Ari’s Space*   est. 2026
Skip to content
ari*s space[ a. ]
← Librarytools and plugins / 2026-05-21

A second brain for taste, not for notes

A 4-stage pipeline that distils visual references and my own corpus into reusable creative context.

6 parts / Hover to explore. Click to read.

A 4-stage pipeline that distils visual references and my own corpus into reusable creative context.

External refs    Own finished work
      |                  |
 Trend cards        DNA profiles
      |                  |
 thesis review      anchor review
      |                  |
 separate approved collections
                 |
        new brief -> check fit
External references and patterns in personal work have different provenance and review needs. Keep the originals; approval adds a card to the reference library, not automatically to the public internet.

A saved reference is useful only if you can do something with it later.

I had references and finished work. What I kept having to reconstruct was the judgement between them: why I liked a frame, what I would take from it and what made my own work recognisable.

Then a new brief arrived and I explained my visual sense again. The files were there. The decisions were harder to retrieve.

LENS was my May 2026 answer: distil a reference into a reusable creative position, while keeping the original and its provenance available. The early line about throwing the reference away contradicted the archive stage. Keep the evidence; reuse the distilled card.

Keep the two sources separate

An external reference and a pattern in your own work make different claims. Give them different homes.

Two streams: Trend Cards and DNA ProfilesTwo System 7 windows side by side. Left window titled TREND CARDS shows three stacked card rows with example slugs. Right window titled DNA PROFILES shows two portrait-frame profile rows. A diagonal-hatched column labelled DO NOT MIX separates them.TREND CARDSDirect flashFeels unapologetic, raw, humanCold openWithholds orientationMaximalismDensity as compositional forcemore cards...EXTERNAL REFS → DISTILLEDDO NOT MIXDNA PROFILESPortraiture24 anchored claimsLighting · palette · what I rejectEditing18 anchored claimsRhythm · structure · what I rejectmore profiles...OWN CORPUS → PROFILED
Two streams. Never mash them.

Trend cards start with external material: images, films, reports and frames that caught my attention. Each card makes one specific argument about how a visual choice works.

The card has five parts:

  1. Thesis: one position you can agree or disagree with.
  2. Operative moves: how you would use it in the work.
  3. Failure modes: where that approach breaks down.
  4. Tooling notes: what matters when realising it.
  5. Sources: the reference material behind the claim.

For example, a direct-flash card might propose that the treatment could feel unapologetic, raw and human. That is an aesthetic interpretation to examine against a reference, not a property that every flash photograph automatically has. The operative moves and failures need to make the position usable.

A tag can find the card. The card should then tell you what to do and why. Its question is: what am I seeing in the visual landscape, and how might I use it?

DNA profiles start with my own finished work. They describe recurring choices: lighting, palette, signature moves and things I reject.

Every claim needs an actual anchor. "I favour hard light" needs a named example, such as the portrait set with an unbroken key and no fill. An anchor makes the claim inspectable; it does not turn one example into proof of an eternal preference.

The profile asks what my work does when it feels like mine. Keep that apart from what is fashionable or interesting elsewhere. One follows external references. The other describes a particular person's corpus.

The two-window illustration shows the separation. Its sample cards and claim counts are teaching examples, not a live inventory of my library.

Move through four stages

The design separates collecting, interpretation, acceptance and reuse.

The four stages: Ingest, Distill, HALT Gate, Publish and IndexFour System 7 windows arranged in a horizontal flow, connected by thick black arrows. INGEST leads to DISTILL, which leads to HALT GATE, the largest window styled as an alert dialog with Approve and Reject buttons. HALT GATE leads to PUBLISH and INDEX. An upward annotation arrow below HALT GATE reads: thesis sentence required.INGESTHash. Extract.Archive original.No drift.DISTILLVision model readsper-genre template.Drafts card.HALT GATENothing auto-promotes.Thesis sentence agreed?Anchor example cited?APPROVEREJECTPUBLISH + INDEXDraft → published.Tag index rebuilds.Library compounds.THESIS SENTENCE REQUIRED
Four stages. One human at the gate.

Ingest. Hash the incoming file, extract useful metadata and archive the original durably. Record enough to find the evidence again. A hash can identify bytes; it cannot prevent loss, guarantee permanent provenance or explain why a source matters on its own.

Distil. Use a template appropriate to the material. Photography, video, writing and direction need different questions. The described model-assisted stage drafts a card from the reference; a draft remains a proposal until someone checks it.

HALT gate. I review the draft. A trend card needs a thesis I agree with. A DNA claim needs a real anchor example. Nothing moves to the published collection automatically in this workflow.

Publish and index. Move an approved card into the reusable collection and rebuild its tags. Lighting, palette, mood, era and other axes help a workspace find the relevant material.

Here, publish means approve for this reference library. It does not automatically mean put private references on the public internet. The workflow still needs its actual permissions and implementation; the diagram is an explanation, not an enforced gate.

Review the claim, not just the polish

A fluent card can still say very little. "Authentic, raw, elevated" is not enough by itself. What specific technique creates the proposed effect? Where is the example? When would the approach fail?

For a trend card, I want a thesis I can defend. For a DNA profile, I want the claimed pattern tied back to my work. If either is missing, return the draft for a better account rather than accepting a good-sounding paragraph.

HALT gate close-up: the review dialogA large System 7 modal dialog titled REVIEW DRAFT. Three checklist items: Thesis sentence agreed with a filled checkbox, Anchor example cited with a filled checkbox, and Adjective list rejected with an empty checkbox. Two buttons at the bottom: APPROVE with double border indicating it is the default action, and REJECT with a single border.REVIEW DRAFTBefore this card publishes, confirm:xThesis sentence agreedOne sentence. Concrete noun + concrete technique.xAnchor example citedA real file from the library. A real frame.Adjective list rejected"Authentic, raw, elevated" earns nothing. Cut it.APPROVEREJECT
No card ships without a thesis. No DNA claim ships without an anchor.

The old two-seconds-per-draft line expressed how light I wanted review to feel. It is not a measured review time or a reason to rush an unfamiliar reference. Take the time the claim needs.

Human approval is also fallible. Its value here is that the library records deliberate judgements rather than silently accumulating whatever the model wrote. The evidence remains available when those judgements need revisiting.

Reuse an accepted position

The intended consumer is a workspace that needs creative context. A brief names a few references; those names resolve to approved cards with a thesis and the moves behind it.

That is more useful than making the next process infer the whole argument from a folder of images. It still has to decide whether the card fits the new brief. Retrieval does not confer relevance or permission to use every source.

The consumer contract: one central LENS folder fanning out to four workspace windowsA System 7 folder icon labelled LENS sits at the centre. Thick black arrows with orthogonal routing fan outward to four smaller System 7 windows labelled Workspace A, Workspace B, Workspace C, and Workspace D. Each arrow is labelled with a small reference tag. A caption reads: one source of taste, many surfaces.WORKSPACE ACreative directionbrief lockedWORKSPACE BStyle guidepalette + lightingWORKSPACE CVisual worldtexture + typeWORKSPACE DCampaign briefdirection lockedLENSPUBLISHED CARDS + PROFILEStrend refdna refdna reftrend refONE SOURCE OF TASTE, MANY SURFACES
Distil once. Consume many times.

The four workspace windows illustrate reuse of one published collection. They are not proof of four running integrations. This essay describes the LENS design; the similarly named public ARI-OS feature has its own implementation and limits.

Separating distillation from consumption lets a useful card serve more than one job. A January reference may still matter in August. A profile based on last year's work may still fit if the work has not changed. Review it when the evidence or your taste moves; a saved claim does not stay current merely because it is saved.

Try a position, then ask for its evidence

The fidget combines five axes into a draft sentence. Select a few and inspect whether the resulting position says anything you could actually use.

LENS CONTROL PANEL

Illustration: combine five fixed phrases. No reference analysis or publication runs here.

SELECT AXES
CARD PREVIEW
DRAFT TREND CARD
untitled
-- select axes to draft a thesis --
-- no axes selected --

It joins five fixed pieces of text. It does not inspect a reference, call a model, search your library or publish a card. Those ready-made phrases are illustrative aesthetic readings, not universal rules about lighting, colour or a decade.

A real card still needs the source, operative moves, failure modes and human review. A real DNA profile still needs the personal work that supports it.

Start with one reference. Write the thesis in one sentence. Name the technique, the intended effect and the example that makes you believe it. Then explain where it would stop working.

You now have something another person can question and use. That is the useful memory: a decision with its evidence, ready to meet the next brief.

//

[ comments ]

guests welcome. members show first in the list

no comments yet. start the thread_

1 / 8