# AIE Flash Homepage Design Benchmark Review — 2026-06-10

Reviewed artifact root: `/root/projects/aieditorrsp/design/aie-prd-v4-flash-homepage-20260610-112815`

Kanban:
- Base design task: `t_ca406e02` — DONE, but verdict overstates readiness because it accepted local HTML patch evidence.
- Canvas brand fix continuation: `t_0c239816` — RUNNING at review time. It must supersede the current screen-index when complete.

## Verdict

`DESIGN_DIRECTION_REVIEW_CONDITIONAL_GO`, not `DESIGN_GO` and not yet a benchmark-quality design process.

Visual direction is materially better than the previous SEO-checklist/text-wall path. It now has a product-first dark catalog/editor feel and a plausible AIE homepage structure. But the process still has critical truth-source and gate gaps. The biggest failure is that `LensEdit` was patched out of local HTML/screenshots while the actual Stitch canvas still showed `LensEdit`. That means local artifacts and official Stitch truth diverged.

## Evidence checked

Files present:
- `content-fit-matrix.md` — present, 8.5 KB
- `DESIGN.md` — present, 3.8 KB
- `DESIGN_VERDICT.json` — present, 3.5 KB
- `screen-index.json` + `stitch/screen-index.json` — present
- desktop/mobile HTML — present
- desktop/mobile PNG + contact sheet — present

HTML audit:
- local HTML `LensEdit`: 0
- local HTML `Unlimited`: 0
- required H1 / short answer / H2 / FAQ strings: present
- desktop title: `AI Editor RSP | AI Editor RSP Editing Prompts`
- mobile title: `AI Editor RSP - Editing Prompts`

Stitch truth audit:
- project: `1491412448695731545`
- stale desktop screen: `4bc55a135912481b9b0102d7327e868d`
- owner observed actual Stitch canvas still displaying `LensEdit`
- therefore current screen-index points to a screen that is not accepted as source of truth.

DESIGN.md lint:
- errors: 0
- warnings: 1 — section order wrong: Layout appears before Typography. Expected: Overview → Colors → Typography → Layout → Elevation & Depth → Shapes → Components → Do's and Don'ts.

Missing benchmark files:
- `visual-style-rationale.md` — missing
- `consistency-matrix.md` — missing
- `nav-seo-audit.json` — missing
- `HANDOFF.md` — missing
- `BRAND_FIX_VERIFICATION.md` — missing at review time

## Visual review

Strengths:
- Direction is now product-first: dark prompt catalog + editor cockpit, not generic SaaS.
- Above fold has the right task chain: H1 → short answer → before/after editor proof → Copy Prompt / Open in Editor.
- Featured prompt cards are visual and actionable; this is closer to a real prompt/effect library.
- SEO blocks are pushed below product proof, which matches the intended Layer A/B/C priority.
- Mobile preserves product modules and does not collapse into a pure SEO article.

Weaknesses:
- Desktop hero balance is still slightly crowded: the right editor proof dominates, but the left value proposition is small. It is usable, but not yet premium.
- Prompt cards look good directionally, but card semantics are still shallow: missing obvious `Best for`, `Avoid if`, variable fields, safety note density expected by the contract.
- Mobile is long and dense. It works for review, but the first 1–2 screens still feel like a compressed desktop rather than a deliberately mobile-native creator flow.
- Visual identity is improved but not distinctive enough to be a reusable brand system. It relies heavily on dark UI + orange CTA + generated images.
- The page has no true interaction-state design package yet: empty/uploaded/loading/result/error/quota/provider-disabled are implied, not delivered as state screens.

## Process review

What worked:
1. PRD six-pack was read and translated into `content-fit-matrix.md` before Stitch.
2. Three-layer priority interpreter is the right direction: Product skeleton > product modules > SEO modules.
3. Model comparison was useful; Flash was a legitimate visual-direction choice for this site.
4. Generating a canonical homepage before full-site expansion is the right gate.

What failed:
1. **Truth-source failure:** Local HTML patch was accepted as brand fix even though actual Stitch canvas remained wrong.
2. **Verdict inflation:** `GO_WITH_NOTES` was too strong. This should have been `DIRECTION_REVIEW_CONDITIONAL_GO` until official Stitch canvas was corrected and owner approved.
3. **Missing formal gates:** no `visual-style-rationale.md`, `consistency-matrix.md`, `nav-seo-audit.json`, `HANDOFF.md`, or brand verification report.
4. **DESIGN.md not canonical enough:** lint only warns, but section order and missing elevation/shapes reduce its value as a design-system source.
5. **No state coverage:** homepage direction exists, but product-state design has not been validated through separate state screens.

## Benchmark process proposal

For AIE and future site-factory design benchmark, use this 8-gate flow:

1. **Inputs Gate**
   - PRD Core
   - Product-state Contract
   - Pricing Contract
   - SEO-copy Freeze + Route Index Contract
   - Design Acceptance Contract
   - Implementation Acceptance Plan
   - If any missing: BLOCK, no Stitch.

2. **Visual Strategy Gate**
   - Write `visual-style-rationale.md` before DESIGN.md.
   - Compare at least 3 visual candidates.
   - Include SERP intent, geo/culture, content form, competitor interaction baseline, legal/material constraints.

3. **Content Fit Gate**
   - Write `content-fit-matrix.md` before Stitch.
   - Split Semantic Fit Matrix and Product/Visual Fit Matrix.
   - Explicitly mark Layer A/B/C priority.

4. **DESIGN.md Gate**
   - Write `DESIGN.md` in official order.
   - Run `npx -y @google/design.md lint DESIGN.md`.
   - Error = BLOCK. Warning = allowed only if explained in HANDOFF.

5. **Canonical Homepage Direction Gate**
   - Generate desktop + mobile official Stitch screens.
   - If model is non-default, record `model_waiver` and comparison evidence.
   - Verdict can only be `DIRECTION_REVIEW`, not `DESIGN_GO`.

6. **Official Stitch Truth Gate**
   - `screen-index.json` must point to the actual accepted Stitch canvas.
   - Fresh `getHtml()` + `getImage()` exports only.
   - Local edits after export cannot be used as source-of-truth proof.
   - If local patch is necessary, it must trigger a new Stitch edit/generate pass or mark `LOCAL_PATCH_ONLY_BLOCKED`.

7. **Design Acceptance Gate**
   - Produce `nav-seo-audit.json`, `consistency-matrix.md`, `BRAND_FIX_VERIFICATION.md`, and `HANDOFF.md`.
   - Visual review must answer: can a user understand the task in 5 seconds? Is Copy Prompt visible? Is Open/Upload secondary? Is SEO supporting rather than dominating?
   - For product sites, require key state screens before full frontend handoff.

8. **Owner Expansion Gate**
   - Owner reviews canonical homepage direction.
   - Only after owner says GO: expand to route/state matrix.
   - Full-site expansion must use same project/design-system truth and cannot silently reuse stale screens.

## Current next action

Wait for `t_0c239816` to finish. It must create or edit a real official Stitch screen/project where the canvas itself says `AI Editor RSP`. After that, re-run:

- local HTML scan: `LensEdit=0`
- Stitch screen-index freshness check
- screenshot/contact sheet visual review
- DESIGN_VERDICT downgrade/fix: no `pass_after_html_patch_and_local_rescreenshot` wording
- create `BRAND_FIX_VERIFICATION.md`

Only then is this acceptable as a homepage direction candidate for owner review.