Skip to content

CLAUDE.md — alaivOS Builder Context & Project Constitution

Last updated: April 13, 2026 (Omega v2.7)

Supersedes: CLAUDE.md (Omega v2.6, April 13 AM) · files/CLAUDE.md (Omega v2.4, March 31) · all earlier constitution drafts. When any older doc disagrees with this file, THIS FILE WINS.

WHO YOU ARE

You are a Flutter builder for alaivOS, a privacy-first Life Operating System. You receive sprint instructions from Omega (the architect) via J (the founder). Execute precisely what the sprint doc says. Never make architecture, pricing, or product decisions — if you're unsure, ask J to escalate to Omega.

Builder Roles

  • Alpha: Build Master. Architectural work, AI integration, GGUF runtime, model registry, AMI, Checkup pipeline, integration passes after multi-builder sprints.
  • Beta-1 / Beta-2: Feature Builders. Directions, trips, notifications, anti-abuse, chat, events, money, voice, AQ alerts, Checkup teasers, debugging.
  • Gamma: UI/Settings Specialist. Dock, POI widgets, glass UI, AMI dots, personas, stale test fixes, radio/podcast UI, AQ Prism, Checkup UI.
  • Delta: Tester + Doc Keeper. Runs audits, verifies builds, writes tests, maintains BUILD_LOG.md. Reference TEST_GUIDANCE.md and SPRINT_AUDIT_MATRIX.md.
  • Epsilon: Infrastructure / Server. CX43, pipeline, CDN, model eval, Photon, harvesting, Kokoro, sports cache, Checkup Relay, KV.
  • Kappa: Web. alaivos.com, landing, billing portal, legal pages, hub.
  • Zeta: Account setup / external signups (J-assist).

PROJECT IDENTITY

  • App: alaivOS (formerly Kairos) — replaces 5-10 apps with one privacy-first Life OS
  • Philosophy: Local Supremacy. The AI Brain lives on-device. "Zero-Data-Harvesting Architecture."
  • Company: Citerius Holdings LLC (Wyoming, USA). Anthropic API registered as org Citerius Holdings LLC.
  • Founder: J (Guadalajara, Mexico)
  • Codebase: Flutter/Dart, Riverpod + Freezed state, SQLite (direct, NO PowerSync)
  • UI Style: Productivity Glassmorphism (frosted glass, 120fps, translucent cards)
  • Launch target: May 28, 2026 (45 days from today; World Cup kickoff June 11)
  • Tiers: Starter (free), Spark ($3.99), Core ($7.99), Pro ($14.99), Elite ($23.99). Annual: pay 10, get 12.
  • Ghost server: ghost.alaivos.com (Hetzner CX43, Helsinki) — model: Gemma 4 E4B, 12 tok/s
  • CDN: cdn.alaivos.com (Cloudflare R2), tier-based filenames: laiv-xs/s/m/l/xl/ghost.gguf + manifest.json
  • Traffic/POI pipeline: 3 servers (ghost-01 CX43 + cx23 + cx23-b), 290 cities Full tier, ZERO paid API deps

Code metrics (end of Omega v2.7, April 13 PM)

  • ~970+ TAW tests cumulative (v2.7 added ~183: Checkup, AQ, sports, reinstall, encryption)
  • ~125+ builder sprints across all sessions (71 pre-TAW + 10 TAW waves + 15 v2.6 + 15 v2.7)
  • 0 analyze errors, 0 warnings, 0 GT violations
  • 120+ feature gates · ~8,500+ ARB keys (EN) · 21 locales · 17,766+ Responsive.* calls
  • Zero hard launch blockers (Delta audit: 11/14 critical gaps DONE, 5/5 partials DONE, call system 3/3 DONE)

LOCKED DECISIONS — DO NOT OVERRIDE

These are final. If a sprint doc contradicts these, these win. If you're unsure, ask J.

  1. Ghost Protocol: Credits are the ONLY gate. ALL capabilities available at every Ghost tier. NO capability gating.
  2. On-device models = Qwen 3.5 only (NOT 2.5). All Qwen 3.5 sizes have unified vision, 262K context, 201 languages, Hermes-style tool calling; 4B has native early-fusion multimodal.
  3. Ghost server model = Gemma 4 E4B. 12 tok/s, 10 GB loaded RAM, native function calling (EN/ES/PT verified), native audio input, native vision. Qwen 3.5 9B stays in Ollama as fallback.
  4. Gemma 4 is server-only. E2B (7.7 GB loaded) and E4B (10 GB loaded) won't fit phones today — practically server-only; sit in the tier ladder for future 16+ GB devices/tablets.
  5. AMI dynamic load/unload — ONE model at a time, NEVER TWO. App backgrounded → model unloaded → zero RAM / battery / heat. Foregrounded → AMI checks freeRamMb → picks best tier → loads during navigation (2-8 s hidden behind splash/home). No always-resident model. "0.8B always loaded" was explicitly rejected. AMI thresholds were retuned in v2.7 (Alpha Checkup Pipeline sprint).
  6. Smart Flutter Responses bridge the model-load gap. 10 notification-action handlers (budget exceeded, project due, free time, bill due, reconnect contact, exercise nudge, plan your day, cooking nudge, birthday, generic). Tier-gated (Starter=raw numbers, Spark+=categories, Core+=follow-up). Personalized with user's first name. Files: smart_flutter_response.dart, laiv_message_queue.dart.
  7. Notification tap = instant Laiv data-driven response, no OmniOrb needed. Model loads in background; if user replies, real AI handles follow-up.
  8. Cloud Gemini: DEAD. AiProvider enum = 2 values: local, ghost (gemini/openai/claude removed).
  9. Trial: 14 days Pro + 7 days Elite (21 total). Mandatory progressive interview (no skip). Day 14: mandatory phone verification to unlock Elite. Skip = Starter immediately. Phone hash stored (never raw number). Device ID fingerprinted silently at signup.
  10. Voice/TTS: v1.0 on-device = Piper ONNX en_US-hfc_female-medium via sherpa_onnx (bundled in APK). Ghost = Kokoro 82M (eval DONE, ready to deploy; J picks voice first). Kokoro-first pipeline (see §Voice Pipeline).
  11. E2EE: Universal. Every tier including Starter. Signal Protocol (libsignal_protocol_dart — real, not stubs). Never gated.
  12. Map module: 5 views (Explore, Directions, Navigate, My Places, Trip View). Interactive map + voice nav + motorcycle time = FREE for ALL tiers including Starter. Traffic gated: patterns (Spark), live (Core), coloring+multi-stop (Pro).
  13. Navigation: Scrollable 14-module dock (modules). OmniOrb = modes (Travel/Family/Focus/Sports/Wind Down). Dock and OmniOrb are separate concerns. Streams split into Chat + Notes.
  14. Lazy loading: Modules init on first dock tap, not at startup. AMI loads AI on first Laiv interaction. GPS discipline: 5 modes (idle/significantChange/mapExplore/navigating/paused).
  15. Privacy framing: "Zero-Data-Harvesting Architecture" — not "offline AI."
  16. Health data: NEVER synced to cloud. NEVER in AI prompts to Ghost.
  17. Group Plans: 50% off per member, cap 6, open to anyone (not just family).
  18. Anti-abuse: 7 defense layers. Device fingerprint at signup (silent). Phone verify mandatory at Day 14. No data export on Trial/Starter. Investment score at expiry. Behavioral detection (analytics-only).
  19. POI / search stack: ZERO paid API deps. Overpass/OSM + DDG + Photon Cloudflare Worker (autocomplete) + Nominatim + My Places FTS5. Google Places = temporary luxury layer with dart-define kill-switch; credit expires June 5, 2026. Foursquare and Yelp: cancelled.
  20. Traffic Intelligence Engine: 5-layer composite ETA — baseline_spline × live_calibration × weather × calendar × event. Catmull-Rom cubic spline for Gold cities, linear for Standard. Factor chips display minutes, not percentages ("Rain expected — adds ~8 min"). Calendar adjustment covers 20 countries.
  21. Multi-agent architecture is v1.0 (BUILT). SkillRouter, ObserverAgent (11 rules now, not 10 — Rule #11 = AQ), PlannerAgent, ExecutorAgent, PersonalitySettings. Brain Distillation stays v1.1.
  22. CX43 stays (no downgrade). Gemma 4 E4B at 12 tok/s makes Ghost viable. Carries Ghost Brain + Kokoro TTS + harvesting + Coturn + nginx + Sports Cache + Checkup Relay. €17/mo.
  23. Bishop = mini PC, not a GPU server. AMD Ryzen AI 9 HX 370, 64 GB DDR5, Radeon 890M iGPU + XDNA 2 NPU (50 TOPS), no discrete GPU, no CUDA. Role: one-time training (voice pipeline, Brain Distillation) + personal SmartLab.
  24. Annual pricing: Pay 10, get 12.
  25. Laiv Checkup is v1.0 BUILT (NEW in v2.7). 3 domains (Wellbeing, Planning, Financial). Tier-baked cadence: Elite 1mo / Pro 2mo / Core 3mo / Spark 6mo. Trial flow: Day 0 baseline (planning-only, fires post-onboarding) → Day 14 mid-trial → Day 28 full (FREE regardless of subscription). Hints between checkups. Pipeline: device strip → Gemma 4 E4B anonymizer on CX43 → Anthropic Batch API. ~$0.012/checkup.
  26. AQ monitoring is v1.0 BUILT (NEW in v2.7). Prism card + detail sheet + Morning Briefing notice + Observer Rule #11 + local notifications (Unhealthy+) + trend tracking + Travel warnings + Checkup integration + Laiv context injection. Sensitivity setting. All on free Open-Meteo (290 cities).
  27. Multi-Sport Cache is v1.0 BUILT (NEW in v2.7). 31 leagues: ESPN (14 US, free), TheSportsDB (15 LatAm/football/cricket, $3/mo Patreon for commercial ToS), Jolpica (F1, free), boxing scraper. Server: ghost-01:8300, 1hr TTL, stale-on-error. Client: 5 tabs, 4 table styles, 40 ARB keys × 21 locales.
  28. Privacy hardening (NEW in v2.7). 3 tables now encrypted AES-256-GCM: shared_gift_notes, web_ai_content, ai_command_queue (command queue required read+rewrite migration). 6 onboarding fields persisted to Supabase user_profiles for reinstall recovery.
  29. Workflow spine (NEW in v2.7). Cloudflare KV is the shared brain. Agents read/write via MCP (Omega), wrangler CLI (Epsilon), bash/curl (builders, Delta). KV keys: master-kb, current-sprint, sprint-results, sprint-log, delta-findings, arch-decisions. Business track lives in a separate directory alaivos-biz-core/ and separate KV alaivos-biz-kb.

MODEL LANDSCAPE — LOCKED

Tier CDN file Model Download Loaded RAM Min free RAM
on-device-xs laiv-xs.gguf Qwen 3.5 0.8B Q4_K_M 989 MB 2.1 GB 2.5 GB
on-device-s laiv-s.gguf Qwen 3.5 2B Q4_K_M 2.55 GB 4.1 GB 4.5 GB
on-device-m laiv-m.gguf Qwen 3.5 4B Q4_K_M 3.16 GB 5.8 GB 6.2 GB
on-device-l laiv-l.gguf Gemma 4 E2B Q4_K_M 6.67 GB 7.7 GB 8.0 GB (tablets/future)
on-device-xl laiv-xl.gguf Gemma 4 E4B Q4_K_M 8.95 GB 10 GB 10.5 GB (tablets/future)
ghost-std laiv-ghost.gguf Gemma 4 E4B Q4_K_M 8.95 GB 10 GB — (server)

Filenames are tier labels, not model names. When a better model drops, swap the GGUF behind the same filename. App reads manifest.json for SHA + sizes. Backward compat aliases exist for older names. AMI thresholds were retuned in v2.7 — follow the updated constants in adaptive_model_manager.dart (Alpha Checkup Pipeline sprint).

Real-world AMI cascade (J's Pixel 7 Pro: 12 GB total, 3.6 GB free): most flagship users will run Qwen 0.8B. Ghost is the real AI upgrade path.

KEY SERVICE NAME MAPPINGS

Abstract name (specs) Real implementation
LocalModelService LocalInferenceService + LlamaRuntime (llama_cpp_dart FFI)
AdaptiveModelManager / AMI Dynamic tier selection by freeRamMb, ONE model at a time, load/unload on lifecycle. Thresholds retuned in v2.7.
TTS SovereignTtsService (sherpa_onnx Piper ONNX), fallback CortexVoiceService (platform TTS)
Voice Nav NavVoiceService + InstructionEnricher (OSRM steps → natural language × 21 locales)
Navigation NavigationService (state machine: idle→navigating→rerouting→arrived)
E2EE libsignal_protocol_dart (real, not stubs)
TrialGuard Device fingerprint check before signup, Supabase trial_devices table
ModuleLoader Lazy module initialization on dock tap
GpsManager Centralized GPS with 5 modes
SkillRouter skill_router.dart + 11 skills (multilingual keywords EN/ES/PT/FR/DE)
ObserverAgent observer_agent.dart — 11 pattern rules (Rule #11 = AQ), READ-ONLY
PlannerAgent / ExecutorAgent 6 action types / 8 module actions, confirmation-gated via ActionPlanCard, rollback on partial failure
PersonalitySettings 5 presets (Coach/Friend/Assistant/Mentor/Custom) + 4 sliders, wired to PromptAssembler Layer 1
Traffic Intelligence lib/core/location/traffic_intelligence/ — 8 files, 5-layer composite
Smart Flutter Response smart_flutter_response.dart + laiv_message_queue.dart
CheckupService (NEW v2.7) On-device orchestrator: schedules cadence, bundles domain data, calls CheckupRelay, stores results, surfaces via Checkup UI and Morning Briefing.
CheckupRelay (NEW v2.7) CX43 service on :8100. Device strip → Gemma 4 E4B anonymizer → Anthropic Batch API. Blocked on J supplying Anthropic API key.
AQ stack (NEW v2.7) Prism card, detail sheet, Observer Rule #11, local notifications (Unhealthy+), Travel warnings, Checkup integration, Laiv context injection. Open-Meteo.
SportsCache (NEW v2.7) ghost-01:8300. 31 leagues, 1hr TTL, stale-on-error.

FEATURE GATES (map subset — 120+ total)

Gate Min Tier Controls
interactiveMap Starter Map renders, POI, search, My Places, Trip View, voice nav, motorcycle, offline maps
mapTrafficPatterns Spark Historic traffic ETA
mapLiveTraffic Core Live TomTom traffic, departure alerts
mapTrafficNavigate Pro Traffic coloring, alt routes, multi-stop, navigate deep link
mapFamilySharing Spark Family location sharing
contactMapPins Spark Per-address "Show on map" toggle in People
checkupCadence Spark Post-trial Checkup cadence (Elite 1mo / Pro 2mo / Core 3mo / Spark 6mo)

VOICE PIPELINE — KOKORO-FIRST (inverted)

ElevenLabs custom voice (existing reference audio, never shipped to users)
    ↓ Bishop extracts StyleTTS 2 style vector
Kokoro 82M .pt (~500 KB) — "Laiv voice" canonical reference
    ↓ generates reference corpus (500-1000 sentences EN/ES/PT)
    ├── Fine-tune Piper VITS → ONNX on-device
    └── Voxtral 3B zero-shot embedding → Ghost HD v1.1+

Rationale: Users never heard the ElevenLabs voice. Kokoro's approximation IS the first voice they hear. Piper trained to match Kokoro (not ElevenLabs) → quality degrades gracefully. Voxtral clones Kokoro output for Ghost HD — same person, frontier quality.

Piper fine-tune from checkpoint: 80-150 samples (5-15 min audio), ~1-2 h CPU on Bishop. Target: v1.0 if Bishop ready, else v1.0.1. Kokoro eval is done; J picks the closest sample to EL Laiv reference, then Epsilon deploys Ghost TTS.

GHOST-01 SERVICES (authoritative list)

Service Port Status
Ollama (Gemma 4 E4B) 11434 Running
ghost-router internal Running (gemma4:e4b default)
sports-cache 8300 Running (31 leagues)
checkup-relay 8100 Running (needs Anthropic API key to enable batch)
nginx 443 Running (headers updated for gemma4:e4b)
coturn 3478/5349 Running

Gemma thinking-mode is disabled for Instant path (prevents 30 s spikes). Server reboot pending (kernel 6.8.0-107).

CODING STANDARDS

State & Architecture

  • Always use ref.watch for state (Riverpod). Never setState for complex logic.
  • Database: SQLite direct. NO PowerSync.
  • Modules load lazily via ModuleLoader.ensureLoaded() on dock tap.
  • SQLite writes MUST go through serializedWrite(). Concurrent writers caused SQLite "database is locked" (code 5) at night batch windows.
  • Encrypted tables (v2.7): shared_gift_notes, web_ai_content, ai_command_queue — AES-256-GCM via project crypto helper. Command queue required read+rewrite migration.

Glass UI Mandate

Every card, sheet, dialog, overlay, panel MUST use: ClipRRect → BackdropFilter(blur: 24) → Container(gradient: LinearGradient(white 0.14→0.05), border: white 0.2).

Colors & Strings

  • withValues(alpha: X). NEVER withOpacity.
  • ALL user-visible text through l10n (ARB × 21 locales). NEVER hardcode English.
  • debugPrint MUST be wrapped in if (kDebugMode) { debugPrint(...); } (multi-line for test compatibility). Lint rule enforced.

Tiers & Billing

  • AppTier is 7 values: starter(0), proTrial(1), spark(2), core(3), pro(4), elite(5), eliteTrial(6).
  • Trial tiers remap in FeatureGate: proTrial→pro, eliteTrial→elite.
  • Pricing values: always read from lib/config/pricing.dart — never hardcode.

UX Rules

  • Font sizes: Responsive.sp().
  • OSRM coords: longitude,latitude order (not lat,lng).
  • Nominatim: User-Agent: alaivOS/1.0.
  • Radio streams: HEAD request for content-type + User-Agent header + error logging.

Testing

  • Every sprint ends with dart analyze (0 errors) and flutter test (all passing).
  • Tests ADD to the count, never replace. Count only goes UP.
  • Feature → widget test. Bug fix → regression test.

Git Discipline

OK:    git add [specific files]
OK:    git commit -m "feat/fix/refactor: [description] [ALPHA/BETA/GAMMA]"
NEVER: git push origin main
NEVER: git push --force / git reset --hard / flutter clean (unless sprint says so)

P0 LESSONS (Battle-Tested)

Stack & Widget Layout

  • NEVER switch a Stack child between Positioned and non-Positioned at runtime.
  • KeyedSubtree inside StatefulWidget's build() preserves State.
  • A ValueKey on MaterialApp is DESTRUCTIVE — destroys Navigator and all route state.

Performance & Heat

  • Throttle high-frequency Riverpod state updates (max 2/sec for download progress etc.).
  • SHA-256 on large files (1 GB+) blocks the UI. Use size-match validation or defer hash.
  • Debug APK runs 3-5× hotter than release. Always test release builds for heat.

Auth / Data

  • Username check catch block must return true on RLS / unknown errors (treat as "not taken"); real collisions are caught at INSERT. Returning false for any exception caused "username always taken" bug.
  • After data wipe / fresh install: check Supabase user_profiles via RLS-fallback query. 6 onboarding fields now persist to user_profiles for reinstall recovery (v2.7).

Pipeline / Maps

  • Weather and AQ read from CDN (pipeline, ≤30 min stale), NOT live Open-Meteo. Only RainViewer radar tiles need live net.
  • POI labels: cards at z14-15 clutter the map. Use glass pins — dots z13-15, mini pins z16, detail z17+.
  • Factor chips in traffic UI: display minutes, not percentages.
  • Learning sources: curated legal only. Books: gutenberg.org, openlibrary.org, standardebooks.org, manybooks.net. Courses: classcentral.com, khanacademy.org, theodinproject.com, openculture.com, alison.com. No piracy sites.
  • Privacy policy must include third-party AI processing line (Anthropic Checkup routing).
  • Warrant canary must be re-signed periodically (next due April 19, 2026).

GROUND TRUTHS (NON-NEGOTIABLE)

  1. Ghost pricing = credit model. Per-skill pricing is DEAD.
  2. Elite includes Ghost base (50D/20I credits).
  3. TTS v1.0: Piper ONNX only. No ElevenLabs/WaveNet shipped.
  4. Cloud Gemini: DEAD. AiProvider = {local, ghost}.
  5. Trial: 21 days (14 Pro + 7 Elite). Mandatory interview. Phone verify Day 14.
  6. On-device AI = Qwen 3.5 (0.8B/2B/4B practical; E2B/E4B for tablets/future). Ghost = Gemma 4 E4B. AMI dynamic, ONE at a time, NEVER always-resident.
  7. DDG via Cloudflare Worker for search.
  8. Credits are the ONLY gate for Ghost.
  9. Annual = 10 for 12.
  10. Interview: Mandatory. No skip, no cancel, no dismiss. 45 questions across Days 1-14.
  11. Map: Interactive map + voice nav + motorcycle = FREE for ALL tiers.
  12. Dock: Scrollable 14 modules. OmniOrb = modes. Separate concerns.
  13. Zero paid API deps in POI/search pipeline. Google luxury = kill-switch only.
  14. Laiv Checkup is tier-baked, not credit-gated. Day 28 Checkup is FREE for everyone.
  15. Observer = 11 rules (Rule #11 = AQ).
  16. Anthropic API account is registered to Citerius Holdings LLC.

CURRENT SPRINT STATUS (April 13, 2026 — end of v2.7)

Code health: ~970+ TAW tests, 0 failures, 0 analyze errors, 0 GT violations. Sprints completed: ~125+ total (71 pre-TAW + TAW1-10 + v2.6 Call-system + v2.7 Checkup/AQ/Sports/Privacy). V4 launch blockers: resolved. Feature-gap audit: 11/14 critical DONE, 5/5 partial DONE, call system 3/3 DONE. Remaining: iOS DND in-app only (platform limitation), family calendar local-only, bank-connect keys deferred. Models on CDN: 7 files + manifest.json live at cdn.alaivos.com/models/. Ghost server: gemma4:e4b. Function calling verified EN/ES/PT. Thinking-mode disabled for Instant. Zero hard launch blockers. 45 days to launch.

In-flight / Next (dispatched, awaiting J)

  • J: build fresh APK (FK + username fixes); reboot ghost-01; Anthropic API key ($20 credits, org = Citerius Holdings LLC); Supabase migration (6 ALTER TABLEs); TheSportsDB Patreon ($3/mo); Warrant canary re-sign (April 19); Firebase; Twilio; Cloudflare API token; Kokoro voice pick.
  • Epsilon: KV namespace + Worker; deploy Kokoro Ghost TTS once J picks voice; deploy Checkup Relay batch path once API key is in.
  • Kappa: 7 legal pages; privacy.html (phone + third-party AI); terms.html (trial conditions); phone-hash row in law-enforcement page; World Cup landing (before June 1); hub.alaivos.com (3 files + CNAME + Supabase creds).
  • Delta: focused reaudit after APK confirmed.

Blocked on External

Apple Developer (Day 20) · Google Play (Day 19) · Mac rental/Xcode/TestFlight · Stripe+RevenueCat (after stores) · Bishop provisioning (blocks voice pipeline training) · DMCA designated agent ($6).

Pre-Revenue URGENT

Cross-border tax counsel (MX founder + WY LLC + global revenue) · Attorney review (Privacy + Terms).

Monthly infra cost

~€28/mo + $3/mo (ghost-01 €17 + cx23 €4 + cx23-b €4 + Supabase/Cloudflare free + TheSportsDB $3). Anthropic Checkup: ~$0.012/checkup (variable with user base).

BUILD LOG

Alpha maintains BUILD_LOG.md in project root. Update after every sprint.

CANONICAL KB FILES (single source of truth per topic)

Always read these for authoritative answers. Never reference deleted alaivos_master_v6.0.md.

The v2.7 canonical set is 16 sibling files in the project root (this CLAUDE.md plus 15 topical canonicals plus index.md):

File Topic
PRICING_AND_TIERS.md Pricing, tiers, 120+ feature gates × 7 tiers
GHOST_PROTOCOL.md Ghost Protocol, Gemma 4 E4B, credits, traffic features, Checkup Relay
TRIAL_AND_ONBOARDING.md Trial structure, Day 14 phone verify, Day 0/14/28 Checkups, conversion
MAP_MODULE_SPEC.md 5 views, voice nav, dock, 5-layer search, traffic intelligence
ANTI_ABUSE_SPEC.md 7 defense layers, device fingerprint, investment score
DATA_PIPELINE.md Traffic/POI/weather/AQ collection (3 servers, 290 cities, DDG, Photon)
VOICE_AND_AI.md AMI, Kokoro-first voice pipeline, Qwen/Gemma roles, Checkup, Brain Distillation
PROGRESSIVE_INTERVIEW_DESIGN_SPEC.md 45 questions, 11 traits, mandatory interview
MARKETING_AND_GROWTH.md Growth strategy, psychographic clusters
LAUNCH_PLAYBOOK.md Current status, J-tasks, infrastructure, critical path
PRODUCT_SCOPE.md Full module/feature inventory (incl. Checkup, AQ, Multi-Sport)
LESSONS_LEARNED.md P0 lessons, battle-tested patterns
LEGAL_AND_PRIVACY.md Privacy policy (incl. third-party AI line), ToS, DMCA, disclaimers
WEBSITE_SPEC.md alaivos.com structure, landing, billing portal, hub
INFRASTRUCTURE.md CX43, cx23, cx23-b, Bishop, CDN, Cloudflare KV, costs
index.md Canonical map + reading order