Methodology
AI-assisted, human-edited, verifiable.
Collect (APIs/feeds only)
Data comes from official APIs and RSS/XML feeds — OpenRouter, Vast.ai, Hugging Face Hub, GitHub API, arXiv, confs.tech, and the East-Asian sources listed on the About page. No HTML scraping: when a source has no API, we don't use it. (Note: every date shown is 'verified DD/MM/YYYY'.)
Structure
Collectors write timestamped JSON. Tables and numbers on the site are rendered deterministically from those JSON files — a language model never generates a price, a license, or a benchmark.
Draft (AI)
A language model drafts the narrative: the angle, the verdict, the 'Editor's take'. Generation runs with reasoning disabled to guarantee clean output.
Gate (automatic)
Every indexable page must pass the quality gate: real author present, data dated, primary source linked, at least one element not findable in the top-10 of search results, no near-duplicate in the same batch, valid structured data.
Human layer
A human editor reviews the day's batch, signs the analysis, and decides what is published vs. kept for the newsletter.
Publish
Static build → deployed → sitemap. Only pages that pass the gate appear in the sitemap; anything else stays readable but carries a noindex meta tag.
Why the site is intentionally small
Search engines penalize volume without value. Our answer is structural: the site only contains pages that are current and useful — dated trackers, verified comparisons, and an editor-signed digest. Old, archived, or superseded content is removed rather than hidden, so the crawl budget goes to pages that deserve it. The premium index caps each section at its strongest articles until the domain earns broader authority.
Corrections
Found an error — a price that moved, a license that changed, a link that broke? Contact us. Every correction is dated.
Last updated: September 3, 2026