Marketing Video System

Automated pipeline to generate real screen-recording demo videos from a running iOS simulator and display them in the website's phone mockup. Mirrors the existing screenshot generation system (scripts/gen-screenshots.sh + MarketingSnapshotTests.swift).

Architecture Overview

iOS Simulator (XCUITest drives app)
    + xcrun simctl io recordVideo
    + MarketingTapOverlayWindow (orange tap ripple)
    + MarketingMockLLMService (instant scripted responses)
    + MarketingDataSeeder (fake realistic data)
        ↓
gen-videos.sh (ffmpeg encode → WebM + MP4)
        ↓
apps/lucidpal-website/public/videos/
        ↓
lp-phone-video Angular component (A/B crossfade, tab nav)
        ↓
lp-hero (replaces static phone HTML)

Scenes

Scene	File	Duration	Launch arg
Onboarding	`onboarding.webm/.mp4`	8s	`--scene onboarding`
Agent	`agent.webm/.mp4`	8s	`--scene agent`
Chat	`chat.webm/.mp4`	8s	`--scene chat`
Notes	`notes.webm/.mp4`	8s	`--scene notes`
Habits	`habits.webm/.mp4`	8s	`--scene habits`
Live Notes	`live-notes.webm/.mp4`	10s	`--scene live-notes`

Tab labels (conversion-optimized, not feature names):

Scene	Tab label
onboarding	Setup in 30 seconds
agent	Ask your AI anything
chat	Fully private chat
notes	Capture every thought
habits	Build real habits

Implementation Phases

Phase 1 — Add XCUITest Target to `project.yml`

No UITest target currently exists. Add LucidPalUITests:

yaml

LucidPalUITests:
  type: bundle.ui-testing
  platform: iOS
  deploymentTarget: '18.0'
  sources:
    - path: UITests
  settings:
    base:
      SWIFT_VERSION: '6.0'
      PRODUCT_BUNDLE_IDENTIFIER: app.lucidpal.uitests
      GENERATE_INFOPLIST_FILE: YES
      CODE_SIGN_STYLE: Automatic
      DEVELOPMENT_TEAM: 'KRPUAN3FFA'
      TEST_TARGET_NAME: LucidPal
  dependencies:
    - target: LucidPal

Add to scheme test targets:

yaml

test:
  targets:
    - LucidPalTests
    - LucidPalUITests

Run xcodegen generate after editing. Creates apps/lucidpal-ios/UITests/ directory.

Phase 2 — iOS Marketing Infrastructure (`#if DEBUG` only)

All files in Sources/Marketing/. Zero prod impact — entire directory excluded from Release builds.

`MarketingEnvironment.swift`

Central launch-arg reader. Everything else reads from here.

swift

#if DEBUG
enum MarketingEnvironment {
    static var isActive: Bool {
        ProcessInfo.processInfo.arguments.contains("--marketing-demo-mode")
    }
    static var scene: String? {
        guard let i = ProcessInfo.processInfo.arguments.firstIndex(of: "--scene"),
              ProcessInfo.processInfo.arguments.indices.contains(i + 1)
        else { return nil }
        return ProcessInfo.processInfo.arguments[i + 1]
    }
}
#endif

`MarketingDataSeeder.swift`

Seeds fake but realistic data into app stores at launch. Called once from LucidPalApp.init() when MarketingEnvironment.isActive.

Seeded content:

Calendar: 6 events across next 3 days (Dentist, Team Standup, Lunch with Alex, Gym, Weekly Review, Pick up kids)
Chat history: 4 pre-scripted conversation pairs demonstrating agent capability
Notes: 3 notes (Meeting notes, Grocery list, Book ideas)
Habits: 4 habits with streaks (Morning run 12-day, Read 20min 7-day, Sleep 8h 3-day, Cold shower 1-day)

No personal data. All content is generic and fictional.

`MarketingMockLLMService.swift`

LLMServiceProtocol conformance for demo mode:

isLoaded = true immediately (skips model download screen)
generate(...) streams pre-scripted tokens at 25ms intervals (looks like real LLM typing)
Response bank keyed by scene name

Wiring in LucidPalApp.swift:

swift

#if DEBUG
private static func makeLLMService() -> any LLMServiceProtocol {
    MarketingEnvironment.isActive ? MarketingMockLLMService() : LLMService()
}
#else
private static func makeLLMService() -> any LLMServiceProtocol { LLMService() }
#endif

private let llmService = LucidPalApp.makeLLMService()

`MarketingTapOverlayWindow.swift`

UIWindow subclass that intercepts sendEvent(_:) and shows a branded tap ripple at every .began touch.

Visual: orange circle (#f97316, 22pt) → spring expands to 60pt → fades → gone. Total: 450ms.

swift

#if DEBUG
final class MarketingTapOverlayWindow: UIWindow {
    override func sendEvent(_ event: UIEvent) {
        super.sendEvent(event)  // never consumes touches
        guard let touches = event.allTouches else { return }
        for touch in touches where touch.phase == .began {
            showRipple(at: touch.location(in: self))
        }
    }
    // ...
}
#endif

Wired in AppDelegate.application(_:didFinishLaunchingWithOptions:) inside existing #if DEBUG block. Window level .alert + 1, isUserInteractionEnabled = false.

LucidPalApp reads MarketingEnvironment.scene at launch and sets the initial navigation state:

Value	Behaviour
`onboarding`	Clear all state, show fresh onboarding
`agent`	Skip onboarding, navigate to agent tab, pre-seed 2 past messages
`chat`	Skip onboarding, navigate to chat tab, fresh session
`notes`	Skip onboarding, navigate to notes tab
`habits`	Skip onboarding, navigate to habits tab
`live-notes`	Skip onboarding, navigate to live notes

Phase 3 — `MarketingVideoTests.swift` (XCUITest)

Location: apps/lucidpal-ios/UITests/MarketingVideoTests.swift

One test method per scene. Not testing correctness — testing visual storytelling. continueAfterFailure = true so one flaky tap doesn't kill the recording.

Sleep pauses are deliberate storytelling beats, not just technical waits. Every pause gives the viewer time to register what happened.

Known element identifiers (from source reading)

View	Element	XCUITest query
`OnboardingCarouselView`	CTA button	`app.buttons.matching(.button, identifier: "Continue").firstMatch` or `app.buttons["Get Started"]` — label set via `.accessibilityLabel(ctaLabel)`
`OnboardingCarouselView`	Skip	`app.buttons["Skip"]`
`ContentView` tab bar	Agent tab	`app.tabBars.buttons["Agent"]`
`ContentView` tab bar	Chat tab	`app.tabBars.buttons["Chat"]`
`ContentView` tab bar	Notes tab	`app.tabBars.buttons["Notes"]`
`ContentView` tab bar	Habits tab	`app.tabBars.buttons["Habits"]`
`AgentView`	Text input	`app.textFields["Ask the agent…"]`
`ChatView+InputBar`	Text input	`app.textFields["Ask anything…"]`
`NoteEditorView`	Title field	`app.textFields["Title"]`
`NoteEditorView`	Save button	`app.buttons["Save"]`
`NotesListView`	New note (toolbar)	`app.navigationBars.buttons.element(boundBy: 0)` — `square.and.pencil` icon, no label
`HabitDashboardView`	Add habit (toolbar)	`app.navigationBars.buttons["Add"]` — `plus` icon, no explicit label
`HabitDetailView`	Log button	`app.buttons["Log"]` — from `Label("Log", ...)`

Identifiers to add before XCUITests can be written reliably:

Several buttons have no accessibilityIdentifier. Add these during Phase 2 implementation:

swift

// NotesListView — new note toolbar button
Image(systemName: "square.and.pencil")
    .accessibilityIdentifier("new-note-button")

// HabitDashboardView — add habit toolbar button
Image(systemName: "plus")
    .accessibilityIdentifier("add-habit-button")

// AgentView — send button (currently no label)
Button { ... }
    .accessibilityIdentifier("agent-send-button")

// ChatView+InputBar — send button
Button { ... }
    .accessibilityIdentifier("chat-send-button")

Mock LLM scripted responses

MarketingMockLLMService returns these token streams keyed by the text typed:

Query	Streamed response
`"What's on my calendar tomorrow?"`	`"You have 3 events tomorrow: Team Standup at 9 AM, Lunch with Alex at 12:30 PM, and Gym at 6 PM."`
`"Schedule dentist Friday at 3pm"`	`"Done! Added \"Dentist\" on Friday at 3:00 PM → 4:00 PM."`
`"When am I free this afternoon?"`	`"You have a 2-hour gap from 2–4 PM today. Nothing scheduled."`
`"Remind me to call mom at 6pm"`	`"Reminder set for 6:00 PM — \"Call mom\"."`

Token delay: 22ms between tokens. Gives natural typing feel without being slow.

Per-scene choreography

`testOnboardingFlow` — 8s

Goal: show how fast setup is. Viewer sees 3 info pages then lands on agent tab.

0.0s  app.launchArguments += ["--scene", "onboarding"]
      app.launch()
      // Onboarding page 1 visible: "Your Pocket AI"
1.2s  sleep(1.2)                          // viewer reads page 1
      app.buttons["Continue"].tap()
1.8s  sleep(0.6)                          // page 2 transition
      // Page 2: "Knows Your Schedule"
2.8s  sleep(1.0)                          // viewer reads page 2
      app.buttons["Continue"].tap()
3.4s  sleep(0.6)
      // Page 3: "Type or Speak"
4.4s  sleep(1.0)                          // viewer reads page 3
      app.buttons["Continue"].tap()
5.0s  sleep(0.6)
      // Model selection page — mock LLM already loaded, "Get Started" visible
5.8s  sleep(0.8)                          // viewer sees "no download needed"
      app.buttons["Get Started"].tap()
6.5s  sleep(1.5)                          // agent tab appears — REVEAL moment
      // End

`testAgentFlow` — 8s

Goal: show agent answering a calendar query instantly. Showcase on-device speed.

0.0s  app.launchArguments += ["--scene", "agent"]
      app.launch()
      // Agent tab visible, orb idle, seeded calendar data in background
1.2s  sleep(1.2)                          // viewer registers orb + abilities
      app.textFields["Ask the agent…"].tap()
1.6s  sleep(0.4)                          // keyboard slides up
      app.typeText("What's on my calendar tomorrow?")
3.2s  sleep(0.4)                          // viewer sees typed query
      app.buttons["agent-send-button"].tap()
      // Mock LLM streams response tokens at 22ms each (~1.8s for full response)
5.4s  sleep(2.2)                          // response finishes streaming
      // Calendar event cards rendered below response
6.8s  sleep(1.4)                          // viewer reads the answer
      // End

`testChatFlow` — 8s

Goal: show private chat. Emphasise "no cloud" — tap send, answer appears, no spinner.

0.0s  app.launchArguments += ["--scene", "chat"]
      app.launch()
      // Chat tab, empty session, clean slate
1.0s  sleep(1.0)
      app.textFields["Ask anything…"].tap()
1.4s  sleep(0.4)
      app.typeText("When am I free this afternoon?")
2.8s  sleep(0.3)
      app.buttons["chat-send-button"].tap()
      // Typing indicator briefly visible, then response streams
4.6s  sleep(1.8)                          // response finishes
      // "You have a 2-hour gap from 2–4 PM today."
6.2s  sleep(1.8)                          // viewer reads — no cloud icon, no spinner
      // End

`testNotesFlow` — 8s

Goal: capture an idea fast. Show note created in seconds.

0.0s  app.launchArguments += ["--scene", "notes"]
      app.launch()
      // Notes list with 3 seeded notes visible
1.0s  sleep(1.0)                          // viewer sees existing notes
      app.buttons["new-note-button"].tap()
1.5s  sleep(0.5)                          // NoteEditorView slides up
      app.textFields["Title"].tap()
2.0s  sleep(0.3)
      app.typeText("Product ideas")
3.0s  sleep(0.4)                          // cursor moves to body
      // Type into body (tap below title)
      app.textViews.firstMatch.tap()
      app.typeText("AI-powered habit suggestions based on calendar patterns")
5.5s  sleep(0.6)
      app.buttons["Save"].tap()
6.2s  sleep(0.5)                          // note list reappears
      // New note "Product ideas" visible at top of list
7.5s  sleep(1.3)                          // viewer sees note saved
      // End

`testHabitsFlow` — 8s

Goal: show streaks are alive and logging is one tap.

0.0s  app.launchArguments += ["--scene", "habits"]
      app.launch()
      // Habits list: Morning run (12🔥), Read 20min (7🔥), Sleep 8h (3🔥)
1.2s  sleep(1.2)                          // viewer registers streaks
      // Tap "Morning run" habit card (first item in list)
      app.collectionViews.cells.firstMatch.tap()
2.0s  sleep(0.8)                          // HabitDetailView slides in, streak calendar visible
      // Viewer sees streak calendar filled in
3.5s  sleep(1.5)
      app.buttons["Log"].tap()
4.2s  sleep(0.4)                          // success animation plays
      // Checkmark / celebration animation
5.5s  sleep(1.3)                          // viewer sees it logged
      // Streak counter increments to 13
6.5s  sleep(1.0)
      // End

`testLiveNotesFlow` — 10s

Goal: show real-time transcription. Microphone active, text streams in live.

0.0s  app.launchArguments += ["--scene", "live-notes"]
      app.launch()
      // Notes list visible
1.0s  sleep(1.0)
      // Tap "Start Live Note" button (onStart action in NotesListView)
      app.buttons.matching(NSPredicate(format: "label CONTAINS 'Live'")).firstMatch.tap()
1.6s  sleep(0.6)                          // live notes recording view opens
      // Microphone active indicator, waveform animating
      // Mock transcription streams in: "Team meeting notes — Q2 roadmap discussion..."
3.0s  sleep(4.0)                          // viewer watches text appear in real time
      // Full sentence visible: "Next sprint: focus on onboarding improvements and live note export."
7.5s  sleep(1.5)                          // viewer reads the transcribed text
      app.buttons["Cancel"].tap()         // or let it run to clip end
      // End

`setUp` / boilerplate

swift

final class MarketingVideoTests: XCTestCase {
    var app: XCUIApplication!

    override func setUp() {
        super.setUp()
        continueAfterFailure = true
        app = XCUIApplication()
        app.launchArguments = ["--marketing-demo-mode"]
    }

    private func launch(scene: String) {
        app.launchArguments += ["--scene", scene]
        app.launch()
    }
}

Phase 4 — `scripts/gen-videos.sh`

Mirrors gen-screenshots.sh structure. Same flags:

bare → generate + copy
--only copy → copy last encoded files without re-running tests
--deploy-dev → generate + copy + build + deploy to dev.lucidpal.pages.dev
--scene <name> → regenerate a single scene only

Key implementation details

Recording: use SIGINT not SIGTERM to stop xcrun simctl io — MP4 container needs a clean close. SIGTERM produces a malformed file.

bash

xcrun simctl io booted recordVideo --codec=h264 --force "/tmp/lp_${name}.mp4" &
REC_PID=$!
sleep 0.5   # let recorder initialize before xcodebuild starts

xcodebuild test -project "$PROJ" -scheme "$SCHEME" \
  -destination "$DEST" \
  -only-testing:"LucidPalUITests/MarketingVideoTests/${test}"

kill -INT $REC_PID
wait $REC_PID 2>/dev/null || true

Encoding: trim 1.5s off head (app launch flash), encode to both formats.

bash

# WebM VP9 — smaller, all modern browsers
ffmpeg -ss 1.5 -i "/tmp/lp_${name}.mp4" \
  -vf "scale=390:844,fps=30" \
  -c:v libvpx-vp9 -crf 35 -b:v 0 -an \
  "$WEBSITE_DIR/${name}.webm" -y

# MP4 H.264 — Safari fallback, faststart for HTTP partial content
ffmpeg -ss 1.5 -i "/tmp/lp_${name}.mp4" \
  -vf "scale=390:844,fps=30" \
  -c:v libx264 -crf 28 -preset slow -an \
  -movflags +faststart \
  "$WEBSITE_DIR/${name}.mp4" -y

Prerequisite check:

bash

command -v ffmpeg >/dev/null || { echo "✗ ffmpeg not found — brew install ffmpeg"; exit 1; }
command -v xcodegen >/dev/null || { echo "✗ xcodegen not found — brew install xcodegen"; exit 1; }

Target file sizes

Clip	Duration	WebM	MP4
onboarding	8s	~380KB	~580KB
agent	8s	~380KB	~580KB
chat	8s	~380KB	~580KB
notes	8s	~380KB	~580KB
habits	8s	~380KB	~580KB
live-notes	10s	~480KB	~720KB
Total		~2.4MB	~3.6MB

Total committed: ~6MB. Within Cloudflare Pages limits, no LFS needed.

Phase 5 — Website: `lp-phone-video` Component

New standalone Angular component.

src/app/shared/components/lp-phone-video/
  lp-phone-video.component.ts
  lp-phone-video.component.html
  lp-phone-video.component.scss

State

typescript

const SCENES = [
  { id: 'onboarding', label: 'Setup in 30 seconds', src: '/videos/onboarding' },
  { id: 'agent', label: 'Ask your AI anything', src: '/videos/agent' },
  { id: 'chat', label: 'Fully private chat', src: '/videos/chat' },
  { id: 'notes', label: 'Capture every thought', src: '/videos/notes' },
  { id: 'habits', label: 'Build real habits', src: '/videos/habits' },
];

activeIndex = signal(0);
transitioning = signal(false);
progress = signal(0); // 0–1, drives thin progress bar
showFallback = signal(false); // prefers-reduced-motion

A/B Crossfade

Two stacked <video> elements. On scene change:

Set next video src, call .load()
Start playing (opacity 0)
transitioning = true → CSS transition: opacity 300ms ease swaps them
After transition: swap roles, transitioning = false

No library. Pure CSS transition.

Preload Strategy

When current video reaches 50% playback, inject <link rel="preload" as="video"> for next scene. Next video is in browser cache before crossfade begins.

Browser Compatibility

html

<video autoplay muted playsinline>
  <source [src]="scene.src + '.webm'" type="video/webm" />
  <source [src]="scene.src + '.mp4'" type="video/mp4" />
</video>

WebM VP9 first (Chrome, Firefox, Edge). MP4 H.264 fallback (Safari).

Guards

typescript

// SSR: skip entirely, render static screenshot fallback
if (!isPlatformBrowser(this.platformId)) return;

// Reduced motion: show static phone mockup (existing screenshots)
if (window.matchMedia('(prefers-reduced-motion: reduce)').matches) {
  this.showFallback.set(true);
  return;
}

// Battery: pause when tab hidden
document.addEventListener('visibilitychange', () =>
  document.hidden ? this.pauseAll() : this.resumeCurrent(),
);

Phase 6 — Website: Hero Integration

Swap static .phone-screen HTML in lp-hero.component.html for <lp-phone-video />.

Remove: .chat-bubble divs, .phone-cal-card, .phone-input, .phone-status, .phone-header — all replaced by video content.

Keep: .phone shell div (hardware frame), .orbit-ring-1/2 (CSS, now ambient/subtle), .hero-float-cards.

Reduce orbit ring animation speed and opacity once video is the focal point.

Phase 7 — Three.js Enhancements (separate PR)

Independent of Phase 6. Can ship any time after the video player is confirmed working.

Hero neural net

Replace #particle-canvas (custom 2D canvas) with Three.js Points + LineSegments. Adds true z-axis depth and GPU-accelerated rendering. Reuses existing mouse/scroll signal wiring.

Phone orbital rings

Replace CSS .orbit-ring-1/2 with Three.js 3D orbital planes at 40° and 70° inclination. Nodes travel ellipses with z-depth. Rendered on a position: absolute canvas behind the phone div — nodes naturally appear "behind" the phone frame.

Both use dynamic import('three') — single shared chunk, loads only in browser, zero SSR impact.

Dependency Graph

Phase 1 (UITest target in project.yml)
    └── Phase 3 (MarketingVideoTests) needs Phase 1 + 2

Phase 2 (iOS infra) ──────────────────────── independent, start now
Phase 5 (website component) ──────────────── independent, start now

Phase 3 (XCUITests) ── needs Phase 1 + 2
Phase 4 (gen-videos.sh) ── needs Phase 3 working + ffmpeg

Phase 6 (hero integration) ── needs Phase 5 + video files from Phase 4
Phase 7 (Three.js) ── fully independent, any time

File Inventory

New iOS files

File	Notes
`Sources/Marketing/MarketingEnvironment.swift`	`#if DEBUG`
`Sources/Marketing/MarketingDataSeeder.swift`	`#if DEBUG`
`Sources/Marketing/MarketingMockLLMService.swift`	`#if DEBUG`
`Sources/Marketing/MarketingTapOverlayWindow.swift`	`#if DEBUG`
`UITests/MarketingVideoTests.swift`	XCUITest target

Modified iOS files

File	Change
`project.yml`	Add `LucidPalUITests` target + scheme entry
`Sources/App/LucidPalApp.swift`	`makeLLMService()` factory + scene nav
`Sources/App/AppDelegate.swift`	Wire tap overlay window in `#if DEBUG` block

New website files

File	Notes
`src/app/shared/components/lp-phone-video/lp-phone-video.component.ts`
`src/app/shared/components/lp-phone-video/lp-phone-video.component.html`
`src/app/shared/components/lp-phone-video/lp-phone-video.component.scss`
`public/videos/onboarding.webm` + `.mp4`	Binary, ~960KB
`public/videos/agent.webm` + `.mp4`	Binary, ~960KB
`public/videos/chat.webm` + `.mp4`	Binary, ~960KB
`public/videos/notes.webm` + `.mp4`	Binary, ~960KB
`public/videos/habits.webm` + `.mp4`	Binary, ~960KB
`public/videos/live-notes.webm` + `.mp4`	Binary, ~1.2MB

Modified website files

File	Change
`src/app/sections/lp-hero/lp-hero.component.html`	Swap phone content for `<lp-phone-video />`
`src/app/sections/lp-hero/lp-hero.component.ts`	Import component, lighten orbit ring styles
`src/app/sections/lp-hero/lp-hero.component.scss`	Reduce orbit ring opacity/speed

New scripts

File	Notes
`scripts/gen-videos.sh`	Mirrors `gen-screenshots.sh`

Current Status

Phase	Status
Phase 1 — UITest target in project.yml	✅ Done
Phase 2 — iOS marketing infrastructure	✅ Done
Phase 3 — MarketingVideoTests.swift	✅ Done
Phase 4 — gen-videos.sh	✅ Done
Phase 5 — lp-phone-video Angular component	✅ Done
Phase 6 — Hero integration	⬜ Not started
Phase 7 — Three.js enhancements	⬜ Not started

Branch: feat/llm-control-plan-gemini (active branch when plan was written).

Risks

Risk	Mitigation
XCUITest timing flaky	`continueAfterFailure = true`; mock LLM makes responses deterministic
MP4 malformed on stop	Use `SIGINT` not `SIGTERM` to stop recorder
Personal data in videos	`MarketingDataSeeder` wipes real stores, seeds only fake content
`#if DEBUG` leaks to prod	All marketing code in `#if DEBUG` blocks; App Store = Release config
Video autoplay blocked	`autoplay muted playsinline` — muted autoplay is universally allowed
SSR plays video on server	`isPlatformBrowser` guard in `afterViewInit`
Cloudflare Pages file limit	~6MB total well within 25MB per-file limit

Marketing Video System ​

Architecture Overview ​

Scenes ​

Implementation Phases ​

Phase 1 — Add XCUITest Target to project.yml ​

Phase 2 — iOS Marketing Infrastructure (#if DEBUG only) ​

MarketingEnvironment.swift ​

MarketingDataSeeder.swift ​

MarketingMockLLMService.swift ​

MarketingTapOverlayWindow.swift ​

Scene Navigation ​

Phase 3 — MarketingVideoTests.swift (XCUITest) ​

Known element identifiers (from source reading) ​

Mock LLM scripted responses ​

Per-scene choreography ​

testOnboardingFlow — 8s ​

testAgentFlow — 8s ​

testChatFlow — 8s ​

testNotesFlow — 8s ​

testHabitsFlow — 8s ​

testLiveNotesFlow — 10s ​

setUp / boilerplate ​

Phase 4 — scripts/gen-videos.sh ​

Key implementation details ​

Target file sizes ​

Phase 5 — Website: lp-phone-video Component ​

State ​

A/B Crossfade ​

Preload Strategy ​

Browser Compatibility ​

Guards ​

Phase 6 — Website: Hero Integration ​

Phase 7 — Three.js Enhancements (separate PR) ​

Hero neural net ​

Phone orbital rings ​

Dependency Graph ​

File Inventory ​

New iOS files ​

Modified iOS files ​

New website files ​

Modified website files ​

New scripts ​

Current Status ​

Risks ​

Marketing Video System

Architecture Overview

Scenes

Implementation Phases

Phase 1 — Add XCUITest Target to `project.yml`

Phase 2 — iOS Marketing Infrastructure (`#if DEBUG` only)

`MarketingEnvironment.swift`

`MarketingDataSeeder.swift`

`MarketingMockLLMService.swift`

`MarketingTapOverlayWindow.swift`

Scene Navigation

Phase 3 — `MarketingVideoTests.swift` (XCUITest)

Known element identifiers (from source reading)

Mock LLM scripted responses

Per-scene choreography

`testOnboardingFlow` — 8s

`testAgentFlow` — 8s

`testChatFlow` — 8s

`testNotesFlow` — 8s

`testHabitsFlow` — 8s

`testLiveNotesFlow` — 10s

`setUp` / boilerplate

Phase 4 — `scripts/gen-videos.sh`

Key implementation details

Target file sizes

Phase 5 — Website: `lp-phone-video` Component

State

A/B Crossfade

Preload Strategy

Browser Compatibility

Guards

Phase 6 — Website: Hero Integration

Phase 7 — Three.js Enhancements (separate PR)

Hero neural net

Phone orbital rings

Dependency Graph

File Inventory

New iOS files

Modified iOS files

New website files

Modified website files

New scripts

Current Status

Risks