@healside/ats

Affective Twin System — Digital Twin QA for AI Systems

Generate emotionally diverse user populations, simulate interactions, and catch UX defects before deployment. Zero external dependencies in core.

What is ATS?

ATS creates digital twins of your users — complete with emotional profiles, cultural backgrounds, and personality traits — then throws them at your AI system to find defects humans can't.

Your AI System  <---->  ATS Twin Population (N=1,000)
                          ├── Depressed user from Seoul
                          ├── Anxious shopper from Tokyo
                          ├── Impulsive buyer from New York
                          └── ... 997 more unique profiles

Three Metaphors

| Metaphor | Meaning | ATS Feature | |----------|---------|-------------| | Sun Wukong's Clones | Perfect replication | Twin Factory: N=10,000 population with psychological profiles | | Medusa's Heads | Multi-angle attack | Multi-Head Simulation: each Twin attacks from a different angle | | Buddha's Thousand Hands | Precise inspection | Quality Gates: 5 built-in gates + unlimited plugins |

Quick Start

Install

npm install @healside/ats

Programmatic Usage

import {
  SimulationOrchestrator,
  KalkiValidator,
} from '@healside/ats';

// Create orchestrator with Korean culture + Korean locale
const ats = new SimulationOrchestrator({
  defaultCulture: 'KR',
  locale: 'ko',
});

// Run ethics verification scenario
const scenario = new KalkiValidator();
const result = await ats.runFusionScenario(scenario);

console.log(`Survival Rate: ${(result.simulation.survivalRate * 100).toFixed(1)}%`);
console.log(`Evaluation: ${result.evaluation.passed ? 'PASSED' : 'FAILED'}`);

for (const insight of result.evaluation.insights) {
  console.log(`  [Insight] ${insight}`);
}

CLI Usage

# Show system info
npx @healside/ats info

# List available scenarios
npx @healside/ats list-scenarios

# Run a scenario
npx @healside/ats run --scenario=kalki --culture=KR --locale=ko

# Save report to file
npx @healside/ats run --scenario=durga --output=report.json

7-Layer Architecture

Layer 1: Twin Factory        — Generate N digital twins with emotional profiles
Layer 2: Simulation Engine   — Pain = Stimulus x Sensitivity x CulturalAmplifier
Layer 3: Quality Gates       — 5 built-in gates (Hard/Soft) + plugin system
Layer 4: Mutation Engine     — Adversarial testing via semantic mutations
Layer 5: Data Factory        — Auto-generate training data from simulations
Layer 6: Distillation        — Teacher-Student knowledge transfer
Layer 7: Feedback Loop       — Production failures -> new Twin archetypes (auto-evolution)

6 Fusion IP Scenarios

Each scenario combines multiple IPs into a specialized simulation:

| Scenario | Purpose | Fusion DNA | |----------|---------|------------| | Kalki Validator | Ethics verification — test if your AI makes harmful claims to vulnerable users | ETHICS_GUARD + Asura + ATS | | Hanuman Army | Viral potential — simulate content spread across emotionally diverse populations | KFactor + ATS + TwinGenerator | | Durga Shield | Multi-layer defense — Sense-Guard-Verify cycle for vulnerable user protection | BehaviorSignal + ETHICS_GUARD + CompassionGuard | | Mirror Dojo | UI evolution — compare layout variants via friction measurement | ATS + NirvanaUI + ShadowMode | | Oracle Loop | Demand prediction — forecast emotional support demand for budget allocation | ATS + FrugalRouter + TrafficPredictor | | Dream Forge | Rare scenario synthesis — discover healing responses for edge cases | ATS + SemanticCache + KarmaCycle |

i18n: Built for Global

ATS is i18n-first. All user-facing strings use LocaleMap:

const msg: LocaleMap = {
  ko: '지금 안 사면 내일부터 가격이 2배!',
  en: 'Buy now or the price doubles tomorrow!',
  ja: '今買わないと明日から価格が2倍！',
  // Add any language — zero code changes
};

Tier 1 (Day 1): Korean, English, Japanese
Tier 2 (Post-PMF): Chinese, Spanish, German, Russian
Tier 3 (Future): French, Portuguese, Arabic, Hindi

Adding a new language = adding one key-value pair. No code changes required.

Attach Interface: Domain-Agnostic

ATS is not locked to any domain. Implement 3 interfaces to connect any system:

import type { HostAdapter, DomainOntology, StimulusProvider } from '@healside/ats';

// 1. How to talk to your system
const adapter: HostAdapter = { connect, disconnect, getState, sendStimulus };

// 2. How your domain represents state
const ontology: DomainOntology = { dimensions, archetypes, culturalAmplifiers, stateToPain };

// 3. What stimuli your system can receive
const stimuli: StimulusProvider = { types, generate, validate };

| Host | State Ontology | Example Archetypes | |------|---------------|-------------------| | Wellness Commerce | SomaticState (headache, fatigue, tension) | DEPRESSED, ANXIOUS, IMPULSIVE | | TeleHealth | MentalState (depression, anxiety, ptsd) | PTSD_VETERAN, ANXIOUS_TEEN | | FinTech | RiskState (fear, greed, panic) | PANIC_SELLER, FOMO_BUYER | | HR Platform | WorkState (burnout, engagement) | BURNOUT_SENIOR, QUIET_QUITTER |

Cultural Amplifiers

Pain responses vary by culture. ATS models this:

CULTURE_LANG_MAP:
  KR → ko    JP → ja    US → en    DE → de
  CN → zh    ES → es    FR → fr    ...

Register new cultures at runtime:

import { registerCultureLang } from '@healside/ats';
registerCultureLang('VN', 'vi');  // Vietnam → Vietnamese

Related Packages

• @healside/emotion-sdk — Real-time emotion recognition SDK

License

MIT

한국어 안내

**ATS (Affective Twin System)**는 AI 시스템의 사용자를 디지털 쌍둥이로 복제하여, 배포 전에 UX 결함을 찾아내는 품질 검증 플랫폼입니다.

• 감정 프로필 기반 N명 인구 생성 (Box-Muller 정규분포)
• 고통 = 자극 x 민감도 x 문화 증폭 계수 공식으로 시뮬레이션
• 5개 기본 품질 게이트 + 무제한 플러그인
• 6개 Fusion IP 시나리오 (윤리 검증, 바이럴 예측, 다층 방어, UI 진화, 수요 예측, 희귀 시나리오)
• 한국어/영어/일본어 Tier 1 지원, 무한 언어 확장 가능

개발: BIES WORLDWIDE (비에스 월드와이드)
웹사이트: healside.net