About Sparklet

Sparklet is a learning feed. Instead of an endless scroll you finish feeling worse about, you get short, specific, sourced facts — each one backed by a full article you can read if it catches your interest.

How a card is made

Cards are drafted with the help of large language models, then put through automated checks before anything is published. We say this plainly because you deserve to know how what you are reading was produced.

Every cited link is fetched. A card whose source URL is dead, or which sits behind a paywall a reader cannot get past, is not published.
A second model checks the first. The provider that did not write a card reads the actual text of its sources and judges whether they support its specific claims — names, numbers, dates. Contradicted cards are held for review rather than published.
Near-duplicates are dropped. New cards are compared against everything already published so the feed does not repeat itself in different words.
Publishers are derived from the URL. The name shown on a source is taken from the link's own domain, never from a label a model wrote — so an attribution can never claim an institution the link does not actually go to.

Corrections

Automated checks catch a lot and will never catch everything. Every card has a report control, and reported cards are pulled from the feed for human review. If you spot something wrong, telling us is the fastest way to get it fixed — get in touch.

Start reading

Browse every topic or jump straight into the feed. No account needed to read.