Method

How a record gets written, what has been read, and everything the research decided not to assert.

How much we know

The corpus is uneven on purpose. Rather than hide that, every Aesthetic reports two separate facts: how much outside research stands behind it, and how much build guidance it holds.

Research status

Partly sourced
57 of 73 — at least one claim has a source attached. The rest are graded by our research brief, carried over from our catalogue entry, or our reading, each labelled where it appears.
Graded, not yet sourced
11 of 73 — a research brief graded some claims documented, but no source is attached yet. A grade is not a source.
Not yet researched
5 of 73 — no source attached, and nothing graded by a research brief.

Build guidance

A separate count, not a rung on the same ladder. 73 of 73 Aesthetics hold build guidance: rules you could make something from. Where a research ledger exists it is counted in research dimensions, as in “guidance recorded in 5 of 8 research dimensions”. It says what usable material is recorded, never how well it is evidenced: an Aesthetic can be not yet researched and still carry guidance from our catalogue entry and our own decisions, each labelled where it appears.

Depth describes our research, not the aesthetic. A thinly documented language is not a lesser one. Punk Zine is not a lesser aesthetic than Swiss Style. We have just not done the reading.

Research standards

How a record gets written, in the order it happens:

  1. 01preflight
  2. 02context assembly
  3. 03research
  4. 04mechanical QA
  5. 05triggered escalation
  6. 06independent audit
  7. 07adjudication

A whole run, discarded An early parallel run produced 50 sources and zero opened-and-matched — 84 claims with nothing behind any of them. It was thrown away rather than ingested, and the requirement became structural: a run that fails it does not enter the corpus at any stage.

AI-assisted research. Aesthetary’s research process uses AI models, primarily Claude, to help retrieve, structure, compare and audit research. Claims are tracked separately from their sources, evidence and Aesthetary’s own interpretations, so that model output is not treated as evidence by itself.

Sources have to be opened

The rule that does the most work is the dullest: a source counts when someone opened it and matched it to the claim it supports. An early parallel run produced 50 sources and zero opened-and-matched — 84 claims with no support behind any of them. That run was discarded rather than ingested, and the requirement is now structural.

Evidence that generalises, and evidence that does not

One poster showing a three-size type scale supports the pattern and not the number. So evidence carries a mode: whether it attests a single occurrence or licenses a general claim. A figure derived from one instance gets removed from anything an agent would read as binding — that has already happened to this corpus more than once, and those withdrawals are recorded in the changelog.

Adjudication is a separate step

Research proposes; adjudication decides. A claim can survive the reading and still be demoted because the evidence does not reach as far as the wording did. Nothing is published because it was produced.

Sources

The 73 Aesthetics in the collection cite 1,042 sources between them, and each Aesthetic's data file lists its own. The 50 below have their publisher and licence recorded here.

14 archive · 12 primary work · 9 press · 8 academic · 5 community · 2 external corpus

What the research changed

15 editorial decisions about Aesthetics in the collection: what was concluded, and at what confidence.

Withheld 3 of 15

Decisions NOT to assert. A reference that only records what it added is telling you half the story.

  • Scandinavian Minimalism Intent · low confidence

    Held back as provisional: the definition leans on its two neighbours rather than on anything internal to the style.

  • Streetwear Intent · low confidence

    Held back: every source retrieved was commercial or industry press, with no museum, university or peer-reviewed source.

  • Victorian Intent · medium confidence

    Held back as too broad: the intent covers Victorian commercial letterpress printing, not Victorian architecture, interiors, dress or painting.

Asserted 12

  • Art Deco Intent · high confidence

    The name postdates the work by roughly forty years.

  • Editorial Modernism Intent · medium confidence

    The contrast with Swiss Style is the only firm evidence.

  • Japanese Minimalism Intent · medium confidence

    “Ma” is not asserted: no scholarly source for it has been retrieved.

  • Manga Intent · high confidence

    Posture is conventional, not refined: a mature publishing grammar working within its own norms, not a position taken against something.

  • Manga Writing system · high confidence

    Reading direction and vertical setting belong to the Japanese writing system; drawn onomatopoeia belongs to the visual language.

  • Maximalism Intent · medium confidence

    Sincerity separates it from Camp: both are dense and excessive, and Maximalism means it.

  • Memphis Intent · high confidence

    Knowing rather than ironic: the group meant the attack, and the attack depends on the viewer recognising what is being attacked.

  • Modern SaaS Intent · medium confidence

    An editorial abstraction of recurring contemporary practice, not a documented movement: no manifesto, no named originator.

  • Swiss Style Intent · high confidence

    Posture is reactive, not refined: the programme was formulated against prewar decorative advertising.

  • Terminal Intent · medium confidence

    Knowing, not sincere: a website does not need a command line, and both author and audience know it.

  • Vaporwave Intent · high confidence

    The ambivalence is documented, not inferred.

  • Y2K Intent · medium confidence

    The original register and the 2020s revival are not the same thing; the intent describes the original.

For machines

Every Aesthetic on this site also exists as structured JSON, generated from the same canonical record as the page you are reading.

One file per published Aesthetic, at data/v1/<slug>.json, listed in data/v1/index.json. Each follows one versioned contract, aesthetary.record/1.0. Every claim carries its basis, the sources that support it and the ones that dispute it, and a label from one shared vocabulary, data/v1/vocabulary.json. Web translations and general good practice sit in lists of their own, so a tool cannot mistake them for the Aesthetic's rules. A value that is still owed, or withheld for cultural review, is marked as such rather than left out.

data/v1/swiss-style.json is the fullest one. The older data/<slug>.json files stay for one release and point to their replacement.

What it is not

Static files, not an API. There is no endpoint and nothing to rate-limit. The contract carries a version number, so a change in what a field means arrives as a new version, not a silent edit.

What we will not do

  • Fill a gap with plausible copy. If a page looks thin, the record is thin. Nothing here is written to make a screen look finished.
  • Show a picture we cannot account for. No moodboards yet. Reserved plates stay empty until there is a licensed image with a source behind it.
  • Turn an estimate into a measurement. The dimension figures are one editor's guesses with no sample behind them, and they are labelled that way everywhere they appear.
  • Read silence as permission. An area nobody has researched is an open question, and the page says open question.
  • Dress an unbuilt feature as a built one. Mix and Recipes have routes that explain themselves and no mock data.

This is an Alpha. The interface exists to find out where the knowledge model breaks, so the places it looks sparse are the places it is telling the truth.

Not written yet

One section of this page cannot be built honestly, and it is worth naming rather than leaving as an absence.

Contributors
Who did the reading. The corpus records the people it refers to — Warhol, Sottsass, Müller-Brockmann — and not the people who researched it. Rendering those as contributors would be a plain falsehood, so this waits for a field that records authorship. The research itself is AI-assisted, as described under Research standards.
Feedback