# Context Grammar Token Specification # Machine-readable spec for AI systems making design decisions # Source: intentfirst.ai spec: name: Context Grammar Token Specification version: "1.0" date: "2026-04-09" description: "Machine-readable specification for Context Grammar's Intent + 8 Context Tokens" license: "CC BY 4.0" url: "https://intentfirst.ai/spec" architecture_note: > Intent is the starting point. 8 Context Tokens declare the user's current state. Brain (3-layer memory: Identity / Accumulated Learning / Right Now) provides personal context. Rule Engine transforms Token x Brain into concrete design rules. Learning Loop feeds outcomes back into Brain, improving implicit Intent detection over time. # ------------------------------------------------------------------- # Intent — separate from the 8 tokens # ------------------------------------------------------------------- intent: name: Intent name_ja: "Intent (意図)" id: intent role: entry_point description: "The user's goal — what they want to accomplish. Serves as the starting point for all Context Grammar decisions." types: explicit: description: "User-declared intent expressed in words or direct action." examples: - "Find winter shoes for my son" - "Plan a Kyoto trip" - "Start spring cleaning today" detection_method: "Natural language input, search query, direct command, or task initiation." availability: now implicit: description: "Intent inferred from the intersection of Context Tokens and Brain memory, without the user stating it." examples: - "Yogurt at 20%, consumption rate 1 per 5 days, today is Wednesday — order to arrive by Friday." - "Calendar shows dentist at 3pm, traffic building — suggest leaving now." - "Child's shoe size updated 2 months ago, growth rate pattern suggests check soon." detection_method: "Token state + Brain accumulated learning + temporal patterns." availability: partial dependencies: - "Brain memory depth (more history = better inference)" - "Disclosure Dial level (AI must know relevant data to infer)" - "Autonomy Dial level (determines whether inference is acted upon)" design_rules: explicit_intent: response_mode: direct ui_pattern: "Show relevant results immediately. Match output to the specificity of the request." latency_target_ms: 500 implicit_intent: response_mode: contextual ui_pattern: "Surface as a suggestion or ambient notification, never as an interruption. Confidence threshold must be met before surfacing." confidence_threshold: 0.8 fallback: "If confidence < threshold, do not surface. Store as candidate and wait for additional signals." # ------------------------------------------------------------------- # 8 Context Tokens # ------------------------------------------------------------------- tokens: # --------------------------------------------------------------- # SITUATION TOKENS (1-6) # --------------------------------------------------------------- - name: Physical State name_ja: "Physical State (身体状態)" id: physical_state number: 1 category: situation description: "Declares the user's current bodily state — posture, mobility, and hand availability — to determine fundamental UI shape." values: - id: stationary_both_hands label: "Stationary, both hands free" description: "Seated or standing still, both hands available for interaction." - id: stationary_one_hand label: "Stationary, one hand occupied" description: "Seated or standing still, one hand holding something (coffee, bag, child's hand)." - id: walking label: "Walking" description: "On foot, device likely in one hand or pocket." - id: transit_passive label: "Passive transit" description: "On train/bus, possibly standing and holding strap with one hand." - id: driving label: "Driving" description: "Hands on wheel, eyes on road. No manual interaction permitted." - id: exercising label: "Exercising" description: "Running, cycling, gym. Hands largely unavailable, screen glancing only." - id: lying_down label: "Lying down" description: "In bed or on couch. Device likely held above face or propped." signals: - source: "Device motion sensors (accelerometer, gyroscope)" method: "iOS CMMotionActivityManager / Android Activity Recognition API classifies walking, running, cycling, automotive, stationary." availability: now - source: "Smartwatch sensors" method: "Heart rate + accelerometer patterns distinguish exercise from transit. Galaxy Watch blood pressure (US, March 2026)." availability: now - source: "Connected vehicle APIs" method: "CarPlay / Android Auto connection confirms driving state." availability: now - source: "Grip pattern detection" method: "Touch sensor + accelerometer fusion to detect one-hand vs two-hand grip. Research shows 84% accuracy but no shipping product." availability: future design_rules: stationary_both_hands: ui_density: high interaction_complexity: full touch_target_min_dp: 48 recommendation: "Full UI with standard controls. Complex forms, multi-step flows, and detailed content are all appropriate." stationary_one_hand: ui_density: medium interaction_complexity: standard touch_target_min_dp: 56 recommendation: "Bottom-aligned primary actions. Swipe-card UI. Avoid top-corner targets. Increase button size." walking: ui_density: low interaction_complexity: minimal touch_target_min_dp: 64 recommendation: "Large tap targets, glanceable info cards, haptic confirmation. No text input. Audio output preferred." transit_passive: ui_density: medium interaction_complexity: standard touch_target_min_dp: 56 recommendation: "Swipe-card UI optimized for one-hand thumb zone. Scrollable content OK. Avoid precision taps." driving: ui_density: low interaction_complexity: minimal touch_target_min_dp: null recommendation: "Voice-only interaction. No visual UI changes that require attention. Audio confirmation for all actions. HUD-safe glanceable status only." hard_constraint: "NEVER require visual attention or manual interaction while driving state is active." exercising: ui_density: low interaction_complexity: minimal touch_target_min_dp: 72 recommendation: "Wrist-glanceable summaries on watch. Audio cues. Pause/resume as single large target. No scrolling." lying_down: ui_density: medium interaction_complexity: standard touch_target_min_dp: 48 recommendation: "Content-consumption optimized. Auto-rotate awareness. Reduce brightness. Larger text for distance viewing if propped." reality_level: 4 reality_note: "Activity recognition (walk/run/drive/stationary) ships on all major platforms and is mature. One-hand vs two-hand detection and automatic UI reshaping based on physical state remain unshipped." - name: Cognitive Load name_ja: "Cognitive Load (認知負荷)" id: cognitive_load number: 2 category: situation description: "Estimates the user's current mental bandwidth to determine how much information and how many choices the UI should present." values: - id: low label: "Low load" description: "Relaxed, few demands. Open to exploration and complex choices." - id: moderate label: "Moderate load" description: "Normal task engagement. Standard cognitive capacity available." - id: high label: "High load" description: "Multiple demands competing. Limited capacity for new information." - id: overloaded label: "Overloaded" description: "At or past cognitive capacity. Decision-making severely impaired." estimation_method_note: > IMPORTANT: Cognitive Load is NEVER directly measured. It is always estimated from indirect signals. This is a fundamental design decision — no EEG, no biometric brain scanning. Estimation from behavioral and contextual signals only. signals: - source: "Time of day" method: "Late night (23:00-05:00) and post-lunch (13:00-14:00) mapped to higher baseline load via circadian model." availability: now - source: "Calendar density" method: "Count meetings/events in surrounding 2-hour window. 3+ events with <15min gaps = high load signal." availability: now - source: "Recent app-switching frequency" method: "5+ app switches in 10 minutes indicates fragmented attention. Screen Time API / Digital Wellbeing API." availability: now - source: "Previous task complexity" method: "Completing a complex task (tax filing, long form, medical decisions) leaves residual load. Decay over 30-60 minutes." availability: partial - source: "Typing pattern degradation" method: "Increased typo rate, slower typing speed relative to personal baseline. Samsung Brain Health (CES 2026) detects long-term cognitive patterns." availability: partial - source: "Focus Mode state" method: "iOS Focus / Android Focus Mode active = user has self-declared a constrained attention context." availability: now design_rules: low: ui_density: high interaction_complexity: full max_choices: 8 recommendation: "Full exploration UI. Show categories, filters, detailed descriptions. User can handle comparison and nuanced decisions." moderate: ui_density: medium interaction_complexity: standard max_choices: 5 recommendation: "Balanced UI. Show curated options with clear differentiators. Progressive disclosure for details." high: ui_density: low interaction_complexity: minimal max_choices: 3 recommendation: "Simplified UI. Pre-filtered top recommendations. Binary or ternary choices only. Offer 'decide later' as explicit option." overloaded: ui_density: low interaction_complexity: minimal max_choices: 1 recommendation: "Single best recommendation with one-tap accept. 'Save for later' as alternative. No comparisons, no details unless requested. Shield from non-urgent notifications." defer_pattern: "Queue non-critical decisions and resurface when load drops to moderate or below." reality_level: 2 reality_note: "No shipping product performs real-time cognitive load estimation for UI adaptation. Individual signals (calendar, time, Focus Mode) are available today. The composite estimation model and UI response are entirely speculative. Closest analog: Focus Mode as a manual proxy." - name: Social Exposure name_ja: "Social Exposure (社会的露出)" id: social_exposure number: 3 category: situation description: "Declares who can currently see or hear the device output, determining what information is safe to display and in what form." values: - id: private label: "Private" description: "User alone. Full information display is safe." - id: trusted_partner label: "Trusted partner" description: "With spouse/partner. Shared view appropriate. Most personal info OK." - id: family_with_children label: "Family with children" description: "Children present. Price, adult content, and sensitive topics should be filtered." - id: social_acquaintances label: "Social / acquaintances" description: "Friends or colleagues nearby. Professional filter on personal information." - id: public label: "Public" description: "Strangers can see screen. Maximum privacy protection required." signals: - source: "Face detection / Visual ID" method: "Amazon Echo Show Visual ID identifies family members by face. Samsung Smart TV Voice ID identifies by voice. Detects known vs unknown presence." availability: now - source: "Privacy display sensor" method: "Samsung Galaxy S26 Ultra Privacy Display detects nearby people via front-facing sensors and narrows viewing angle automatically." availability: now - source: "Bluetooth/Wi-Fi proximity" method: "Detect known family member devices nearby via BLE beacons or Wi-Fi presence." availability: now - source: "Location context" method: "GPS + venue data. Coffee shop, office, train = likely public. Home address = likely private." availability: now - source: "Audio environment analysis" method: "Ambient noise classification (quiet room vs crowd vs office chatter) as supporting signal." availability: partial design_rules: private: ui_density: high interaction_complexity: full content_filter: none recommendation: "Full information display. Prices, personal data, health info, recommendations all visible. Audio output OK." trusted_partner: ui_density: high interaction_complexity: full content_filter: light recommendation: "Shared view layout when on large screen. Individual surprise/gift items hidden. Joint decision UI available." family_with_children: ui_density: medium interaction_complexity: standard content_filter: moderate recommendation: "Hide prices on gift-related items. Filter age-inappropriate content. Show child-safe alternatives. Parental override available via discrete gesture or auth." hidden_categories: - prices_for_gifts - adult_content - financial_details - medical_details social_acquaintances: ui_density: medium interaction_complexity: standard content_filter: significant recommendation: "Minimize personal information on screen. Use notification previews without content. Professional/neutral presentation." hidden_categories: - financial_information - health_data - personal_messages_preview - relationship_details public: ui_density: low interaction_complexity: minimal content_filter: maximum recommendation: "Privacy mode. No personal info visible. Narrowed viewing angle if hardware supports. Notifications show sender only, no preview. Suggest switching to audio/haptic output." hidden_categories: - all_personal_information - all_financial_data - message_contents - calendar_details reality_level: 4 reality_note: "Privacy Display (Samsung S26 Ultra), Visual ID (Echo Show), Voice ID (Samsung TV), and attention detection (Apple) all ship today. Cross-device unified 'who is watching' protocol and automatic content filtering based on social context do not yet exist." - name: Priority Weight name_ja: "Priority Weight (優先度)" id: priority_weight number: 4 category: situation description: "Resolves collisions when multiple demands compete for the user's attention or the system's resources. Operates at two layers: current urgency (Token) and learned tradeoff patterns (Brain)." values: - id: critical label: "Critical" description: "Time-sensitive and irreversible. Child safety, medical, hard deadlines." urgency_range: [0.9, 1.0] - id: high label: "High" description: "Time-sensitive but recoverable. Pickup times, cooking timers, meeting starts." urgency_range: [0.7, 0.9] - id: standard label: "Standard" description: "Important but flexible. Shopping tasks, meal planning, routine decisions." urgency_range: [0.3, 0.7] - id: low label: "Low" description: "No time pressure. Exploration, wishlist curation, long-term planning." urgency_range: [0.0, 0.3] collision_resolution: description: > When multiple items have competing priority, resolution uses two layers: (1) Token layer — current urgency based on time proximity and irreversibility. (2) Brain layer — learned tradeoff patterns accumulated over time (e.g., "this family prioritizes function over aesthetics", "when rushed, price tolerance increases"). algorithm: > effective_priority = token_urgency * 0.6 + brain_tradeoff_weight * 0.4 If two items score within 0.05, surface both with explicit comparison UI. Irreversible items get +0.2 boost. signals: - source: "Calendar/schedule proximity" method: "Time until next commitment. <15 min = critical time pressure." availability: now - source: "Irreversibility detection" method: "Rule-based classification: flight departure, event start, pharmacy closing = irreversible. Shopping, browsing = recoverable." availability: now - source: "Domain-level priority intelligence" method: "Gmail Priority Inbox (Gemini, 3-5 day learning), Apple Intelligence Priority Notifications (iOS 18.4+), Reclaim.ai calendar optimization." availability: now - source: "Cross-domain priority integration" method: "Unified priority across calendar, shopping, childcare, household tasks. Requires agent orchestration layer." availability: future - source: "Brain tradeoff patterns" method: "Accumulated learning from past decisions: price sensitivity under time pressure, quality vs speed preferences, per-family-member priority structures." availability: partial design_rules: critical: ui_density: low interaction_complexity: minimal recommendation: "Interrupt current flow. Full-screen alert with single primary action. Suppress all non-critical notifications. Countdown if time-bound." notification_level: interrupt high: ui_density: low interaction_complexity: minimal recommendation: "Prominent banner or overlay. Pre-computed best action as one-tap option. Show time remaining. Queue other tasks visibly but deprioritized." notification_level: prominent standard: ui_density: medium interaction_complexity: standard recommendation: "Normal task list ordering. Show relative priority indicators. Allow reordering. Brain-informed sorting (learned preferences influence default order)." notification_level: normal low: ui_density: high interaction_complexity: full recommendation: "Exploration-friendly UI. No time pressure indicators. Rich browsing, comparison, and discovery features. Can be deferred indefinitely." notification_level: ambient reality_level: 3 reality_note: "Per-domain priority (email, calendar) ships today and works well. Cross-domain priority resolution (calendar vs shopping vs childcare vs guests) and agent-to-agent conflict resolution do not exist in any shipping product." - name: Form Factor name_ja: "Form Factor (フォームファクター)" id: form_factor number: 5 category: situation description: "Detects available display surfaces and their capabilities to determine how content should be distributed and shaped across screens." values: - id: phone_handheld label: "Phone (handheld)" screen_size_range: "5-7 inch" input_methods: [touch, voice, gesture] description: "Primary personal device. Touch-first interaction." - id: phone_folded label: "Foldable phone (folded)" screen_size_range: "3-4 inch cover display" input_methods: [touch, voice] description: "Cover display for glanceable info and quick actions." - id: phone_unfolded label: "Foldable phone (unfolded)" screen_size_range: "7-8 inch" input_methods: [touch, voice, stylus, gesture] description: "Tablet-like canvas. Flex Mode splits UI at fold angle." - id: tablet label: "Tablet" screen_size_range: "10-13 inch" input_methods: [touch, voice, stylus, keyboard] description: "Large personal screen. Multi-column layouts appropriate." - id: tv_display label: "TV / Large display" screen_size_range: "40-85 inch" input_methods: [voice, remote, gesture] description: "Shared family screen. Lean-back viewing distance (2-4m)." - id: fridge_display label: "Smart fridge display" screen_size_range: "21-32 inch" input_methods: [touch, voice] description: "Kitchen-anchored. Glanceable info while cooking. Wet/messy hands likely." - id: car_display label: "Car dashboard" screen_size_range: "8-15 inch" input_methods: [voice, limited_touch] description: "Driving context. Voice-primary, glance-safe visuals only." - id: watch label: "Smartwatch" screen_size_range: "1.2-2 inch" input_methods: [touch, voice, crown, gesture] description: "Wrist-glanceable. Notifications and quick actions only." - id: desktop_monitor label: "Desktop monitor (via DeX)" screen_size_range: "13-34 inch" input_methods: [mouse, keyboard, touch] description: "Full desktop experience via Samsung DeX or equivalent." signals: - source: "Device hardware identification" method: "Device model, screen dimensions, and input capabilities reported by OS." availability: now - source: "Connected display detection" method: "Samsung DeX, Apple AirPlay, Google Cast, HDMI connection detection." availability: now - source: "Nearby device discovery" method: "Matter protocol, Bluetooth LE, Wi-Fi Direct for discovering available surfaces in proximity." availability: now - source: "Fold angle sensor" method: "Galaxy Z Fold hinge angle sensor triggers Flex Mode UI split at specific angles." availability: now - source: "Multi-surface orchestration" method: "Dynamic content distribution across all detected surfaces simultaneously (e.g., phone + TV + fridge)." availability: future design_rules: phone_handheld: ui_density: medium interaction_complexity: standard recommendation: "Single-column layout. Bottom navigation. Thumb-zone optimized. When casting to TV, auto-transform to remote-control UI." typography_base_sp: 14 grid_columns: 1 phone_folded: ui_density: low interaction_complexity: minimal recommendation: "Glanceable cards only. Status, notifications, quick replies. No deep navigation." typography_base_sp: 12 grid_columns: 1 phone_unfolded: ui_density: high interaction_complexity: full recommendation: "Multi-pane layout. List-detail pattern. In Flex Mode (partially folded): top half content, bottom half controls." typography_base_sp: 14 grid_columns: 2 tablet: ui_density: high interaction_complexity: full recommendation: "Multi-column layout. Side navigation. Rich media display. Split-screen multitasking support." typography_base_sp: 16 grid_columns: 2-3 tv_display: ui_density: low interaction_complexity: minimal recommendation: "Large image grid. 10-foot UI principles. Focus-based navigation (D-pad). Minimal text, large typography. Shared viewing context — respect Social Exposure token." typography_base_sp: 32 grid_columns: 3-5 fridge_display: ui_density: medium interaction_complexity: standard recommendation: "Large touch targets (hands may be wet/messy). Recipe cards, grocery lists, family calendar. Quick-glance kitchen-relevant info. Voice interaction preferred while cooking." typography_base_sp: 20 grid_columns: 2 touch_target_min_dp: 64 car_display: ui_density: low interaction_complexity: minimal recommendation: "Voice-primary. Glance-safe status only. No scrolling. Pre-computed actions. Inherits driving constraint from Physical State token." typography_base_sp: 24 grid_columns: 1 hard_constraint: "Must comply with NHTSA visual-manual distraction guidelines." watch: ui_density: low interaction_complexity: minimal recommendation: "Notification cards. Single action per screen. Haptic feedback for confirmation. Crown/bezel for scrolling." typography_base_sp: 14 grid_columns: 1 desktop_monitor: ui_density: high interaction_complexity: full recommendation: "Full desktop layout. Multi-window support. Keyboard shortcuts. Dense information display with hover states." typography_base_sp: 14 grid_columns: 3-6 reality_level: 4 reality_note: "Device-to-device handoff (Continuity, DeX, CarPlay/Android Auto) and responsive layouts are mature and shipping. Dynamic multi-surface content distribution across 3+ heterogeneous screens simultaneously is not shipping in any product." - name: Feasibility name_ja: "Feasibility (実現可能性)" id: feasibility number: 6 category: situation description: "Acts as a reality filter — constrains the ideal plan proposed by other tokens against what is actually possible right now in the physical world." values: - id: fully_feasible label: "Fully feasible" description: "All requirements met. Item in stock, within budget, delivery on time, weather OK." - id: partially_feasible label: "Partially feasible" description: "Some constraints met, others require compromise. Substitution or schedule adjustment needed." - id: infeasible label: "Infeasible" description: "Cannot be executed as planned. Alternative approach required." constraint_dimensions: - id: inventory label: "Stock / Availability" description: "Is the item or service available right now?" check_method: "Real-time inventory API, store hours API" - id: budget label: "Budget" description: "Is it within the user's price range or allocated budget?" check_method: "Brain-stored budget preferences, price comparison API" - id: time label: "Time / Delivery" description: "Can it be delivered or completed within the required timeframe?" check_method: "Delivery estimation API, traffic/routing API" - id: weather label: "Weather" description: "Does weather permit the planned activity?" check_method: "Weather API (hourly forecast)" - id: location label: "Geographic access" description: "Is the destination reachable? Is the store/venue nearby?" check_method: "Maps API, routing API, geofencing" - id: regulatory label: "Regulatory / eligibility" description: "Age restrictions, prescriptions, insurance coverage, regional availability." check_method: "Rule-based eligibility check against user profile" signals: - source: "Real-time inventory / availability APIs" method: "OpenTable, Uber Eats, retail inventory APIs for stock and availability checks." availability: now - source: "Maps and traffic APIs" method: "Google Maps / Waze real-time traffic, ETA, fuel efficiency, alternate routes." availability: now - source: "Weather APIs" method: "Hourly and daily forecast data for activity feasibility." availability: now - source: "Calendar and schedule APIs" method: "Time-window availability based on existing commitments." availability: now - source: "Multi-constraint simultaneous optimization" method: "Solving budget x time x inventory x family constraints simultaneously across vendors." availability: future design_rules: fully_feasible: ui_density: medium interaction_complexity: standard recommendation: "Show plan as-is with confidence indicators. Green status. One-tap execution available." visual_indicator: "success" partially_feasible: ui_density: medium interaction_complexity: standard recommendation: "Show original plan with constraint violations highlighted. Offer pre-computed alternatives. Substitution modes apply (Exact / Flexible / Exploring / Surprise)." visual_indicator: "warning" substitution_modes: exact: "Find the same item elsewhere (different store, different delivery window)." flexible: "Find a close equivalent that meets most criteria." exploring: "Suggest alternatives the user might not have considered." surprise: "AI picks the best available option autonomously (requires high Autonomy Dial)." infeasible: ui_density: low interaction_complexity: minimal recommendation: "Clear explanation of why the plan cannot proceed. Offer the closest viable alternative. Option to defer to a feasible time window." visual_indicator: "error" required_actions: - "Show specific constraint that failed" - "Offer at least one alternative path" - "Allow user to adjust constraints (budget, timeline, location)" reality_level: 3 reality_note: "Individual feasibility checks (inventory, maps, weather) are mature and shipping. Multi-constraint simultaneous optimization across vendors and cross-vendor rate limiting at scale remain unsolved." # --------------------------------------------------------------- # RELATIONSHIP DIALS (7-8) # --------------------------------------------------------------- - name: Autonomy Dial name_ja: "Autonomy Dial (自律度ダイアル)" id: autonomy_dial number: 7 category: relationship_dial description: "Declares how much autonomous authority the AI currently has — from suggestion-only to fully automatic execution. Shaped by three simultaneous forces, NOT purely user-controlled." values: - id: suggest label: "Suggest" description: "AI shows candidates. Human decides and executes. AI has no execution authority." ai_authority: none - id: confirm label: "Confirm" description: "AI proposes a specific action. Human approves to execute. AI cannot proceed without approval." ai_authority: propose - id: notify label: "Notify" description: "AI executes autonomously and notifies the user after the fact. User can undo within a window." ai_authority: execute_and_report undo_window_seconds: 300 - id: auto label: "Auto" description: "AI executes autonomously without notification. User trusts the system completely for this domain." ai_authority: full three_forces: description: "The Autonomy Dial's position is determined by three forces acting simultaneously, not by user preference alone." forces: - id: service_default label: "Service default" description: "The starting position set by the service provider. Netflix and Amazon recommendations already operate at Suggest or higher by default." examples: - "Netflix autoplay = Notify level by default" - "Amazon 'frequently bought together' = Suggest level by default" - "Gmail spam filter = Auto level by default" - id: ai_adjustment label: "AI adjustment" description: "AI observes patterns and proposes dial changes. 5 consecutive approvals may trigger 'should I do this automatically next time?' But pushing too far triggers backlash and dial regression." learning_signal: "consecutive_approvals >= 5 → propose_upgrade" backlash_signal: "user_override_or_undo → propose_downgrade" - id: user_active_change label: "User active change" description: "User explicitly adjusts the dial. 'Just do it from now on' (upgrade) or 'Actually, ask me first' (downgrade). Always takes precedence." precedence: highest scope: per_domain: true per_person: true temporal_evolution: true examples: - domain: "Grocery staples reorder" typical_dial: notify rationale: "Routine, low-risk, high-frequency. Dial tends to rise quickly." - domain: "Dinner menu selection" typical_dial: suggest rationale: "Personal preference, variety desired. Dial stays low." - domain: "Child's content filtering" typical_dial: auto rationale: "Safety-critical. Parents set to Auto and rarely revisit." - domain: "Travel hotel booking" typical_dial: confirm rationale: "High cost, personal preference. AI narrows options, human decides." prerequisite: > Autonomy Dial level is bounded by Disclosure Dial level. AI cannot automate what it does not know about. If Disclosure Dial = minimal for a domain, Autonomy Dial cannot exceed Suggest for that domain. design_rules: suggest: ui_density: high interaction_complexity: full recommendation: "Show ranked options with explanations. User browses, compares, and selects. No auto-execution. All decisions require explicit user action." ui_pattern: "Card carousel or ranked list with comparison features." confirm: ui_density: medium interaction_complexity: standard recommendation: "Show AI's top recommendation prominently with 'Approve' as primary action. Alternative options available but deprioritized. Single-tap approval flow." ui_pattern: "Hero recommendation card with approve/modify/reject actions." notify: ui_density: low interaction_complexity: minimal recommendation: "Show post-execution notification with undo option. 'Ordered [item]. Undo within 5 min.' No pre-approval UI needed. Undo must be maximally accessible." ui_pattern: "Toast notification with undo action. Activity log for review." undo_affordance: required auto: ui_density: low interaction_complexity: minimal recommendation: "No notification for individual actions. Periodic summary available in activity log. Trust indicator visible in settings. User can review and downgrade at any time." ui_pattern: "Background execution. Weekly summary digest. Settings-accessible audit trail." multi_person_orchestration: description: "Different family members can have different Autonomy Dial levels for the same domain simultaneously." example: domain: "Grocery shopping" members: - person: "Father" dial: notify ui: "Receives post-purchase notification with undo" - person: "Mother" dial: confirm ui: "Receives approval request before purchase" - person: "10-year-old son" dial: suggest ui: "Can browse and add to wishlist, cannot trigger purchase" reality_level: 3 reality_note: "Developer tools (Claude Code, GitHub Copilot, Cursor) ship 2-3 stage autonomy modes. Smart home automations allow per-device rules. Consumer-facing per-domain, per-person, temporally-evolving Autonomy Dials with multi-person orchestration do not exist in any shipping product." - name: Disclosure Dial name_ja: "Disclosure Dial (開示ダイアル)" id: disclosure_dial number: 8 category: relationship_dial description: "Declares how much personal information the AI has access to, per domain and per person. Prerequisite for Autonomy — you cannot delegate what the AI does not know about." values: - id: none label: "None" description: "AI has no information about this domain. Cannot personalize or automate." - id: minimal label: "Minimal" description: "Basic preferences only. Enough for simple filtering but not proactive suggestions." - id: moderate label: "Moderate" description: "Preferences, history, and patterns shared. AI can make informed suggestions." - id: full label: "Full" description: "Complete information access for this domain. AI can anticipate needs and act proactively." logical_constraint: > Disclosure Dial is the prerequisite for Autonomy Dial. "Not disclosed + delegated" is logically impossible. The AI cannot automate decisions in domains it knows nothing about. Autonomy Dial level for any domain MUST NOT exceed the practical ceiling set by the Disclosure Dial level for that same domain. constraint_matrix: - disclosure: none max_autonomy: null description: "AI has no involvement. Cannot even suggest." - disclosure: minimal max_autonomy: suggest description: "AI can suggest based on limited info. Cannot confirm or auto-execute reliably." - disclosure: moderate max_autonomy: notify description: "AI has enough context to act and report. May occasionally miss preferences." - disclosure: full max_autonomy: auto description: "AI has complete context. Full automation is possible and reliable." three_forces: description: "Like Autonomy Dial, Disclosure Dial is shaped by three forces." forces: - id: service_default label: "Service default" description: "Services already access significant data by default. Google has location, search, and email. Samsung Family Hub stores family profiles." examples: - "Google: location history, search history, email content = moderate-to-full by default" - "Samsung Family Hub: up to 6 family profiles, allergies, dietary restrictions, Voice ID" - id: ai_proposal label: "AI proposal" description: "AI identifies value from additional disclosure. 'Share your music taste and I can suggest concerts on your trip. Share?'" trigger: "Identified value-add from undisclosed domain" - id: user_active_change label: "User active change" description: "User explicitly opens or closes disclosure. 'Share my music taste' or 'Stop accessing my health data.' Always takes precedence." precedence: highest scope: per_domain: true per_person: true examples: - person: "Father" domains: music: full health: none food: moderate finance: minimal - person: "Mother" domains: music: none health: minimal food: full education: full - person: "10-year-old son" domains: anime: full gaming: moderate education: moderate managed_by: "Parents" signals: - source: "App permission grants" method: "iOS/Android per-app permissions (camera, location, contacts, health). Granular opt-in." availability: now - source: "Connected app settings" method: "Google Gemini Connected Apps — per-service opt-in (Gmail, Photos, YouTube, etc)." availability: now - source: "Device profile settings" method: "Samsung Family Hub per-member profiles, dietary restrictions, Voice ID enrollment." availability: now - source: "Unified cross-domain disclosure management" method: "Single interface to manage disclosure levels across all domains (music, food, health, travel, finance). Does not exist today." availability: future - source: "Family-member disclosure governance" method: "Parents managing children's disclosure scope. Partial in parental controls but not as unified dial." availability: partial design_rules: none: ui_density: low interaction_complexity: minimal recommendation: "Generic, non-personalized UI. Offer clear value proposition for enabling disclosure. Show what the user would gain. Never pressure." personalization_level: none ai_capability: "No AI involvement in this domain." minimal: ui_density: medium interaction_complexity: standard recommendation: "Basic preference-based filtering. AI can exclude known dislikes. Cannot proactively suggest. Suggest expanding disclosure when clear value exists." personalization_level: basic_filtering ai_capability: "Filter and basic suggest only." moderate: ui_density: medium interaction_complexity: standard recommendation: "Personalized recommendations based on history and patterns. AI can proactively suggest. Show how disclosed data improves suggestions to build trust." personalization_level: personalized ai_capability: "Suggest, confirm, and notify." full: ui_density: high interaction_complexity: full recommendation: "Deeply personalized, anticipatory UI. AI can predict needs before user articulates them. Implicit Intent detection enabled for this domain. Show transparency — what data is being used and why." personalization_level: anticipatory ai_capability: "Full spectrum including auto, implicit intent detection." reality_level: 2 reality_note: "Per-app permissions and connected app settings ship today but are fragmented across platforms. A unified per-domain, per-person Disclosure Dial with clear Autonomy coupling does not exist. Closest analogs: Google Connected Apps, iOS app permissions, Samsung Family Hub profiles." # ------------------------------------------------------------------- # Cross-Token Interactions # ------------------------------------------------------------------- cross_token_rules: - name: "Driving safety override" condition: physical_state: driving effect: form_factor: car_display cognitive_load: "cap at low density regardless of estimated load" interaction_complexity: minimal hard_rule: "Voice-only interaction. No visual attention demands." - name: "Cognitive overload protection" condition: cognitive_load: overloaded effect: max_choices: 1 defer_non_critical: true notification_filter: "critical_only" priority_weight: "auto-elevate critical, suppress standard and low" - name: "Public privacy cascade" condition: social_exposure: public effect: disclosure_dial: "UI behaves as if disclosure = minimal regardless of actual setting" content_filter: maximum audio_output: "suppress unless headphones detected" - name: "Children present filter" condition: social_exposure: family_with_children effect: content_filter: moderate price_visibility: hidden_for_gifts autonomy_dial: "child members capped at suggest" - name: "Disclosure-Autonomy gate" condition: disclosure_dial: none effect: autonomy_dial: "forced to null — AI has no involvement" condition_2: disclosure_dial: minimal effect_2: autonomy_dial: "capped at suggest" - name: "Multi-surface content distribution" condition: form_factor: "multiple surfaces detected" effect: content_distribution: "allocate content to optimal surface by type" phone_role: "controller / detail view" tv_role: "shared media / large visual" fridge_role: "kitchen-context cards" watch_role: "notifications / quick actions only" - name: "Feasibility constraint override" condition: feasibility: infeasible effect: priority_weight: "deprioritize infeasible items regardless of urgency" substitution_ui: "activate based on Autonomy Dial level" user_notification: "explain constraint and offer alternatives" # ------------------------------------------------------------------- # Substitution Modes (used by Feasibility + Autonomy interaction) # ------------------------------------------------------------------- substitution_modes: - id: exact label: "Exact" description: "Find the same item from a different source (different store, delivery window)." autonomy_requirement: suggest user_control: full - id: flexible label: "Flexible" description: "Find a close equivalent meeting most criteria." autonomy_requirement: confirm user_control: "approves substitution" - id: exploring label: "Exploring" description: "AI suggests alternatives the user might not have considered." autonomy_requirement: confirm user_control: "approves from expanded options" - id: surprise label: "Surprise" description: "AI autonomously picks the best available option." autonomy_requirement: notify user_control: "undo after execution"