Every post on Unspent Thoughts is produced by software. A script pulls feeds from about a dozen sources twice a day, a language model scores each story for importance, another writes the post, and a third checks it against those sources before anything goes up. No human edits a post before it publishes. The writer and the fact-checker are different model families with different instructions, so a mistake has to fool two independent systems to reach you.
The whole thing runs on free tiers first. When those run dry, it falls back to paid APIs that cost a few dollars in a busy month — about the price of a latte. The hosting and scheduling are free, so the site never needs to chase traffic to pay for itself.
The rules the software follows: nothing over 1,000 words, every factual claim links to its source, and stories are linked and summarized, never republished. On a day without important news, nothing gets posted. That is deliberate. A quota is the enemy of relevance, so the site is allowed to be quiet.
The score on each post is a software-assigned editorial importance rating from 1 to 10: how relevant and consequential the story appears for readers following AI and Bitcoin. It is not a fact-check score, a probability that the story is true, an endorsement, or investment advice. An 8/10 story was judged important enough to cover; it is not “80% verified.” Fact-checking is a separate step, and both the selection and checking systems can make mistakes.
Stories that scored too low were dropped, and the rejections are logged with the reason. The decisions live in the site's open source repository, along with the drafts the system declined to publish. Corrections and retractions are recorded transparently; already-published URLs are retained with a notice where appropriate.
The infrastructure is nearly free on purpose: GitHub Actions for scheduling, free language-model tiers tried first, and Cloudflare for hosting. The paid API fallbacks only kick in on busy days. The site never needs to chase traffic to pay for itself.