Eclecta

Story

Eclecta is a daily newspaper for technology, AI, and the sciences, written by a machine and edited by a taste. Every weekday morning there is a brief; on Fridays a weekly digest; then monthly, quarterly, and yearly reviews — five cadences, one front page. It is a static Astro 5 site, so the reading experience is instant and the whole thing survives even if the machinery behind it is offline.

That machinery is a pipeline called signalpipe. It ingests thousands of verified sources — RSS, Hacker News, Lobsters, Reddit, arXiv, Mastodon trends, Google News, Wikipedia current events, GDELT — then canonicalizes and clusters them, because one story shows up on twenty surfaces and should count as one story. What survives clustering is scored deterministically: consensus times engagement times reputation times recency times topic fit. No model decides what is important. A formula does, and the formula is the same every day.

Only then does the LLM earn its keep. Claude reads the day’s finalists closely and explains what actually matters, and writes the digests under an Economist-school style guide — plain, confident, no hedging. The pipeline writes its output into the repository and pushes; GitHub Actions builds and deploys. The site never depends on an API being up at read time, which is the whole trick: the intelligence happens at build, the reading happens on plain static HTML.

It is tested like something people rely on — unit tests with Vitest, end-to-end flows with Playwright, and accessibility checked with axe. The source and the pipeline are both public under MIT. Read it, run it, fork it — or just read it: eclecta.co.