Recursive mirror

I have five years of raw journal entries. Unfiltered. No audience in mind. Just me talking to myself about games, music production, philosophy, frustration, and the occasional boat gender question.[1]

This site exists because I wanted to share some of that thinking. But there's a transformation involved. Raw thought → readable prose. And I use AI to help with that transformation.

So here's the meta question: Can an AI analyze my raw voice well enough to help preserve it during the transformation?

Feeding journals to Claude

I pointed Claude at my journal directory[2] and asked it to identify patterns. Then compare those patterns to what's already on this site.

The result is a reference document - a catalog of my voice patterns with examples, transformation rules, and guidelines for AI collaboration.[3] A style guide, but for something harder to pin down than fonts and colors.

Patterns I didn't know I had

Some patterns I knew. Fragment punches. Self-questioning. The "Please." at the end of earnest requests. Others I didn't consciously recognize:

• Staccato rhythm - Short. Short. Longer flowing thought. Short.
• Profanity as punctuation - Not filler. Emphasis. ("This is all fucking insane but I have a mortgage.")
• Rhetorical escalation - Questions that answer themselves through absurdity.
• Call to action closers - "Figure it out." "Let's get to work."

The weird part

This post was written using that analysis. The AI knows what patterns to preserve, what to add, what to avoid. It's reading my journals, reading its own analysis of my journals, then writing in my voice based on that analysis.

Recursive mirror.

Does it work? You're reading the output. The ideas are mine - the phrasing is a collaboration. Not sure that's different from working with any editor who knows your style.[4]

The second opinion

Then I did something obvious in retrospect: asked a different AI session to review the analysis against source material the first session never saw. Work documents, sprint retros, technical walkthroughs.

Turns out the original analysis captured a refined voice. The aspirational version. What emerges when writing for an audience.

Raw work notes are messier. More tentative. And they have patterns the polished journals don't:

• Absurdist narration - Gallows humor for unbearable code. ("yes, really", shrug emoji, dread built into document layout)
• Scale humor - "we did one (1) log! only 100 more, and 100 more, and so on"
• Quantified frustration - "it took 1.5h of 2 devs' time" — numbers make waste harder to argue with.
• Empathetic observation - Ethnographic attention to humans in systems, not just the systems.

The mirror shows you at your best, not your baseline.

Which might be fine. Maybe the transformation is supposed to elevate. But it's worth knowing where the gaps are.

Buried project ideas

Beyond the obvious outputs (this post, the reference doc), the process surfaced project ideas I'd forgotten about. A debug-themed social deduction game. A "cynical originality checker" that shows you how unoriginal your ideas are. A keyboard-to-synth that turns typing patterns into melodies.

Five years of notes. Plenty of material. Some of it might actually become something.

Let's get to work.