The daily Digest, explained

A smaller internet, by design.

Top 5 uses a three-stage AI editorial council to evaluate a bounded set of curated sources. It scores for importance, learning, and wonder, then publishes five. It is not built to maximize watch time, outrage, or endless engagement.

The starting point

A bounded source universe

The general Digest starts with selected publications and feeds, then adds discovery signals from public sources including RSS, Reddit, Hacker News, YouTube, Lobste.rs, and web discovery. It can consider articles, videos, and podcasts. The pool changes with what is published and discovered; it is not a fixed source count.

The constraint

5

The final output is five items. There is no ranked stream below it and no next page designed to keep the session going.

The selection funnel

Three stages, one short edition.

  1. 01

    Filter

    A fast AI pass removes obvious noise such as press releases, routine coverage, weak material, and known duplicates.

  2. 02

    Score

    The survivors are scored through three editorial lenses, with source and cross-reference signals as supporting evidence.

  3. 03

    Select

    A final AI pass chooses five as a collection and writes the short note explaining why each item made the cut.

What the scoring asks

Not just what is loud.

40%

Learning

Will this help someone understand something better?

30%

Need to know

Would missing this leave someone out of the loop?

30%

Wonder

Is this the kind of thing you would send to a friend?

Guardrails and records

Fresh, distinct, inspectable.

The final stage limits a source to one pick, requires at least three fresh items, and checks for near-duplicate stories within the edition and against recent editions. It also applies a domain cap and limits AI-topic concentration.

Each run writes structured records for ingestion, filtering, scoring, selection, and the journey of each item through the system. These are operational records for diagnosing and improving the process—not a public live dashboard.

Where automation can fail

AI can miss context, overweight a weak signal, or make a poor trade-off between otherwise good choices.

That is a limitation of the system, not something we hide behind an editorial claim. If a pick is wrong, repetitive, broken, or simply misses the point, tell us.

Send product feedback

Personalization boundary

One public edition for everyone.

The public Digest does not use an individual account profile, clicks, reading history, or item feedback to rank the day's five. Optional Custom Top 5 topics are separate: they can use the interests a reader supplies and feedback on that custom topic.

What we do not do

  • No infinite feed to keep you scrolling.
  • No engagement-based ranking for the general Digest.
  • No paid placement presented as an editorial pick.
  • No claim that a human reviewed an edition when that did not happen.