MDN

A real-time annotation embedded in conversation. The thing that holds a thought in external memory until it becomes internal memory. Not a note-taking system. A cognitive exhaust capture layer.

MDN stands for Meta Dave Note. The name is deliberately personal. It is not a framework. It is not a methodology you adopt. It is a specific operator's working memory made legible during active construction.

The purpose is simple: capture the thought now, without derailing the work. The corpus gets the signal. You get to keep moving. When the thought becomes internalized — when the scaffolding is no longer needed — the MDN stops appearing. The corpus keeps the record. The operator keeps the understanding.

The format is minimal by design. A prefix, a colon, the thought. M-count indicates nesting depth — each additional M means a thought within a thought.

Claude's UI may render the asterisks as bold or italic. That's fine. The *MDN: tag is distinctive enough to serve as an identifier regardless of how it renders.

Dex Jr. sweeps for *MDN: automatically at corpus ingestion. The operator does not need to manage them.

MDNs are not permanent. They exist until they are no longer needed. Internalization is the exit condition.

The system was not designed. It emerged. The behavior existed before the format had a name. The operator was already leaving annotations in conversation — thoughts that were too important to lose, not important enough to stop for.

Naming it made it retrievable. It did not create it. The pattern was already there. The protocol is the recognition, not the invention.

The format stabilized through use, not through design. The operator used it until it worked, then noticed what they were actually doing. The protocol wrote itself.

The extractor auto-assigns categories based on content. They can be adjusted manually. Categories are for routing — they tell the corpus what kind of signal this is.

Paste any document — council review, session log, conversation transcript. The extractor pulls every *MDN: and its variants, auto-categorizes, and lets you sort and export before routing to the corpus.