Mostly AI systems. When something doesn't add up, I keep digging until it does, then write it down. Now and then I build the thing I wished existed.
Just published: ArticleLog 002, on what changes when a cache comes with a promiseSmall things I noticed and followed further than I meant to. Measured where I could, labelled where I couldn't.
OpenAI's prompt cache went from "up to" to "at least". A 12-hour write pays only if you come back twice after a 30-minute break. And "forever" in Codex is probably a billing term, not a storage one.
Why long-running agents lose their cache, not their history—and when keeping it warm is worth paying for. An exploratory simulation with public coding-agent traces.
Every entry carries an evidence labelExploratory · Measured · Documentation analysis · Opinion
Things I wished existed, so I built a rough version. Rough on purpose. Mostly.
