---
name: "convention-aesthetic"
type: convention
description: 'The cross-venture aesthetic canon — Aled''s visual sensibility above the per-venture brand entries: written taste principles, named references and why, anti-patterns, and pointers to the Figma reference library. Read before any brand, design-system, or exploration work so the aesthetic baseline is inherited rather than re-derived each engagement; per-venture convention-brand instances inherit from and specialise it. Kept evolving by the taste-capture loop defined here. Owned by Pixel via skill-brand-system; first draft from the 2026-07-10 capture, grown as signals accrue.'
license: CC BY-NC-SA 4.0
---

# Aesthetic canon

The taste layer above the brands. `convention-brand` holds one instance per venture; this convention holds what precedes and informs all of them — Aled's cross-venture visual sensibility, recorded so Pixel inherits a baseline instead of re-deriving (or guessing) it each engagement. Owned by Pixel via `skill-brand-system`; read before any brand, design, or exploration work.

## The two-layer model

* **This convention** — the cross-venture layer: taste principles, references, anti-patterns, and the reference-corpus pointers. Venture-agnostic.
* **`convention-brand` instances** — the per-venture layer: each inherits from here and specialises (BARA sharpens "natural materiality" into wabi-welsh and Welsh craft; another venture may specialise the same principle differently). An instance states its divergences from this layer rather than restating it.

Nothing here licenses bleeding one venture's *identity* into another — the shared layer is sensibility, not livery.

## Taste principles

Each entry is tagged **known** (stated or repeatedly confirmed by Aled) or **assumed** (generalised from one venture's evidence, awaiting confirmation). The seed set below is generalised from the BARA brand entry and is therefore largely assumed at the cross-venture level; the capture loop below is what hardens or corrects it.

* **Quiet over loud** *(assumed)* — understated, grounded, warm minimalism; quality read through restraint, never through gloss or volume. "Quietly luxurious."
* **Natural materiality** *(assumed)* — earthy palettes, real texture, visible process and maker; an affinity for wabi-sabi restraint — imperfection, rootedness, weather. (Known for BARA as "wabi-welsh".)
* **Editorial over commercial** *(assumed)* — surfaces read like a journal, story-first; never salesy, never templated ecommerce.
* **Cultural specificity without cliché** *(assumed)* — rooted, particular references (place, language, tradition) that earn their place; never folksy pastiche.

**Natural materiality narrows: the shared principle is restraint and one accent, not a hue family.** A1 is the first venture other than BARA with concrete palette evidence, and it is not earthy — Aled supplied bone, near-black, slate, a hot red-orange accent, rust and salmon (2026-07-16, APP-572; recorded in `convention-brand`). That does not contradict *quiet over loud*: the ground carries the quiet and a single hot accent is still a single accent. It does show that "earthy palettes" was BARA's specialisation being read as the cross-venture rule, which is exactly the risk the **assumed** tag exists to flag — the seed set was generalised from one venture, and the second venture corrected it. So the principle stays **assumed** and now reads narrower: restraint, real texture, one accent used for meaning. The hue family belongs to the instance.

## References

Named references carry a *why*, so they steer rather than decorate. They come in two kinds, and keeping them apart is what stops a layout lesson being mistaken for a brand-identity one: a **brand-sensibility** reference steers tone, materiality and livery; a **layout / document-craft** reference steers hierarchy, grid and how information is presented. The second kind is legitimately borrowed from outside the venture's category — a software company's deck craft can steer a homeware report without licensing anything about how that homeware brand looks.

**A third kind is open, not filled: product and data UI.** Both kinds above were written for brand and deck work, and both reference sets are drawn from it — fashion, apothecary, homeware and furniture houses for sensibility, a software company's deck craft for layout. Neither steers an **internal product or console surface**: a dashboard, an operator console, an admin or analytics view. A builder of one reads this file, finds nothing that names the target taste for dense data UI, and takes it from whatever the tool in hand supplies — which is the anti-pattern below, reached by following this canon rather than by ignoring it. So the kind is named here with its set **empty and marked as a gap**, rather than filled from research: a reference steers only once Aled has pulled it in or confirmed it, and a reference chosen for him would put a tool's taste in this file under his name. Until one lands, the nearest thing a product or console surface has is the kit and token bar already stated — `standard-design-system` (*Default UI kit — product and console surfaces*, and *The bar*) — which answers what to build with and never what it should feel like. Fill this with the first product-UI reference Aled names, with its why and its kind, per *Taste capture* below. (Gap pointer: APP-2525, raised 2026-10-21 from an internal dashboard the Owner rejected on sight. The candidate references floated in that note are a session's research reading, **reported**, and are deliberately not recorded here.)

**Brand sensibility.** Seed set — evidenced in the BARA entry, assumed to generalise:

* **TOAST** — craft plus a sharp editorial lens; the closest peer for tone and restraint.
* **Aesop** — restrained luxury, apothecary utility, typographic discipline.
* **Frama** — materials-led, architectural calm.
* **Margent Farm** — process- and material-honesty.
* **Ffern** — small-batch ritual, seasonal storytelling.

**Layout and document craft.**

* **Chronicle** (chroniclehq.com) — one idea per surface, stated as a headline; small-caps eyebrow in a single accent; asymmetric thirds grids with hairline structure; charts as flat single-accent figures in soft cards; full-bleed statement covers carrying heavy negative space; near-monochrome grounds with one accent per deck. The reference for deck and report layout — **not** for brand identity. (Pulled in by Aled 2026-07-15 and studied by Pixel; full study `pixel-reference-chronicle-layout.md`. APP-561.)

Chronicle is supporting evidence for *quiet over loud* and *editorial over commercial* — both still tagged **assumed**, and staying so: a reference Aled pulls in evidences the sensibility, but the tag turns on him **stating** the principle or confirming it repeatedly, which this is not. Recorded here so the next signal in the same direction hardens both rather than starting over.

**A hero image is a conceptual illustration; the diagram does the explaining** *(assumed, 1 instance)*. Where one surface carries both — an article with a hero above and a diagram below — the two do different jobs, and a hero that restates the diagram leaves the surface explaining the same thing twice and introducing it not at all. The hero's job is to put the reader in the idea; the diagram's is to show the mechanism. So a hero is an illustrative representation of the concept, not the diagram rendered in another style, and least of all the diagram in isometric form, which reads as explanatory while explaining nothing the diagram does not already. Aled's verdict, on a first hero set of three isometric images each restating its article's diagram: *"the hero image is not linked to the diagram but is an illustrative representation like this reference"* (2026-10-21, the Mr Pritchard three-layers series). One instance, so **assumed** — a second verdict in the same direction hardens it, per *Taste capture* below. The **material** qualities of the reference he supplied alongside it — ground, texture, monochrome, the scale-anchor figure — are Mr Pritchard identity rather than cross-venture sensibility, and belong in that venture's `convention-brand` instance under *Imagery*; this entry carries the principle only. Those qualities are the filing session's reading of an image not attached to the note and not read here, so they are **reported**. (APP-2554.)

Add a reference only with its why and its kind; retire one when a verdict contradicts it.

## Anti-patterns

* Over-polished, couture gloss; anything that reads as luxury-by-volume.
* The generic AI/UGC look — synthetic sheen, unplaceable interiors, stock-adjacent styling.
* Cliché treatments of cultural material (for BARA: Welsh kitsch; the rule generalises).
* Trend-pastiche — "contemporary" asserted by copying the current feed rather than argued from the concept.
* **The tool-default look** — warm cream ground with a muted terracotta accent: the palette an AI tool reaches for when nothing is briefed. Aled's verdict on it: *"I don't like how you have gone back to the anthropic look and feel"* (2026-07-16, a1.Muse deck-04). The objection is not that the palette is ugly — it is that it reads as **tool-default rather than chosen**, so the surface asserts nothing about the venture. Distinct from *the generic AI/UGC look* above, which is about imagery; this is the ground and accent an unbriefed render falls back to, and it is the more insidious of the two because it looks tasteful. Where no palette is briefed, take it from the venture's `convention-brand` instance; where that instance has no palette yet, **ask rather than fall back** — an unbriefed default is a decision made by the tool. (APP-572.)

**A design default supplied by the tool is an instance of the same anti-pattern, and now the commonest one.** The bullet above is written about a render with nothing briefed. The identical failure arrives pre-briefed: a host, a template or an artefact type ships an advisory look — a named face pairing on a host token set — a build follows it literally, and the surface asserts the tool's identity rather than the venture's. It is the harder case to catch, because there was something to follow, so the build reads as compliant and the inherited look never surfaces as a decision anyone made. Treat an advisory look exactly as the bullet above treats a missing palette: **a supplied default is a starting point to replace, not a brief to satisfy.** Take the palette and type from the venture's `convention-brand` instance; where the surface is an internal product or console one, from the bar in `standard-design-system` (*Default UI kit — product and console surfaces*); ask where neither carries it. (Worked case: an internal dashboard rebuilt on an artefact type that ships an advisory look, 2026-10-21 — the look was followed as written and the Owner's verdict on the result was *"so Anthropic looking"*. That the type ships that look and that the build followed it are the filing session's account and are **reported**; no artefact type or host surface is readable from the refine surface. APP-2525.)

## The Figma reference library

The image corpus lives in **Figma as curated reference boards** (per theme and per venture), read by Pixel via the Figma MCP; this convention holds the pointers plus the written principles — never the images themselves. **Gap:** the boards are not yet stood up (a human/Pixel action); until they exist, the written entries above are the whole corpus. When a board lands, record its pointer here.

## Taste capture — how this canon evolves

Aled's design verdicts are the source of truth for this file, and they must not evaporate. The loop mirrors `skill-ops-retro`:

* **Capture.** When a Pixel skill's output receives a verdict that reveals a taste signal — a rejection with a reason, a "more like this", a reference Aled pulls in — the session files a `FRICTION:` note to the Pulse queue naming this convention (or the venture's `convention-brand` instance) as the target, quoting the verdict. Capture only; the session never edits this file on the back of one verdict.
* **Drain.** The weekly Pulse retro folds accepted signals in — adding, hardening (assumed → known), or retiring entries. Recurrence is the escalation signal, exactly as for process friction.
* Taste notes reuse the existing `FRICTION:` prefix and queue — no separate prefix or intake; the named target is what lets the drain group them (drain decision, 2026-07-11).

## Guardrails

* This layer records *sensibility*; identity stays per-venture. Do not use a shared principle to justify one brand's look appearing in another.
* Tag every entry known or assumed; never present a generalisation as confirmed.
* Entries change through the capture-and-drain loop or Aled directly — never silently mid-engagement.
