The Method
The Daily Edition is an AI-curated daily news portal. Every morning we read 141 sources — think tanks, investigative outlets, specialist publications, and quality newspapers from every continent — score each article on six dimensions, and present the 11–16 best analytical pieces on a single page. Every reader worldwide sees the same edition. No personalisation, no algorithmic feed, no engagement optimisation.
This page is the short version. Two longer pages explain the editorial method and the technical architecture in depth.
What we do, and what we don't
We select, connect, and frame. We read widely, score for explanatory power, and identify the mechanisms connecting the day's most important stories. The three editorial cards on each edition — Force, Theme, Spotlight — name those connections directly. The synthesis across articles is our editorial work.
We do not write articles. Every link goes to the original publisher. We do not summarise articles either. Our card descriptions name what each piece distinctively contributes — its frame, method, or evidence — not the conclusion it reaches. The arguments stay in the articles. Reading the article is how you learn what its author argues.
We do not personalise. Every reader sees the same edition. We do not take money from any source to appear. We do not imitate any journalist's voice. The AI-written text is editorial framing, and we say so plainly.
How it's built
The Daily Edition is not one AI call producing a page. It is a 15-step pipeline. Different models generate, verify, and challenge each other's work, and deterministic code validates the result at every stage. We use three model tiers — Opus for editorial judgment, Sonnet for classification, Haiku for triage — assigned by cognitive type, not just cost.
Every factual claim in a card description is mapped to a specific article and a quoted phrase. Articles whose evidence cannot be verified are removed from the description. Hooks that give away an article's conclusion are automatically retried. A separate model reviews the assembled edition before it reaches you. Twenty-nine deterministic structural checks run before any AI review begins.
The scoring rubric is cross-validated against GPT-5.4 and Gemini 3.1 Pro. Eight of the nine worked examples in the rubric are articles where all three models agreed closely on both the overall score and the dimension scores. The dimension values shown are the medians across the three models. The remaining example — at the bottom of the scale — is a hypothetical anchor, flagged as such so the scorer knows to be cautious.
How this differs from other news products
News aggregators (Google News, Apple News, Flipboard) show what's popular, not what's important. They personalise feeds based on clicks, creating filter bubbles. The Daily Edition shows every reader the same edition, selects for explanatory depth over popularity, and enforces source diversity so one story cannot dominate.
AI summaries (asking ChatGPT "what happened today?") give a confident-sounding answer with no sources, no methodology, and no way to verify the claims. The Daily Edition links every claim in a card description to a specific article from a named publisher. Each claim is mapped to a quoted phrase before publication. The AI's role is curation and framing, not reporting.
Morning briefing newsletters (Morning Brew, The Skimm, Axios AM) summarise headlines in a breezy tone. The Daily Edition's cards identify shared mechanisms across sources, and the continuing/returning system tracks stories across the week.
Think tank newsletters (Brookings, CSIS, CFR daily briefs) are excellent but each covers its own institution's work. The Daily Edition reads across 141 sources regardless of who published them.
Why this matters
Most AI news products work the same way: the AI reads journalism, summarises it, and hands the reader a digest. That assumption produces products that are convenient, legally exposed, and quietly corrosive to the journalism they depend on.
There is another way. AI can help the reader choose what to read, then get out of the way. It's slower, less convenient, more respectful of sources. Products built this way have a different shape. The Daily Edition is one of them.
The full source list, scoring details, and failure modes are documented in How We Read and How We Build.