The article is the summary. The log is the thing.
What a log carries that an article doesn't
- Decisions and their reasons. Which alternatives were considered, why they lost.
- False starts. The wrong approach that taught something. Almost never in the final version.
- Timing. When ideas arrived, how long they sat, what pushed them forward.
- Multi-voice. A conversation has two people. An article usually has one.
- Raw phrasings. The scrappy first sentence often captures the idea better than the polished rewrite.
The shape
Every article gets an optional "log" link. Flat, searchable record of conversations, edits, and decisions that produced it. Private by default, published when comfortable. The article is the entry point; the log is there for depth.
Progressive disclosure: top layer is the claim, middle is the reasoning, bottom is the conversation that produced both. Each optional, each for a different reader.
AI softens the friction
Most self-censoring in a live log comes from the gap between a half-formed thought and something you'd let strangers read. AI collapses that gap. Capture raw, publish presentable, at the speed of the original thought. The fast private thought can now safely reach others.
And there's a second reader. Not many humans will sit through a full log. But AIs will. They'll read all of it, index it, connect it across logs, extract patterns a skim would miss. The log isn't a vanity artifact; it's a corpus. Publishing it is partly for the one person who wants depth, partly for the machine readers quietly becoming the majority audience for technical material.
Status
Shape is clear: a blog where the article is the tip, the log is readable on request. Subscribers and architecture prompts sit next to this when they're ready.