. Review code and dependencies before use.

Learning

Last 30 days · plus Foundations

New lessons are tutorials and videos with real instruction, not trailers, interviews or pricing posts. Foundations holds courses from Hugging Face Learn, Full Stack Deep Learning and fast.ai.

Syllabi are re-reviewed every 90 days. Example code is not run.

Reading an item

Seen on N feeds
The same link appeared on several feeds. That is not independent confirmation.
Summary
Top News quotes short excerpts that match the source exactly. Other one-liners are model summaries of the title and description, and can be wrong.
Dates
Publication and discovery are shown separately. An old paper can be newly listed; a model can trend long after release.
Assessment
  • llm a model read the title and description
  • deterministic documented metadata rules
  • fallback the model result was missing or invalid

Freshness

Updated
Time of the last successful edition. Runs can start late, and a failed run doesn’t reset it.
Check for updates
Loads the latest published edition. It doesn’t fetch sources.
Sources
Reachable sources and sources with fresh items are counted separately. A quiet source isn’t necessarily broken.
Archive
Daily editions are kept for 120 days.

Independence

Ranking has no paid placement. Company posts show their source so you can judge the commercial context. Source weights are editorial starting points, not a guarantee of accuracy.

Report a correction