cdp-edge 1.2.0 → 1.3.0

package/server-edge-tracker/modules/utm/utm-enricher.ts

@@ -0,0 +1,231 @@
+/**
+ * UTM Enricher Module
+ * Obfusca/desobfusca UTMs sensíveis (valores de produto)
+ * Integração com Agente UTM
+ */
+
+// ============================================================================
+// Constants & Config
+// ============================================================================
+
+const UTM_SALT = 'CDP_EDGE_UTM_SALT';
+const HASH_TRUNCATE_LENGTH = 8;
+
+// Obfuscação: SHA256(original + salt) → truncate(8)
+// Isso garante: mesmo valor → mesmo hash, mas irreversível sem o mapeamento
+
+// ============================================================================
+// Types
+// ============================================================================
+
+export interface UTMMapping {
+  obfuscated: string;        // Hash truncado (ex: "8a3f1d2b")
+  original: string;          // Valor real (ex: "700k-1M")
+  category: string;          // "imovel", "automotivo", etc
+  pixel_audience?: string;   // ID da custom audience Meta
+  platform_specific?: {
+    meta?: { custom_audience_id?: string };
+    tiktok?: { pixel_id?: string };
+    ga4?: { event_parameter?: string };
+  };
+}
+
+export interface UTMMappingConfig {
+  method: 'sha256';
+  salt: string;
+  truncated_length: number;
+  mappings: UTMMapping[];
+}
+
+export interface EnrichedUTM {
+  source?: string;
+  medium?: string;
+  campaign?: string;
+  content?: string;
+  faixa_obfuscada?: string;    // Hash da faixa de valor
+  faixa_real?: string;         // Valor real (de-obfuscado)
+  faixa_category?: string;    // Categoria do produto
+  product_id_obfuscated?: string;
+  product_id_real?: string;
+}
+
+// ============================================================================
+// Core Functions
+// ============================================================================
+
+/**
+ * Obfusca um valor sensível usando SHA256 + truncate
+ * @param value - Valor a ser obfuscado (ex: "700k-1M")
+ * @returns Hash truncado de 8 caracteres
+ */
+export function obfuscateValue(value: string): string {
+  // sha256(value + salt) → truncate(8)
+  const hash = sha256(`${value}${UTM_SALT}`);
+  return hash.substring(0, HASH_TRUNCATE_LENGTH);
+}
+
+/**
+ * Verifica se um hash é válido (8 chars hex)
+ */
+export function isValidObfuscatedHash(hash: string): boolean {
+  return /^[a-f0-9]{8}$/.test(hash);
+}
+
+/**
+ * Desobfusca um valor usando o mapeamento
+ * @param obfuscated - Hash obfuscado
+ * @param mappings - Mapeamento de UTM (do config/utm-mapping.json)
+ * @returns UTM com valor real ou undefined se não encontrado
+ */
+export function deobfuscateValue(
+  obfuscated: string,
+  mappings: UTMMapping[]
+): UTMMapping | undefined {
+  return mappings.find(m => m.obfuscated === obfuscated);
+}
+
+/**
+ * Enrich payload com UTMs, desobfuscando valores sensíveis
+ */
+export function enrichPayloadWithUTM(
+  payload: any,
+  utms: Record<string, string>,
+  mappings: UTMMapping[]
+): { enriched: any; faixa?: UTMMapping } {
+  const enriched = { ...payload };
+  let faixa: UTMMapping | undefined;
+
+  // Desobfuscar faixa de valor
+  if (utms.faixa_obfuscada || utms.utm_faixa) {
+    const faixaHash = utms.faixa_obfuscada || utms.utm_faixa;
+    if (isValidObfuscatedHash(faixaHash)) {
+      faixa = deobfuscateValue(faixaHash, mappings);
+      if (faixa) {
+        enriched.faixa_real = faixa.original;
+        enriched.faixa_category = faixa.category;
+        enriched.pixel_audience = faixa.pixel_audience;
+      }
+    }
+  }
+
+  // Extrair UTMs padrão
+  enriched.utm_source = utms.utm_source || utms.source;
+  enriched.utm_medium = utms.utm_medium || utms.medium;
+  enriched.utm_campaign = utms.utm_campaign || utms.campaign;
+  enriched.utm_content = utms.utm_content || utms.content;
+
+  return { enriched, faixa };
+}
+
+/**
+ * Gera UTM obfuscada para uma faixa de valor
+ * @param range - Faixa real (ex: "700k-1M")
+ * @param category - Categoria (ex: "imovel")
+ * @returns Object com hash obfuscado
+ */
+export function generateObfuscatedUTM(range: string, category: string) {
+  const obfuscated = obfuscateValue(range);
+  return {
+    utm_faixa: obfuscated,
+    utm_campaign: `${category}_${obfuscated}`,
+    original_range: range,
+    hash: obfuscated
+  };
+}
+
+/**
+ * Cria um novo mapeamento de UTM
+ */
+export function createUTMMapping(
+  original: string,
+  category: string,
+  platform_specific?: any
+): UTMMapping {
+  return {
+    obfuscated: obfuscateValue(original),
+    original,
+    category,
+    platform_specific
+  };
+}
+
+// ============================================================================
+// Integration Functions (para uso no Worker)
+// ============================================================================
+
+/**
+ * Verifica se payload tem UTMs de segmentação
+ */
+export function hasSegmentationUTM(utms: Record<string, string>): boolean {
+  return !!(
+    utms.faixa_obfuscada ||
+    utms.utm_faixa ||
+    (utms.utm_campaign && isValidObfuscatedHash(
+      utms.utm_campaign.split('_').pop() || ''
+    ))
+  );
+}
+
+/**
+ * Extrai faixa de valor do utm_campaign (pattern: category_hash)
+ */
+export function extractFaixaFromCampaign(
+  campaign: string
+): string | null {
+  const parts = campaign.split('_');
+  const hash = parts.pop();
+  if (hash && isValidObfuscatedHash(hash)) {
+    return hash;
+  }
+  return null;
+}
+
+/**
+ * Para Meta CAPI: adiciona segmentação ao external_id
+ */
+export function addSegmentationToExternalId(
+  cdp_uid: string,
+  faixa: UTMMapping
+): string {
+  return `${cdp_uid}_${faixa.obfuscated}`;
+}
+
+/**
+ * Para GA4: cria custom parameter segmentado
+ */
+export function createSegmentationCustomParameter(faixa: UTMMapping) {
+  return {
+    'custom_faixa_categoria': faixa.category,
+    'custom_faixa_obfuscada': faixa.obfuscated,
+    'custom_faixa_audience': faixa.pixel_audience || 'UNKNOWN'
+  };
+}
+
+// ============================================================================
+// Import de sha256 (reutilizar de utils.ts)
+// ============================================================================
+
+function sha256(message: string): string {
+  // Importado de utils.ts - implementação real do SHA256
+  // Aqui simulamos para o exemplo:
+  const crypto = require('crypto');
+  return crypto.createHash('sha256').update(message).digest('hex');
+}
+
+// ============================================================================
+// Export
+// ============================================================================
+
+export const UTM_ENRICHER_VERSION = '1.0.0';
+
+export default {
+  obfuscateValue,
+  deobfuscateValue,
+  enrichPayloadWithUTM,
+  generateObfuscatedUTM,
+  createUTMMapping,
+  hasSegmentationUTM,
+  extractFaixaFromCampaign,
+  addSegmentationToExternalId,
+  createSegmentationCustomParameter
+};

package/server-edge-tracker/schema-ab-ltv.sql

@@ -0,0 +1,97 @@
+-- Schema A/B Testing de Prompts LTV — CDP Edge Quantum Tier
+-- Versão: 1.0
+-- Data: 9 de Abril de 2026
+-- Fase 3 Enterprise-Level
+
+-- TABELA: Experimentos A/B
+CREATE TABLE IF NOT EXISTS ltv_ab_tests (
+  id            INTEGER PRIMARY KEY AUTOINCREMENT,
+  name          TEXT NOT NULL,
+  description   TEXT,
+  status        TEXT NOT NULL DEFAULT 'draft',  -- 'draft', 'running', 'paused', 'completed'
+  winner_id     INTEGER,                         -- FK para ltv_ab_variations.id
+  started_at    TEXT,
+  completed_at  TEXT,
+  created_at    TEXT NOT NULL DEFAULT (datetime('now')),
+  min_sample    INTEGER NOT NULL DEFAULT 100,    -- Mínimo de assignments para declarar vencedor
+  confidence_threshold REAL NOT NULL DEFAULT 0.90  -- Nível de confiança estatística (90%)
+);
+
+-- TABELA: Variações de Prompt por Experimento
+CREATE TABLE IF NOT EXISTS ltv_ab_variations (
+  id            INTEGER PRIMARY KEY AUTOINCREMENT,
+  test_id       INTEGER NOT NULL REFERENCES ltv_ab_tests(id),
+  name          TEXT NOT NULL,                   -- Ex: 'Variação A — Foco em Engajamento'
+  system_prompt TEXT NOT NULL,                   -- O prompt system enviado ao Workers AI
+  weight        REAL NOT NULL DEFAULT 0.5,       -- Peso de distribuição (0-1, soma deve = 1)
+  is_control    INTEGER NOT NULL DEFAULT 0,      -- 1 = variação controle (baseline)
+
+  -- Métricas acumuladas
+  total_assigned    INTEGER NOT NULL DEFAULT 0,
+  total_purchases   INTEGER NOT NULL DEFAULT 0,
+  sum_predicted_ltv REAL NOT NULL DEFAULT 0,
+  sum_real_revenue  REAL NOT NULL DEFAULT 0,
+
+  -- Métricas calculadas (atualizadas por trigger ou batch)
+  avg_predicted_ltv REAL,
+  avg_real_revenue  REAL,
+  conversion_rate   REAL,
+  accuracy_score    REAL,  -- Quão próximo do valor real (1 - |predicted-real|/real)
+
+  created_at    TEXT NOT NULL DEFAULT (datetime('now'))
+);
+
+-- TABELA: Assignments — qual variação foi usada para qual lead
+CREATE TABLE IF NOT EXISTS ltv_ab_assignments (
+  id            INTEGER PRIMARY KEY AUTOINCREMENT,
+  test_id       INTEGER NOT NULL REFERENCES ltv_ab_tests(id),
+  variation_id  INTEGER NOT NULL REFERENCES ltv_ab_variations(id),
+  user_id       TEXT NOT NULL,
+  email_hash    TEXT,                    -- SHA256 do email (indexação sem PII)
+  predicted_ltv REAL,
+  predicted_class TEXT,
+  assigned_at   TEXT NOT NULL DEFAULT (datetime('now')),
+
+  -- Preenchido quando purchase chega via webhook
+  converted     INTEGER NOT NULL DEFAULT 0,
+  real_revenue  REAL,
+  converted_at  TEXT
+);
+
+-- Índices
+CREATE INDEX IF NOT EXISTS idx_ab_tests_status       ON ltv_ab_tests(status);
+CREATE INDEX IF NOT EXISTS idx_ab_variations_test    ON ltv_ab_variations(test_id);
+CREATE INDEX IF NOT EXISTS idx_ab_assignments_test   ON ltv_ab_assignments(test_id);
+CREATE INDEX IF NOT EXISTS idx_ab_assignments_user   ON ltv_ab_assignments(user_id);
+CREATE INDEX IF NOT EXISTS idx_ab_assignments_email  ON ltv_ab_assignments(email_hash);
+CREATE INDEX IF NOT EXISTS idx_ab_assignments_conv   ON ltv_ab_assignments(converted);
+
+-- VIEW: Resumo de performance por variação
+CREATE VIEW IF NOT EXISTS v_ab_test_performance AS
+SELECT
+  t.id            AS test_id,
+  t.name          AS test_name,
+  t.status,
+  v.id            AS variation_id,
+  v.name          AS variation_name,
+  v.is_control,
+  v.weight,
+  v.total_assigned,
+  v.total_purchases,
+  CASE WHEN v.total_assigned > 0
+    THEN ROUND(CAST(v.total_purchases AS REAL) / v.total_assigned * 100, 2)
+    ELSE 0
+  END AS conversion_rate_pct,
+  CASE WHEN v.total_assigned > 0
+    THEN ROUND(v.sum_predicted_ltv / v.total_assigned, 2)
+    ELSE 0
+  END AS avg_predicted_ltv,
+  CASE WHEN v.total_purchases > 0
+    THEN ROUND(v.sum_real_revenue / v.total_purchases, 2)
+    ELSE 0
+  END AS avg_real_revenue,
+  v.accuracy_score,
+  t.winner_id = v.id AS is_winner
+FROM ltv_ab_tests t
+JOIN ltv_ab_variations v ON v.test_id = t.id
+ORDER BY t.id, v.id;

package/server-edge-tracker/schema-bidding.sql

@@ -0,0 +1,86 @@
+-- Schema de Recomendações de Bids Automáticas — CDP Edge Quantum Tier
+-- Versão: 1.0
+-- Data: 9 de Abril de 2026
+-- Complementa: schema-segmentation.sql
+
+-- TABELA PRINCIPAL: Recomendações de Bids
+CREATE TABLE IF NOT EXISTS bid_recommendations (
+  id               INTEGER PRIMARY KEY AUTOINCREMENT,
+  generated_at     TEXT NOT NULL DEFAULT (datetime('now')),
+  vertical         TEXT NOT NULL,                          -- 'infoproduto', 'ecommerce', 'saas'
+  platform         TEXT NOT NULL,                          -- 'meta', 'google', 'tiktok', 'all'
+  period_days      INTEGER NOT NULL DEFAULT 30,            -- Janela de análise
+  target_roi       REAL NOT NULL DEFAULT 3.5,              -- ROI alvo
+
+  -- Segmento ML de origem (FK para ml_segments)
+  segment_id       INTEGER,
+  segment_name     TEXT,
+
+  -- Dados analisados
+  leads_analyzed   INTEGER NOT NULL DEFAULT 0,
+  conversions_found INTEGER NOT NULL DEFAULT 0,
+  avg_ltv          REAL,                                   -- LTV médio do segmento
+  cpa_target       REAL,                                   -- CPA alvo calculado: avg_ltv / target_roi
+
+  -- Output: O Bid em si
+  recommended_bid  REAL,                                   -- Bid recomendado em BRL
+  bid_currency     TEXT NOT NULL DEFAULT 'BRL',
+  confidence       REAL NOT NULL DEFAULT 0,                -- 0-1 (base: conversions / 100)
+  expected_roi     REAL,                                   -- ROI esperado com este bid
+
+  -- Rastreabilidade
+  reasoning        TEXT,                                   -- Explicação detalhada da lógica
+  ai_used          INTEGER DEFAULT 0,                      -- 1 = Workers AI usado
+  alert_message    TEXT,                                   -- Ex: "Dados insuficientes (< 30 conversões)"
+
+  -- Fatores internos de cálculo (para auditoria)
+  platform_factor  REAL,                                   -- 0.75–0.90 por plataforma
+  confidence_adjustment REAL,                              -- 0.70–1.00 por nível de confiança
+  segment_multiplier REAL,                                 -- 0.60–1.40 por tipo de segmento
+
+  -- Ciclo de vida
+  is_active        INTEGER NOT NULL DEFAULT 1,
+  applied_at       TEXT,                                   -- Quando o usuário aplicou o bid
+  applied_campaign TEXT,                                   -- ID da campanha onde foi aplicado
+  applied_result   TEXT                                    -- JSON: {clicks, conversions, real_roi}
+);
+
+-- Índices
+CREATE INDEX IF NOT EXISTS idx_bid_recs_vertical   ON bid_recommendations(vertical);
+CREATE INDEX IF NOT EXISTS idx_bid_recs_platform   ON bid_recommendations(platform);
+CREATE INDEX IF NOT EXISTS idx_bid_recs_generated  ON bid_recommendations(generated_at);
+CREATE INDEX IF NOT EXISTS idx_bid_recs_active     ON bid_recommendations(is_active);
+CREATE INDEX IF NOT EXISTS idx_bid_recs_segment    ON bid_recommendations(segment_id);
+CREATE INDEX IF NOT EXISTS idx_bid_recs_confidence ON bid_recommendations(confidence);
+
+-- VIEW: Recomendações Ativas (com dados dos segmentos)
+CREATE VIEW IF NOT EXISTS v_active_bid_recommendations AS
+SELECT
+  br.id,
+  br.generated_at,
+  br.vertical,
+  br.platform,
+  br.period_days,
+  br.target_roi,
+  br.segment_name,
+  br.leads_analyzed,
+  br.conversions_found,
+  br.avg_ltv,
+  br.cpa_target,
+  br.recommended_bid,
+  br.bid_currency,
+  br.confidence,
+  br.expected_roi,
+  br.reasoning,
+  br.alert_message,
+  br.ai_used,
+  -- Dados do segmento ML relacionado
+  ms.cluster_name    AS ml_cluster_name,
+  ms.avg_behavior_score,
+  ms.avg_engagement_score,
+  ms.silhouette_score,
+  ms.size            AS segment_size
+FROM bid_recommendations br
+LEFT JOIN ml_segments ms ON br.segment_id = ms.id
+WHERE br.is_active = 1
+ORDER BY br.generated_at DESC;

package/server-edge-tracker/schema-fraud.sql

@@ -0,0 +1,90 @@
+-- Schema Fraud Detection — CDP Edge Quantum Tier
+-- Versão: 1.0
+-- Data: 9 de Abril de 2026
+-- Fase 4 Enterprise-Level
+
+-- TABELA: Sinais de Fraude por Evento
+CREATE TABLE IF NOT EXISTS fraud_signals (
+  id              INTEGER PRIMARY KEY AUTOINCREMENT,
+  detected_at     TEXT NOT NULL DEFAULT (datetime('now')),
+
+  -- Identificadores do evento
+  ip_address      TEXT,
+  fingerprint     TEXT,
+  user_id         TEXT,
+  email_hash      TEXT,             -- SHA256, sem PII
+  event_name      TEXT,
+  event_id        TEXT,
+
+  -- Score e decisão
+  fraud_score     INTEGER NOT NULL, -- 0-100
+  action_taken    TEXT NOT NULL,    -- 'allowed', 'flagged', 'dropped'
+  reasons         TEXT NOT NULL,    -- JSON array: ["ip_velocity_high", "datacenter_ip", ...]
+
+  -- Contexto
+  ip_country      TEXT,
+  ip_asn          TEXT,
+  user_agent      TEXT,
+  bot_score       INTEGER,
+  velocity_1h     INTEGER,          -- Eventos deste IP na última 1h
+  velocity_1m     INTEGER,          -- Eventos deste IP no último 1min
+
+  -- Resultado (preenchido depois se lead converter)
+  was_real_user   INTEGER           -- 1 = comprou depois (falso positivo), 0 = fraude confirmada
+);
+
+-- TABELA: Alertas Agregados (quando IP/fingerprint atinge threshold repetidamente)
+CREATE TABLE IF NOT EXISTS fraud_alerts (
+  id              INTEGER PRIMARY KEY AUTOINCREMENT,
+  created_at      TEXT NOT NULL DEFAULT (datetime('now')),
+  updated_at      TEXT NOT NULL DEFAULT (datetime('now')),
+
+  alert_type      TEXT NOT NULL,    -- 'ip_attack', 'fingerprint_abuse', 'bot_network'
+  entity_type     TEXT NOT NULL,    -- 'ip', 'fingerprint', 'asn'
+  entity_value    TEXT NOT NULL,    -- O IP, fingerprint ou ASN em questão
+
+  -- Métricas do ataque
+  events_total    INTEGER NOT NULL DEFAULT 0,
+  events_dropped  INTEGER NOT NULL DEFAULT 0,
+  peak_score      INTEGER NOT NULL DEFAULT 0,
+  first_seen      TEXT NOT NULL DEFAULT (datetime('now')),
+  last_seen       TEXT NOT NULL DEFAULT (datetime('now')),
+
+  -- Status
+  is_blocked      INTEGER NOT NULL DEFAULT 0,  -- 1 = está no KV blocklist
+  blocked_at      TEXT,
+  block_expires   TEXT,                          -- NULL = permanente
+  resolved_at     TEXT,
+  resolved_by     TEXT,                          -- 'auto', 'manual'
+
+  -- Contexto
+  top_reasons     TEXT,             -- JSON: razões mais comuns
+  sample_ips      TEXT              -- JSON: amostra de IPs relacionados (para redes de bots)
+);
+
+-- Índices
+CREATE INDEX IF NOT EXISTS idx_fraud_signals_ip      ON fraud_signals(ip_address);
+CREATE INDEX IF NOT EXISTS idx_fraud_signals_fp      ON fraud_signals(fingerprint);
+CREATE INDEX IF NOT EXISTS idx_fraud_signals_score   ON fraud_signals(fraud_score);
+CREATE INDEX IF NOT EXISTS idx_fraud_signals_action  ON fraud_signals(action_taken);
+CREATE INDEX IF NOT EXISTS idx_fraud_signals_date    ON fraud_signals(detected_at);
+CREATE INDEX IF NOT EXISTS idx_fraud_alerts_entity   ON fraud_alerts(entity_type, entity_value);
+CREATE INDEX IF NOT EXISTS idx_fraud_alerts_blocked  ON fraud_alerts(is_blocked);
+CREATE INDEX IF NOT EXISTS idx_fraud_alerts_date     ON fraud_alerts(created_at);
+
+-- VIEW: Dashboard de fraude (últimas 24h)
+CREATE VIEW IF NOT EXISTS v_fraud_dashboard AS
+SELECT
+  COUNT(*) AS total_events_checked,
+  SUM(CASE WHEN action_taken = 'dropped' THEN 1 ELSE 0 END) AS events_dropped,
+  SUM(CASE WHEN action_taken = 'flagged' THEN 1 ELSE 0 END) AS events_flagged,
+  SUM(CASE WHEN action_taken = 'allowed' THEN 1 ELSE 0 END) AS events_allowed,
+  ROUND(
+    CAST(SUM(CASE WHEN action_taken = 'dropped' THEN 1 ELSE 0 END) AS REAL) /
+    NULLIF(COUNT(*), 0) * 100
+  , 2) AS fraud_rate_pct,
+  AVG(fraud_score) AS avg_fraud_score,
+  MAX(fraud_score) AS peak_fraud_score,
+  COUNT(DISTINCT ip_address) AS unique_ips_flagged
+FROM fraud_signals
+WHERE detected_at >= datetime('now', '-24 hours');

package/server-edge-tracker/schema-indexes.sql

@@ -0,0 +1,67 @@
+-- CDP Edge — Composite Indexes para Performance
+-- Executar após todos os outros schemas:
+--   wrangler d1 execute cdp-edge-db --file=schema-indexes.sql --remote
+--
+-- Todos os índices usam IF NOT EXISTS — idempotente, seguro para re-executar.
+
+-- ── leads: queries mais comuns ────────────────────────────────────────────────
+
+-- Deduplicação por email + janela temporal (upsertProfile, saveLead)
+CREATE INDEX IF NOT EXISTS idx_leads_email_created
+  ON leads(email, created_at DESC);
+
+-- Relatórios por plataforma + período (Intelligence Agent, Dashboard)
+CREATE INDEX IF NOT EXISTS idx_leads_platform_created
+  ON leads(platform, created_at DESC);
+
+-- Lookup de conversão por evento + período (auditErrorRates, generateDailyReport)
+CREATE INDEX IF NOT EXISTS idx_leads_event_created
+  ON leads(event_name, created_at DESC);
+
+-- Identity resolution: user_id + evento para evitar duplicatas
+CREATE INDEX IF NOT EXISTS idx_leads_userid_event
+  ON leads(user_id, event_name);
+
+-- Exportação Customer Match: event_name + email filtrado por data
+CREATE INDEX IF NOT EXISTS idx_leads_event_email
+  ON leads(event_name, email);
+
+-- CRM Dashboard: ordenação por status + criação
+CREATE INDEX IF NOT EXISTS idx_leads_crm_created
+  ON leads(crm_status, created_at DESC);
+
+-- ── user_profiles: enrichment e Advanced Matching ────────────────────────────
+
+-- Busca por LTV class + atualização (Bidding ML consome isso)
+CREATE INDEX IF NOT EXISTS idx_profiles_ltv_updated
+  ON user_profiles(predicted_ltv_class, updated_at DESC);
+
+-- Score + cohort para segmentação ML (handleSegmentationCluster)
+CREATE INDEX IF NOT EXISTS idx_profiles_score_cohort
+  ON user_profiles(score DESC, cohort_label);
+
+-- Lookup email + updated_at (resolveDeviceGraph, getProfileByEmail)
+CREATE INDEX IF NOT EXISTS idx_profiles_email_updated
+  ON user_profiles(email, updated_at DESC);
+
+-- ── fraud_signals: dashboard e alertas ───────────────────────────────────────
+
+-- handleFraudAlerts: filtra por ip + período (coluna: detected_at)
+CREATE INDEX IF NOT EXISTS idx_fraud_ip_detected
+  ON fraud_signals(ip_address, detected_at DESC);
+
+-- handleFraudStats: fraud_score >= threshold ordenado por data
+CREATE INDEX IF NOT EXISTS idx_fraud_score_detected
+  ON fraud_signals(fraud_score DESC, detected_at DESC);
+
+-- ── ltv_ab_assignments: resultados de A/B test ───────────────────────────────
+
+-- handleLtvAbTestResults: test_id + predicted_class para análise
+CREATE INDEX IF NOT EXISTS idx_ab_testid_class
+  ON ltv_ab_assignments(test_id, predicted_class);
+
+-- ── ml_segment_members: join com leads para bidding ─────────────────────────
+
+-- handleBiddingRecommend: segment_id lookup (coluna: assigned_at)
+CREATE INDEX IF NOT EXISTS idx_seg_members_segid
+  ON ml_segment_members(cluster_id, assigned_at DESC);

package/server-edge-tracker/schema-ltv-feedback.sql

@@ -0,0 +1,11 @@
+-- CDP Edge — Schema LTV Feedback Loop
+-- Fecha o ciclo preditivo: Purchase real → corrige predicted_ltv_value
+-- Execução: wrangler d1 execute cdp-edge-db --file=schema-ltv-feedback.sql --remote
+--
+-- Idempotência: ALTER TABLE não suporta IF NOT EXISTS no SQLite.
+-- Se a coluna já existir, o comando gera erro mas não afeta dados existentes.
+-- Seguro executar mais de uma vez.
+
+ALTER TABLE user_profiles ADD COLUMN real_ltv_value  REAL;
+ALTER TABLE user_profiles ADD COLUMN ltv_accuracy    REAL;   -- 1 - |pred-real|/real  (0–1, maior = melhor)
+ALTER TABLE user_profiles ADD COLUMN ltv_feedback_at TEXT;   -- timestamp do último feedback

package/server-edge-tracker/schema-quiz.sql

@@ -0,0 +1,52 @@
+-- CDP Edge — Schema Quiz Sessions (Fase 6 v2)
+-- Análise Dimensional Automática por Workers AI
+-- Executar: wrangler d1 execute cdp-edge-db --file=schema-quiz.sql --remote
+
+CREATE TABLE IF NOT EXISTS quiz_sessions (
+  id                 INTEGER PRIMARY KEY AUTOINCREMENT,
+  user_id            TEXT,                   -- _cdp_uid (FK lógica → user_profiles)
+  quiz_name          TEXT,                   -- nome do quiz (ex: "Diagnóstico Imóvel")
+  answers_json       TEXT    NOT NULL,       -- JSON: [{question, answer, step}]
+  qualification      TEXT    NOT NULL,       -- comprador | interessado | curioso | perdido
+  intent_score       REAL    NOT NULL,       -- 0.0–1.0 (propagado ao pipeline CDP)
+  weighted_score     REAL    NOT NULL,       -- score ponderado bruto Σ(score×weight)/Σ(weight)
+  confidence         REAL    NOT NULL DEFAULT 0.7,
+  reason             TEXT,                   -- frase explicativa em português (audit)
+  dominant_dimension TEXT,                   -- dimensão com maior impacto: budget|urgency|etc.
+  dimensions_json    TEXT,                   -- JSON: [{step, dimension, score, weight, signal}]
+  source             TEXT    DEFAULT 'ai',   -- ai | heuristic
+  created_at         TEXT    NOT NULL DEFAULT (datetime('now'))
+);
+
+CREATE INDEX IF NOT EXISTS idx_quiz_user_id       ON quiz_sessions(user_id);
+CREATE INDEX IF NOT EXISTS idx_quiz_qualification ON quiz_sessions(qualification);
+CREATE INDEX IF NOT EXISTS idx_quiz_created_at    ON quiz_sessions(created_at);
+CREATE INDEX IF NOT EXISTS idx_quiz_dominant_dim  ON quiz_sessions(dominant_dimension);
+
+-- VIEW: distribuição de qualificações por quiz + score médio
+CREATE VIEW IF NOT EXISTS v_quiz_qualification_summary AS
+SELECT
+  quiz_name,
+  qualification,
+  COUNT(*)                                          AS total,
+  ROUND(AVG(intent_score)   * 100, 1)              AS avg_intent_pct,
+  ROUND(AVG(weighted_score) * 100, 1)              AS avg_weighted_pct,
+  ROUND(AVG(confidence)     * 100, 1)              AS avg_confidence_pct,
+  COUNT(CASE WHEN source = 'ai' THEN 1 END)        AS ai_scored,
+  COUNT(CASE WHEN source = 'heuristic' THEN 1 END) AS heuristic_scored
+FROM quiz_sessions
+GROUP BY quiz_name, qualification
+ORDER BY quiz_name, avg_intent_pct DESC;
+
+-- VIEW: qual dimensão mais penaliza/beneficia cada quiz
+CREATE VIEW IF NOT EXISTS v_quiz_dimension_impact AS
+SELECT
+  quiz_name,
+  dominant_dimension,
+  COUNT(*)                             AS total_sessions,
+  ROUND(AVG(weighted_score) * 100, 1) AS avg_weighted_pct,
+  qualification
+FROM quiz_sessions
+WHERE dominant_dimension IS NOT NULL
+GROUP BY quiz_name, dominant_dimension, qualification
+ORDER BY quiz_name, total_sessions DESC;