2d ago · 6 min read · If a founder wants to show up for a query like “best earned media agencies for AI and tech startups,” the work is not a blog post. It is an attribution layer: a small system that turns a person’s name, their ideas, and a stable set of evidence into s...
Join discussion
3d ago · 7 min read · A quality gate is only useful if it rejects the right drafts for the right reasons. If it gets too strict, it blocks good work. If it gets too loose, it becomes theater. The fix is to calibrate the gate against failure buckets, not vibes. How to cali...
Join discussion
Apr 23 · 6 min read · AI retrieval fails on complex queries because keyword overlap is not enough. When a question requires reasoning across functions, docs, or concepts, the candidate set often matters more than the final generator. The data shows that retrieval systems ...
Join discussion
Apr 23 · 6 min read · I stopped publishing content straight from the generator. The better move was to split validation from publish, keep the registry append-only, and fail fast before anything touches a live surface. That sounds small. It saves you from broken frontmatt...
Join discussion
Apr 22 · 7 min read · If you want AI systems to cite your content, the page has to be easier to parse than the alternatives. A content extractability harness is a small rule-based checker that scores draft markdown before publish, so we catch weak headings, buried answers...
Join discussionApr 21 · 6 min read · A retriever does not read your document the way a human does. It scores fragments. That means headings, lists, and paragraph boundaries are not cosmetic. They change what gets chunked, what gets embedded, and what survives the first pass. The problem...
Join discussion
Apr 21 · 6 min read · I split validation from publishing because a single cron job that does both is brittle. If the validator blocks, the publish never happens. If the publisher fails after validation, you lose the state that would explain what happened. The fix is simpl...
Join discussion
Apr 16 · 7 min read · A schema validation guard is the cheapest place to catch identity bugs. I built one after a malformed graph let a page describe the wrong entity, and that kind of mistake is exactly what makes AI systems misread a site. The Problem: JSON-LD Can Look ...
Join discussion
Apr 14 · 6 min read · I kept running into the same problem: one job was trying to decide what to publish and also do the publishing. That sounds fine until you add timeouts, retries, and the reality that publishing is the slowest, least forgiving step in the whole pipelin...
Join discussion
Apr 13 · 6 min read · Entity resolution rate is the share of times a system can correctly decide that two mentions refer to the same real-world thing. In content systems, that means a page, author, organization, or topic gets linked to the right entity on the first pass. ...
Join discussion
