# Outlex Taste v3.1: Approved and Rejected Examples

Version **v3.1**, dated **14 September 2026**. This is the canon's copy of the evidence behind `tokens/PRINCIPLES-v3.md`: two annotated walls of real examples the founder has scored, plus a curated slice of the correction ledger they come from. Look at the images before designing; reading the principles text is not a substitute.

Assets published alongside this file: `brand/taste/wall/wall-great.jpg`, `brand/taste/wall/wall-rejected.jpg`, `brand/taste/founder-taste-ledger-v3.json` (57 curated entries; full 184-entry ledger and every rating/review file: outlex-marketing `main`, `marketing/design-system/evidence/design-study-2026-09/`).

Precedence: current founder correction, then live canon (this file included), then project adapter, then surface-specific skill. Full statement: `brand/precedence.md`.

---

## 1. What the great wall shows

`wall-great.jpg`, "Outlex design wall: what great looks like." Nine panels, each labelled with the one mechanic that makes it work. Borrow the mechanic named on the panel, never the layout, the asset, or the brand it belongs to.

| Panel | One-line mechanic |
|---|---|
| Outlex live site (the bar) | One real person in warm side light, one serif headline with one sage italic phrase, two buttons, nothing else; beat this on restraint, class, and human feeling. |
| Circle / FIELD.IO | One light source fills the frame; the words stay small in a corner. |
| Circle / the one number | One decisive value carries the whole section. |
| FLORA | Strong crops around a protected reading island; serif with italic. |
| Granola | Product re-typeset, legible, small, over scenery beside real people. |
| Attio | Colour lives only inside the product; the page stays quiet. |
| Linear | Restraint: small confident type, the product carries the page. |
| Mercury | A photograph as atmosphere; one calm headline. |
| Cursor | A real product window over a painted landscape, warm and light. |

The Outlex live site is panel 1 and the explicit bar: any new direction has to beat it on restraint, class, human feeling, and fit with context, not merely avoid violating a rule (`tokens/PRINCIPLES-v3.md` P-36).

## 2. What the rejected wall shows

`wall-rejected.jpg`, "Outlex design wall: known rejections." Six panels; every one of these lost to the live site. If a new design resembles one of these, start again rather than iterate on it.

| Rejected pattern | Why it lost |
|---|---|
| Pale sage board, raw screenshot | Pale, boring: a screenshot inside a frame is not product presentation (matches P-19). |
| Small cropped finding | Scored 2/10: tiny pale crop, unclear story, nothing invites a closer look. |
| Decorative journey diagram | Dashed lines and boxes that explain nothing: decoration standing in for a real diagram. |
| Purple glow, giant number | The purple reads corny; devices substituted for genuine class. |
| Closest, still loses | Human touch present, but the product crop is too big and too in-your-face, with a textured background (matches P-18). |
| Generic tile grid | Rule-clean but templated: no photograph, no distinctive move (matches P-05). |

## 3. Reading the curated ledger

`brand/taste/founder-taste-ledger-v3.json` groups 57 entries (of the full 184) by eight areas: restraint, colour and light, type, photography, product presentation, components, motion, copy. `tokens/PRINCIPLES-v3.md` has a ninth section, "Across Surfaces" (§9, P-35 to P-37): P-36 and P-37 draw only from `Brief` and `Plan`, not from a ledger area; P-35 cites `Ledger#restraint` alongside `Brief`, so its underlying founder correction sits in the ledger's existing "restraint" category (§1) rather than a distinct ninth one. The ledger's eight categories are complete on their own terms; "Across Surfaces" is a synthesis section, not evidence of a missing ninth ledger category. Each entry carries a date, surface, correction kind (instruction, correction, rejection, reversal, or approval), the stated principle, and, where recorded, the founder's own words describing the change. Every numbered rule in `PRINCIPLES-v3.md` traces to one or more entries here, or to `Brief`/`Plan` directly; treat a `PRINCIPLES-v3.md` rule as the summary and this file (or the named non-ledger source) as its receipt.

Two calibration facts worth carrying into any new design pass:
- Two full recovery rounds (loop 1, loop 2) scored 20/100 and 25/100 against the live site and were rejected outright as "basic, corny, weak"; agent self-scoring on those same rounds was withdrawn as evidence, and only the founder's own side-by-side comparison counts (`reviews/founder-calibration.md`).
- A bounded product-presentation study scored 2/10 or worse on both its closed and open states even after an independent reviewer had accepted it; reviewer acceptance is not founder acceptance (`visual-proof/founder-review-2026-09-10.md`).

## 4. Full evidence

The complete study, all 1,360 files (the full 184-entry ledger, both calibration rating sets, every founder review, the wall source images, and the archived, do-not-adopt earlier rounds) lives on `outlex-marketing` `main` at `marketing/design-system/evidence/design-study-2026-09/`. Its `PRESERVATION.md` names exactly what is evidence versus what is an explicit do-not-adopt archive (the old dark-only router, the rejected visual-proof-v1, and loops 1 through 3). This canon file is a curated summary for agent grounding; when the two disagree, the full evidence and the founder's most recent word win, not this file.