# Shared UI contract, revision 1

## The outcome
One public, commentable canvas lets Matt and Rita follow both acquisition routes through enrollment, fourteen days of practice and daily personal feedback to the final online consultation. It contains the actual clickable customer screens, not cards saying that screens should exist. It feels like one Komplex product. Persuasion is a hypothesis to test later, not a measured conversion claim.

This run improves existing UI. It does not rebuild the backend, deploy campaigns, send messages, process payments, create real accounts or book real consultations. The old 44-criterion board and its evidence viewer are not deliverables of this run. Do not rebuild them.

## Existing work and exact workspace
- Project: `/Users/agency/Documents/Agty/KomplexLogopédia`. Parent context: `01a0c622-0f03-7480-9f24-003a441af4b3`.
- Immutable copied baseline: `/Users/agency/Documents/Agty/KomplexLogopédia/outputs/rita-ui-parallel-20261002/baseline`.
- Only new implementation root: `/Users/agency/Documents/Agty/KomplexLogopédia/outputs/rita-ui-parallel-20261002/site`. It is already seeded from the baseline. Do not recreate it or copy over a sibling's changes.
- Input identity receipt: `/Users/agency/Documents/Agty/KomplexLogopédia/plans/rita-ui-parallel-20261002/baseline-manifest.json`.
- Previous published UI: https://review.clientsflow.hu/komplex-journey-2026-10-02-v1/ . It is a reference, not the new output.
- Original executor's `outputs/rita-journey-20261002`, original board, reports, Git changes and published URLs are outside all writer ownership. No transcript archaeology, source rediscovery or monitoring is required. The frozen snapshot prevents collisions with that executor.
- The baseline has 20 square ads, 4 landings, 7 landing visuals, 14 participant and 14 Rita screens, 56 daily visual alternatives, and 58 email topics with 116 complete A/B versions. Preserve useful work and these inventories. Fix weaknesses in place. Regenerate a visual only when its actual text, relevance, crop or legibility fails the new screen.

## Fixed product and content decisions
1. Broad adult meta-persona: wants to improve everyday speech, recognises situations such as a call, meeting or explaining something, and needs a concrete starting point and manageable practice. Describe contexts without diagnosing the viewer, inventing their feelings, or narrowing the whole funnel to performers. Reuse `baseline/review/persona-and-canvas.md` and `reasoning.json`; do not commission another persona study.
2. Free route: short spontaneous video, free personal feedback, then useful optional follow-up and a relevant invitation to the challenge. The three-working-day written response is the current proposed service detail. Put operational detail on the landing/form, not automatically in every ad hook. Ads say a short video, not a compulsory one-minute sales hook.
3. Paid proposal: a 14-day Online Beszédalapozó, daily written guide, daily participant practice video, daily personal written feedback from Rita, and a final online review consultation. The pilot can promise daily support in the proposal. Do not silently reinstate the earlier no-feedback or four-batch model. Matt proposed this service model; Rita's final operational acceptance is still open.
4. Proposed price: 24,900 Ft once. Proposed first cohort: 12 people. Real start/end dates, a real deadline and actual availability are unconfirmed. Distinguish proposal status in the reviewer context and at commercial decision points. Do not add a fake ticking countdown, scarcity count, refund guarantee, testimonial, certification, result or fabricated expert recording. No claim that fourteen days resolves every speech difficulty.
5. Preserve the six original lesson texts in `baseline/app/source/module-1.txt` through `module-6.txt`, including meaning, exercise wording and cautionary instructions. Grouping, headings, navigation and scaffolding may improve. The existing 14-day split is a proposed organisation of those six modules. Initial spontaneous video and final agreed 2–3-minute practice video are distinct. The original 20/30-minute module guidance is not proof of a per-day duration; any daily estimate remains a labelled proposal and includes additional recording/reflection time.
6. Existing rewatch checkpoints are days 2, 5, 8 and 12. Show the relevant previous-day clip and feedback before practice, with an honest missing-feedback state. Daily personal feedback is the primary model.
7. Use an optional separate marketing choice, not a condition of free feedback. Weekly helpful email is the baseline; a cohort sales sequence replaces that week's newsletter and excludes paid participants. Demo UI shows the states. No mailing integration.
8. Natural Hungarian. Every ad image hook, caption, copy, destination hero and CTA must tell a coherent story. No Rita name on ads. On later screens her identity is useful where it establishes who actually gives the service. Avoid sentimental pressure, generic benefits without context, em dashes and semicolons in customer copy.

## Brand and authoritative references
- Use the actual Komplex style already in the baseline: blue `#008AE6`, pink `#E14F9B`, yellow `#FEA500`, white, Nunito Sans headings and Quicksand body, the real logo and Rita portrait. Retain the darker accessible action/text colours already used where appropriate. Do not replace this with a generic software-dashboard theme.
- `baseline/review/brand-source-reference.md`, `DESIGN.md`, `brand-reference/`, `source-layout-kit.html`, `landing-layout-map.md`, `source-map.md`, and `promise-destination-matrix.md` contain the useful visual/content evidence.
- Rita Figma context: `/Users/agency/Documents/Agty/KomplexLogopédia/plans/rita-funnel-20260926/source-figma.md`; actual file `g8LRZzfXbjRUgDA8dROzzb`, comment-rich frames `9473:13681`, `9630:1670`, `9610:10899`. Reuse their language and design insight. Old claims are not automatically approved. Matt explicitly superseded comments about avoiding daily work/feedback. Other material content objections remain relevant.
- Current website: https://www.komplexlogopedia.hu/ . Saved source HTML/CSS: `/Users/agency/Documents/Agty/KomplexLogopédia/outputs/rita-journey-20261002/private/landing`.
- Required canvas reference: https://review.clientsflow.hu/ccklima-canvas-2026-10-01-v1/ . Saved original: `/Users/agency/Documents/Agty/KomplexLogopédia/outputs/rita-journey-20261001/private/canvas-live.html`. The working baseline already adapts its pan/zoom/fit/inspector behaviour. Reuse it.
- Required landing kit: `/Users/agency/Desktop/reusable-template-kit/new_template-3.html`, also copied as `baseline/review/source-layout-kit.html`. Preserve the existing justified mapping of all 28 layouts across the four-page family. Do not inflate every page to 28 sections; log any coverage change rationally.

## Parallel ownership and route agreement
You are not alone in this codebase. Never revert another worker, overwrite their directory, reset the root, or mutate the frozen baseline. All paths below are relative to the new `site/`.

| Agent | Exclusive writes | Read-only dependencies |
|---|---|---|
| 01 Ads | `ads/**`, `assets/ads/**` | landing pages and shared reference assets |
| 02 Landings | `landing/**`, `assets/landing/**` | ads, flow and shared reference assets |
| 03 Flow | `flow/**` | landings, app lesson data and app routes |
| 04 Course | `app/**`, `assets/daily/**` | flow and shared reference assets |
| 05 Canvas | `index.html`, `canvas.css`, `canvas.js`, `review/**` | all four module trees |

Everyone also owns their separate `handoffs/<agent-id>/`, `evidence/<agent-id>/` and `reports/<agent-id>.md` beneath `/Users/agency/Documents/Agty/KomplexLogopédia/outputs/rita-ui-parallel-20261002`. Agent05 alone publishes the assembled batch. Root/shared fonts, logo and portrait are read-only to 01–04. If a sibling needs a change, record the exact seam in the handoff; do not patch their files behind them. An owner can delegate within its own partition through `/launch-agent`, with non-overlapping ownership.

Keep these existing routes stable. `?variant=` continues to distinguish the four landing IDs. Preserve the existing participant query for the course, default `primary`, alternate `sample-b`. Do not create another incompatible identity system.
- `ads/ad-01.html` through `ad-10.html` → `landing/free-a.html` or `free-b.html` → `flow/free-upload.html?variant=free-a` or `free-b` → `flow/free-receipt.html` → `flow/free-feedback.html` → `flow/emails.html?topic=nurture-1` → paid landing.
- `ads/ad-11.html` through `ad-20.html` → `landing/challenge-a.html` or `challenge-b.html` → `flow/checkout.html?variant=challenge-a` or `challenge-b` → `flow/confirmation.html` → `app/index.html?from=confirmation` → `app/overview.html` → `app/day-01.html` … `app/day-14.html` → `app/consultation.html`.
- `app/rita-01.html` … `rita-14.html` match participant/day IDs one-to-one. Consultation emails link to `app/consultation.html`, not a separate booking system. Email viewer uses existing stable topic IDs and existing version switching. It accepts and preserves `participant=primary|sample-b` on every course-bound link. The course owner preserves `#feedback-current` on all14 days. A received-feedback email links to an existing response for that participant/day; an absent response shows an honest missing-state fallback. An email click must not fabricate a submission or feedback.
- Preserve the current local state model, but isolate this new implementation from the old demo on the same host. Agent03 owns new key `kl-ui-parallel-20261002-funnel`; Agent04 owns new course key `kl-ui-parallel-20261002-course` and IndexedDB `kl-ui-parallel-20261002-videos`. Never clear or migrate old state automatically. Check that old-demo data remains unchanged after exercising this new implementation. The confirmation-to-course URL is the required handoff; it must work without either agent writing the other's storage. Existing demo/sample entry remains available. Nothing treats a URL as proof of actual payment.
- Landing links must be relative to the new assembled site, not the old host. Preserve variant through cancel/retry/return and through free-flow steps. A free-entry variant maps to challenge-a by default; challenge-b is an explicit alternative. Freeze this default so separate copy agents do not invent different funnels.
- Every main CTA specifies a meaningful action and has a real next screen or a deliberate, visibly labelled demo state. Every expanded preview can return to the canvas without losing the reviewer's location.

## How to work efficiently
Read your contract, this file and the relevant existing module. Inspect one complete current path, then improve visible UI. Use a short keep/fix list in your report, not a new research phase. Finish one coherent route or state before polishing another. Reuse shared components inside your owned module; do not fabricate fourteen unrelated apps.

Each owner writes `handoffs/<agent-id>/manifest.json` with `status` (`ready` or `blocked`), `owned_roots`, `entry_routes`, `recommended_routes`, `changed_files`, `checks` (criterion ID, result, evidence path), `open_seams`, `source_snapshot_sha256`, and `api_spend_usd`. Agent04 also reports the final `lessons.json` hash and changed day IDs. Agent03 records the lesson hash its 42 daily topic families were checked against. Agent05 compares these and rechecks affected email copy if the final lesson hash differs. Site-relative paths in route fields. Evidence stays local. A ready handoff means usable UI exists and the owner's checks pass, not merely copy or a plan. No fixed new report count, screenshot gallery or repeated overall rescoring.

One agent can finish independently against the existing routes. Agent05 can prepare the canvas concurrently, then integrates only declared ready handoffs. If a writer still owns a module, the integrator does not edit it. After final DONE, Agent05 may take exclusive ownership for a named seam repair, recording the handover before touching it. Never race a writer.

## Quality and verification that serve the UI
Customer interfaces are mobile-only: check 360, 390 and 430 CSS px. Do not build or QA larger customer layouts. The outer canvas may use a desktop review viewport. Interactions, source-text parity, correct destinations and consistency are required, not just page counts.

For visual quality, use `/impeccable` where relevant and the actual `/luna-visual-qa-section-screehoshots` workflow. Partition whole page families between independent Luna xhigh reviewers, use Luna max for bounded fixes, and fresh postdeployment review. The user's mobile-only scope overrides the skill's desktop default. Reuse the existing capture harness where helpful. Inventory visible semantic sections, not every DOM descendant; inspect tall sections and readable overlapping 3–4-section runs. Do not multiply every content alternative by every unrelated state. Record equivalence of truly shared unchanged components, inspect all unique customer content, and disclose uncovered scope. Actual saved raster inspection is mandatory for a visual pass; a receipt or generated screenshot filename is not a review.

Each owner supplies one compact local findings table plus actual screenshot links with page, width, state and what the reviewer saw. Agent05 coordinates the final deployed check. This does not mean copying the old thousands-image viewer into Rita's canvas. Resolve copy/flow/source errors before pixel refinements. Do not claim a conversion lift without campaign data. Proposed A/B tests name one hypothesis, one primary metric and the next decision, without a fake winning variant.

## Authority, recovery and completion
No backend, live Stripe, real uploads, email sending, scheduling, campaign changes, new login protection or account provisioning. Browser-local simulation is required and must be distinguishable from real service. Customer screens should remain natural; put engineering notes in reviewer context, not repeatedly inside the sales pitch. Source/pilot uncertainty stays visible where a decision depends on it.

Do not purchase new API generations by default. This pack is a reuse/refinement assignment. Native available tools may be used. The shared USD15 task limit across agents remains binding; Agent05 alone allocates any remaining paid API amount from a single spend ledger after verifying previous spend. No separate $15 per agent. Gemini 3.1 Pro provenance is already present for the four landings; use that model if genuinely regenerating their design, with a recorded permitted allocation. GPT image tooling is the required tool for replacement raster visuals. Record actual access problems instead of silently swapping the required model.

A criterion can be replaced only for a concrete source conflict, misleading test, duplicated proof or a better check of the same outcome. Write the original, replacement, reason and evidence into your report before declaring it passed. Never waive an asset count, original lesson wording, mobile scope, useful next action, independent inspection, or a truthfulness limit just to finish.

When invoked as a new task, establish the native `/goal` using this outcome if it is not already active. Finish and verify your owned module; do not launch the other contracts without Matt's explicit launch instruction. Return once through your normal final response with `DONE` or `STALLED` and `/Users/agency/Documents/Agty/KomplexLogopédia/outputs/rita-ui-parallel-20261002/reports/<agent-id>.md`, using `/Users/agency/.agents/skills/launch-agent/report-template.md`. No separate parent message, audio, automatic preview, sibling supervision or transcript polling. DONE is owner completion; final integrated acceptance belongs to Agent05 and Matt's review. Finish independent work before reporting a precise irreducible blocker.