A second brain for taste, not for notes

Creative reference libraries have a rot problem. You save references. You forget them. Your aesthetic exists somewhere between your eyes and your past work. It's not a document anyone can query.

So you go to brief a workspace. A collaborator. A process. And you're re-explaining your visual sense from scratch. Every time.

The fix isn't a better folder structure. It's distillation. Take the reference, extract the principle, throw the reference away.

Two streams

There's a hard line between two things most people mash together.

Trend cards come from external references. Images, films, reports, frames that caught your eye. Each card distils one into an atomic, opinionated document.

The shape is fixed. A thesis sentence first. Something like "direct flash feels unapologetic, raw, and human." Then operative moves: how to deploy it. Then failure modes: how it goes wrong. Then tooling notes. Then sources. That's the whole anatomy.

Not a mood board. Not a folder of saved images. A context block you can hand to any process and trust.

They answer: what's in the current visual landscape, and how do I use it?

DNA profiles come from my own corpus. Not references. My own library of finished work. Each profile maps what I consistently do. Lighting tendencies. Palette bias. What I reject. Signature moves.

Every claim cites a real anchor example. A profile doesn't say "I favour hard light." It says "I favour hard light. Anchor: the portrait set with the unbroken key, no fill." A claim without an anchor is a vibe. A claim with an anchor is a contract.

They answer: what does my work look like when it sounds like me?

Don't mash them. One tracks the world. One tracks the self.

Four stages

The pipeline runs in four stages.

Ingest. References drop in. The system picks them up. Hashes each file. Extracts metadata. Archives the original somewhere durable. Permanent provenance. No drift.

Distill. A vision model reads each reference through a per-genre template. Photography gets a different lens than video. Video gets a different lens than writing. Writing gets a different lens than direction. The model drafts a card.

HALT gate. Nothing promotes automatically. I review every draft. If a trend card doesn't have a thesis sentence I agree with, it doesn't publish. If a DNA claim doesn't cite a real anchor example, it doesn't publish. The model can draft all it wants. I sign off.

Publish and index. An approved card moves from draft to published. A tag taxonomy rebuilds. Every card becomes searchable by axis. By lighting. By palette. By mood. By era. The library compounds rather than piling up.

Stage 3 is where most systems break down. Without a human approval gate, you accumulate plausible-sounding content that doesn't actually represent your taste. The HALT gate isn't a bottleneck. It's the quality contract.

The criteria aren't about polish. They're about whether the card earns its place. Does the thesis sentence say something specific? Does the DNA claim cite a real frame from a real file? Has the model fallen back on adjective lists like "authentic, raw, elevated"? If yes, reject. If the answers are clean, approve. Two seconds per draft. Months of compounding payoff.

One source of truth

Every workspace I run can pull from the same context layer. A brief that needs style direction names a few reference points. Those names resolve to the published cards. One source of taste. Every surface gets the distilled layer, not a re-explanation of it.

This is the part that beats tagging systems. A tag is a label. A card is a position. When a workspace pulls a card, it isn't filtering by keyword. It's loading a thesis sentence and the operative moves that go with it. The taste arrives ready to use.

The compounding effect comes from separating distillation from consumption. Distil once. Consume many times. A card built from a reference in January is still load-bearing in August. A DNA profile built from last year's corpus is still accurate if the work hasn't drifted.

The alternative is re-briefing your aesthetic every time. That doesn't scale. It also means your reference library is a graveyard. Storing things, not answering questions.

Try composing a card from its axes. Select any combination and watch the thesis synthesise:

The fidget is a toy version of what the real cards do. Pick axes, get a thesis. The real cards add anchor examples, failure modes, and the human gate. But the shape is the same. Taste decomposes into positions you can name and recombine.

Taste is a structured artefact, not a vibe. The discipline is the HALT gate. Nothing publishes without a thesis you can defend and an anchor example you can point to.

The system isn't what makes the work consistent. The approval gate is.

//