traqr-memory-mcp 0.1.0

package/setup.sql

@@ -0,0 +1,1037 @@
+-- ============================================================================
+-- TraqrDB Fresh-Install Setup
+-- ============================================================================
+-- Compiled from migrations 001-011 into a single idempotent fresh-install.
+-- No ALTER TABLE, no DROP, no backfill. Tables have ALL v2 columns from start.
+--
+-- Dependency order:
+--   1. Extensions
+--   2. Immutable wrapper functions (needed by GENERATED columns)
+--   3. Tables: memory_users, memory_domains, traqr_memories, traqr_memory_history,
+--              memory_relationships, memory_entities, memory_entity_links, schema_version
+--   4. Indexes
+--   5. Conditional RLS (auth.role() only exists on Supabase)
+--   6. RPC functions (v2 versions only)
+--   7. Schema version tracking + migration bootstrap
+--
+-- Safe to re-run: all operations use IF NOT EXISTS / CREATE OR REPLACE.
+-- ============================================================================
+
+
+-- ============================================================================
+-- 1. EXTENSIONS
+-- ============================================================================
+
+CREATE EXTENSION IF NOT EXISTS vector;
+CREATE EXTENSION IF NOT EXISTS pg_trgm;
+
+
+-- ============================================================================
+-- 2. IMMUTABLE WRAPPER FUNCTIONS
+-- ============================================================================
+-- to_tsvector() is STABLE, but GENERATED ALWAYS AS requires IMMUTABLE.
+-- Safe because english/simple dictionaries are built-in and never change.
+
+CREATE OR REPLACE FUNCTION traqr_tsvector_en(content text, summary text, tags text[])
+RETURNS tsvector
+LANGUAGE sql IMMUTABLE AS $$
+  SELECT to_tsvector('english',
+    COALESCE(content, '') || ' ' ||
+    COALESCE(summary, '') || ' ' ||
+    COALESCE(array_to_string(tags, ' '), '')
+  );
+$$;
+
+CREATE OR REPLACE FUNCTION traqr_tsvector_simple(content text, summary text, tags text[])
+RETURNS tsvector
+LANGUAGE sql IMMUTABLE AS $$
+  SELECT to_tsvector('simple',
+    COALESCE(content, '') || ' ' ||
+    COALESCE(summary, '') || ' ' ||
+    COALESCE(array_to_string(tags, ' '), '')
+  );
+$$;
+
+
+-- ============================================================================
+-- 3. TABLES
+-- ============================================================================
+
+-- 3a. memory_users — user identity for multi-tenant support
+CREATE TABLE IF NOT EXISTS memory_users (
+    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    api_key VARCHAR(64) UNIQUE NOT NULL,
+    email VARCHAR(255) UNIQUE,
+    created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+    updated_at TIMESTAMP WITH TIME ZONE DEFAULT NOW()
+);
+
+-- 3b. memory_domains — legacy project/domain isolation
+CREATE TABLE IF NOT EXISTS memory_domains (
+    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    user_id UUID REFERENCES memory_users(id) ON DELETE CASCADE,
+    name VARCHAR(100) NOT NULL,
+    description TEXT,
+    is_shareable BOOLEAN DEFAULT FALSE,
+    created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+    updated_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+    UNIQUE(user_id, name)
+);
+
+-- 3c. traqr_memories — core memories table (v1 + v2 columns merged)
+CREATE TABLE IF NOT EXISTS traqr_memories (
+    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+
+    -- Identity / scoping
+    user_id UUID,
+    project_id UUID,
+    domain_id UUID REFERENCES memory_domains(id) ON DELETE CASCADE,
+
+    -- Content
+    content TEXT NOT NULL,
+    summary VARCHAR(500),
+
+    -- Categorization
+    category VARCHAR(50),
+    tags VARCHAR(100)[] DEFAULT '{}',
+    context_tags VARCHAR(100)[] DEFAULT '{}',
+    domain VARCHAR(100),
+    topic VARCHAR(100),
+
+    -- Embedding
+    embedding vector(1536),
+    embedding_model VARCHAR(100) DEFAULT 'openai/text-embedding-3-small',
+    embedding_model_version VARCHAR(20) DEFAULT 'v1',
+    needs_reembedding BOOLEAN DEFAULT FALSE,
+
+    -- Provenance
+    source_type VARCHAR(50),
+    source_ref VARCHAR(255),
+    source_project VARCHAR(100) DEFAULT 'default',
+    source_tool VARCHAR(50),
+
+    -- Confidence & Decay
+    original_confidence FLOAT DEFAULT 1.0,
+    last_validated TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+
+    -- Version relationships
+    related_to UUID[] DEFAULT '{}',
+    is_contradiction BOOLEAN DEFAULT FALSE,
+
+    -- Archive
+    is_archived BOOLEAN DEFAULT FALSE,
+    archived_at TIMESTAMP WITH TIME ZONE,
+    archive_reason VARCHAR(50),
+
+    -- Durability / TTL
+    durability VARCHAR(20) DEFAULT 'permanent',
+    expires_at TIMESTAMP WITH TIME ZONE,
+
+    -- Cross-project / portability
+    is_universal BOOLEAN DEFAULT FALSE,
+    is_portable BOOLEAN DEFAULT TRUE,
+    agent_type VARCHAR(50),
+
+    -- Citation tracking
+    times_returned INTEGER DEFAULT 0,
+    times_cited INTEGER DEFAULT 0,
+    last_returned_at TIMESTAMP WITH TIME ZONE,
+    last_cited_at TIMESTAMP WITH TIME ZONE,
+
+    -- v2: Memory lifecycle (M5 Pipeline Design)
+    memory_type VARCHAR(20) DEFAULT 'pattern',
+    is_latest BOOLEAN DEFAULT TRUE,
+    is_forgotten BOOLEAN DEFAULT FALSE,
+    forgotten_at TIMESTAMP WITH TIME ZONE,
+    forget_after TIMESTAMP WITH TIME ZONE,
+
+    -- v2: Temporal model (M7)
+    valid_at TIMESTAMP WITH TIME ZONE,
+    invalid_at TIMESTAMP WITH TIME ZONE,
+
+    -- v2: Dual tsvector for BM25 (GENERATED ALWAYS, auto-computed)
+    search_vector_en tsvector
+      GENERATED ALWAYS AS (traqr_tsvector_en(content, summary::text, tags::text[])) STORED,
+    search_vector_simple tsvector
+      GENERATED ALWAYS AS (traqr_tsvector_simple(content, summary::text, tags::text[])) STORED,
+
+    -- Timestamps
+    created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+    updated_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+
+    -- Constraints
+    CONSTRAINT chk_memory_type CHECK (memory_type IN ('fact', 'preference', 'pattern'))
+);
+
+-- 3d. traqr_memory_history — evolution tracking
+CREATE TABLE IF NOT EXISTS traqr_memory_history (
+    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    memory_id UUID REFERENCES traqr_memories(id) ON DELETE CASCADE,
+    previous_content TEXT,
+    previous_embedding vector(1536),
+    previous_confidence FLOAT,
+    change_reason TEXT,
+    changed_at TIMESTAMP WITH TIME ZONE DEFAULT NOW()
+);
+
+-- 3e. memory_relationships — memory-to-memory edges for version chains
+CREATE TABLE IF NOT EXISTS memory_relationships (
+    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    source_memory_id UUID NOT NULL REFERENCES traqr_memories(id) ON DELETE CASCADE,
+    target_memory_id UUID NOT NULL REFERENCES traqr_memories(id) ON DELETE CASCADE,
+    edge_type VARCHAR(20) NOT NULL,
+    confidence FLOAT DEFAULT 1.0,
+    created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+    metadata JSONB DEFAULT '{}',
+    CONSTRAINT chk_edge_type CHECK (edge_type IN ('updates', 'extends', 'derives', 'related')),
+    CONSTRAINT uq_memory_rel UNIQUE (source_memory_id, target_memory_id, edge_type)
+);
+
+-- 3f. memory_entities — entities extracted from memories (user-scoped)
+CREATE TABLE IF NOT EXISTS memory_entities (
+    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    user_id UUID NOT NULL,
+    name VARCHAR(255) NOT NULL,
+    entity_type VARCHAR(50) NOT NULL,
+    embedding vector(1536),
+    mentions_count INTEGER DEFAULT 0,
+    first_seen_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+    last_seen_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+    metadata JSONB DEFAULT '{}',
+    is_archived BOOLEAN DEFAULT FALSE,
+    created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+    updated_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+    UNIQUE(user_id, name, entity_type)
+);
+
+-- 3g. memory_entity_links — junction table (many-to-many memories <-> entities)
+CREATE TABLE IF NOT EXISTS memory_entity_links (
+    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    memory_id UUID NOT NULL REFERENCES traqr_memories(id) ON DELETE CASCADE,
+    entity_id UUID NOT NULL REFERENCES memory_entities(id) ON DELETE CASCADE,
+    role VARCHAR(20) DEFAULT 'mention',
+    extracted_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+    UNIQUE(memory_id, entity_id, role)
+);
+
+-- 3h. schema_version — tracks installed schema version
+CREATE TABLE IF NOT EXISTS schema_version (
+    version INTEGER PRIMARY KEY,
+    description TEXT,
+    applied_at TIMESTAMP WITH TIME ZONE DEFAULT NOW()
+);
+
+
+-- ============================================================================
+-- 4. INDEXES
+-- ============================================================================
+
+-- memory_users
+CREATE INDEX IF NOT EXISTS memory_users_api_key_idx
+  ON memory_users(api_key);
+
+-- memory_domains
+CREATE INDEX IF NOT EXISTS memory_domains_user_idx
+  ON memory_domains(user_id);
+
+-- traqr_memories: partial HNSW (only active, non-forgotten memories)
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_active_embedding
+  ON traqr_memories USING hnsw (embedding vector_cosine_ops)
+  WHERE is_archived = FALSE AND is_forgotten = FALSE;
+
+-- traqr_memories: legacy indexes (renamed from memories_* prefix)
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_domain_id
+  ON traqr_memories(domain_id);
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_tags
+  ON traqr_memories USING gin(tags);
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_category
+  ON traqr_memories(category);
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_source_project
+  ON traqr_memories(source_project);
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_archived
+  ON traqr_memories(is_archived);
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_created
+  ON traqr_memories(created_at DESC);
+
+-- traqr_memories: cross-project / portability
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_universal
+  ON traqr_memories(is_universal)
+  WHERE is_universal = TRUE;
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_universal_category
+  ON traqr_memories(is_universal, category)
+  WHERE is_universal = TRUE;
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_agent_type
+  ON traqr_memories(agent_type)
+  WHERE agent_type IS NOT NULL;
+
+-- traqr_memories: durability / TTL
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_expires_at
+  ON traqr_memories(expires_at)
+  WHERE expires_at IS NOT NULL AND durability != 'permanent';
+
+-- traqr_memories: citation tracking
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_citation
+  ON traqr_memories(times_returned, times_cited)
+  WHERE is_archived = FALSE;
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_cited
+  ON traqr_memories(times_cited DESC)
+  WHERE times_cited > 0 AND is_archived = FALSE;
+
+-- traqr_memories: v2 lifecycle indexes
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_memory_type
+  ON traqr_memories(memory_type) WHERE is_archived = FALSE;
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_is_latest
+  ON traqr_memories(is_latest) WHERE is_latest = TRUE AND is_archived = FALSE;
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_forgotten
+  ON traqr_memories(is_forgotten) WHERE is_forgotten = TRUE;
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_forget_after
+  ON traqr_memories(forget_after)
+  WHERE forget_after IS NOT NULL AND is_forgotten = FALSE;
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_source_tool
+  ON traqr_memories(source_tool) WHERE source_tool IS NOT NULL;
+
+-- traqr_memories: v2 temporal indexes
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_valid_at
+  ON traqr_memories(valid_at) WHERE valid_at IS NOT NULL;
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_temporal
+  ON traqr_memories(valid_at, invalid_at)
+  WHERE is_archived = FALSE AND is_forgotten = FALSE;
+
+-- traqr_memories: v2 BM25 tsvector indexes (GIN for full-text search)
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_search_en
+  ON traqr_memories USING gin(search_vector_en);
+CREATE INDEX IF NOT EXISTS idx_traqr_memories_search_simple
+  ON traqr_memories USING gin(search_vector_simple);
+
+-- traqr_memory_history
+CREATE INDEX IF NOT EXISTS idx_traqr_memory_history_memory
+  ON traqr_memory_history(memory_id);
+CREATE INDEX IF NOT EXISTS idx_traqr_memory_history_changed
+  ON traqr_memory_history(changed_at DESC);
+
+-- memory_relationships
+CREATE INDEX IF NOT EXISTS idx_rel_source
+  ON memory_relationships(source_memory_id);
+CREATE INDEX IF NOT EXISTS idx_rel_target
+  ON memory_relationships(target_memory_id);
+CREATE INDEX IF NOT EXISTS idx_rel_edge
+  ON memory_relationships(edge_type);
+
+-- memory_entities
+CREATE INDEX IF NOT EXISTS idx_entities_user
+  ON memory_entities(user_id);
+CREATE INDEX IF NOT EXISTS idx_entities_type
+  ON memory_entities(entity_type);
+CREATE INDEX IF NOT EXISTS idx_entities_name
+  ON memory_entities(name);
+CREATE INDEX IF NOT EXISTS idx_entities_mentions
+  ON memory_entities(mentions_count DESC);
+CREATE INDEX IF NOT EXISTS idx_entities_embedding
+  ON memory_entities USING hnsw (embedding vector_cosine_ops);
+
+-- memory_entity_links
+CREATE INDEX IF NOT EXISTS idx_entity_links_memory
+  ON memory_entity_links(memory_id);
+CREATE INDEX IF NOT EXISTS idx_entity_links_entity
+  ON memory_entity_links(entity_id);
+
+
+-- ============================================================================
+-- 5. CONDITIONAL ROW-LEVEL SECURITY
+-- ============================================================================
+-- auth.role() only exists on Supabase. Wrap ALL RLS in a DO block that
+-- checks for the function's existence first.
+
+DO $$
+DECLARE
+  has_auth_role BOOLEAN;
+BEGIN
+  -- Check if auth.role() exists (Supabase-specific)
+  SELECT EXISTS(
+    SELECT 1 FROM pg_proc p
+    JOIN pg_namespace n ON p.pronamespace = n.oid
+    WHERE n.nspname = 'auth' AND p.proname = 'role'
+  ) INTO has_auth_role;
+
+  -- Enable RLS on all tables (safe everywhere)
+  ALTER TABLE memory_users ENABLE ROW LEVEL SECURITY;
+  ALTER TABLE memory_domains ENABLE ROW LEVEL SECURITY;
+  ALTER TABLE traqr_memories ENABLE ROW LEVEL SECURITY;
+  ALTER TABLE traqr_memory_history ENABLE ROW LEVEL SECURITY;
+  ALTER TABLE memory_relationships ENABLE ROW LEVEL SECURITY;
+  ALTER TABLE memory_entities ENABLE ROW LEVEL SECURITY;
+  ALTER TABLE memory_entity_links ENABLE ROW LEVEL SECURITY;
+  ALTER TABLE schema_version ENABLE ROW LEVEL SECURITY;
+
+  IF has_auth_role THEN
+    -- Supabase: restrict to service_role
+    BEGIN
+      CREATE POLICY "Service role full access on memory_users"
+        ON memory_users FOR ALL
+        USING (auth.role() = 'service_role')
+        WITH CHECK (auth.role() = 'service_role');
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Service role full access on memory_domains"
+        ON memory_domains FOR ALL
+        USING (auth.role() = 'service_role')
+        WITH CHECK (auth.role() = 'service_role');
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Service role full access on traqr_memories"
+        ON traqr_memories FOR ALL
+        USING (auth.role() = 'service_role')
+        WITH CHECK (auth.role() = 'service_role');
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Service role full access on traqr_memory_history"
+        ON traqr_memory_history FOR ALL
+        USING (auth.role() = 'service_role')
+        WITH CHECK (auth.role() = 'service_role');
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Service role full access on relationships"
+        ON memory_relationships FOR ALL
+        USING (auth.role() = 'service_role')
+        WITH CHECK (auth.role() = 'service_role');
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Service role full access on entities"
+        ON memory_entities FOR ALL
+        USING (auth.role() = 'service_role')
+        WITH CHECK (auth.role() = 'service_role');
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Service role full access on entity_links"
+        ON memory_entity_links FOR ALL
+        USING (auth.role() = 'service_role')
+        WITH CHECK (auth.role() = 'service_role');
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Service role full access on schema_version"
+        ON schema_version FOR ALL
+        USING (auth.role() = 'service_role')
+        WITH CHECK (auth.role() = 'service_role');
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+  ELSE
+    -- Non-Supabase (RDS, local): open access policies
+    BEGIN
+      CREATE POLICY "Full access on memory_users"
+        ON memory_users FOR ALL USING (true) WITH CHECK (true);
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Full access on memory_domains"
+        ON memory_domains FOR ALL USING (true) WITH CHECK (true);
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Full access on traqr_memories"
+        ON traqr_memories FOR ALL USING (true) WITH CHECK (true);
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Full access on traqr_memory_history"
+        ON traqr_memory_history FOR ALL USING (true) WITH CHECK (true);
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Full access on relationships"
+        ON memory_relationships FOR ALL USING (true) WITH CHECK (true);
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Full access on entities"
+        ON memory_entities FOR ALL USING (true) WITH CHECK (true);
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Full access on entity_links"
+        ON memory_entity_links FOR ALL USING (true) WITH CHECK (true);
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+
+    BEGIN
+      CREATE POLICY "Full access on schema_version"
+        ON schema_version FOR ALL USING (true) WITH CHECK (true);
+    EXCEPTION WHEN duplicate_object THEN NULL;
+    END;
+  END IF;
+END $$;
+
+
+-- ============================================================================
+-- 6. RPC FUNCTIONS (v2 versions only)
+-- ============================================================================
+
+-- 6a. calculate_current_confidence — v2: 5-arg, STABLE (uses NOW())
+CREATE OR REPLACE FUNCTION calculate_current_confidence(
+  p_original_confidence FLOAT,
+  p_created_at TIMESTAMP WITH TIME ZONE,
+  p_times_cited INTEGER DEFAULT 0,
+  p_times_returned INTEGER DEFAULT 0,
+  p_memory_type VARCHAR DEFAULT 'pattern'
+)
+RETURNS FLOAT
+LANGUAGE plpgsql STABLE AS $$
+DECLARE
+  years_elapsed FLOAT;
+  decay_rate FLOAT;
+BEGIN
+  years_elapsed := EXTRACT(EPOCH FROM (NOW() - p_created_at)) / 31536000.0;
+
+  -- Type-aware decay rates (M5 Pipeline Design ADR)
+  IF p_memory_type = 'fact' THEN
+    decay_rate := 0.98;  -- 2%/year (facts barely decay)
+  ELSIF p_memory_type = 'preference' THEN
+    IF p_times_cited > 0 THEN
+      decay_rate := 0.90;  -- 10%/year if cited
+    ELSE
+      decay_rate := 0.85;  -- 15%/year if uncited
+    END IF;
+  ELSE  -- 'pattern' (default)
+    IF p_times_cited > 3 THEN
+      decay_rate := 0.95;  -- 5%/year (proven valuable)
+    ELSIF p_times_cited >= 1 THEN
+      decay_rate := 0.90;  -- 10%/year (moderate usage)
+    ELSIF p_times_returned > 5 AND p_times_cited = 0 THEN
+      decay_rate := 0.60;  -- 40%/year (noise: returned but never cited)
+    ELSE
+      decay_rate := 0.70;  -- 30%/year (default uncited)
+    END IF;
+  END IF;
+
+  RETURN GREATEST(0.1, p_original_confidence * POWER(decay_rate, years_elapsed));
+END;
+$$;
+
+-- 6b. search_memories — v2: with p_latest_only + v2 columns
+CREATE OR REPLACE FUNCTION search_memories(
+  p_query_embedding vector(1536),
+  p_project_id UUID DEFAULT NULL,
+  p_category VARCHAR DEFAULT NULL,
+  p_tags VARCHAR[] DEFAULT NULL,
+  p_include_archived BOOLEAN DEFAULT FALSE,
+  p_limit INTEGER DEFAULT 10,
+  p_similarity_threshold FLOAT DEFAULT 0.35,
+  p_latest_only BOOLEAN DEFAULT TRUE
+)
+RETURNS TABLE (
+  id UUID,
+  content TEXT,
+  summary VARCHAR,
+  category VARCHAR,
+  tags VARCHAR[],
+  context_tags VARCHAR[],
+  source_type VARCHAR,
+  source_ref VARCHAR,
+  source_project VARCHAR,
+  original_confidence FLOAT,
+  current_confidence FLOAT,
+  similarity FLOAT,
+  relevance_score FLOAT,
+  created_at TIMESTAMP WITH TIME ZONE,
+  updated_at TIMESTAMP WITH TIME ZONE,
+  durability VARCHAR,
+  is_universal BOOLEAN,
+  times_returned INTEGER,
+  times_cited INTEGER,
+  last_returned_at TIMESTAMP WITH TIME ZONE,
+  last_cited_at TIMESTAMP WITH TIME ZONE,
+  memory_type VARCHAR,
+  valid_at TIMESTAMP WITH TIME ZONE,
+  invalid_at TIMESTAMP WITH TIME ZONE,
+  is_latest BOOLEAN,
+  source_tool VARCHAR
+)
+LANGUAGE plpgsql AS $$
+BEGIN
+  RETURN QUERY
+  SELECT
+    m.id, m.content, m.summary, m.category, m.tags, m.context_tags,
+    m.source_type, m.source_ref, m.source_project,
+    m.original_confidence,
+    calculate_current_confidence(
+      m.original_confidence, m.created_at, m.times_cited, m.times_returned, m.memory_type
+    ) AS current_confidence,
+    1 - (m.embedding <=> p_query_embedding) AS similarity,
+    (1 - (m.embedding <=> p_query_embedding))
+      * calculate_current_confidence(
+          m.original_confidence, m.created_at, m.times_cited, m.times_returned, m.memory_type
+        )
+      * (1 + ln(1 + m.times_cited) * 0.1) AS relevance_score,
+    m.created_at, m.updated_at, m.durability, m.is_universal,
+    m.times_returned, m.times_cited, m.last_returned_at, m.last_cited_at,
+    m.memory_type, m.valid_at, m.invalid_at, m.is_latest, m.source_tool
+  FROM traqr_memories m
+  WHERE (p_include_archived OR m.is_archived = FALSE)
+    AND m.is_forgotten = FALSE
+    AND (NOT p_latest_only OR m.is_latest = TRUE)
+    AND (m.invalid_at IS NULL OR m.invalid_at > NOW())
+    AND (p_project_id IS NULL OR m.project_id = p_project_id)
+    AND (p_category IS NULL OR m.category = p_category)
+    AND (p_tags IS NULL OR m.tags && p_tags)
+    AND 1 - (m.embedding <=> p_query_embedding) >= p_similarity_threshold
+  ORDER BY relevance_score DESC
+  LIMIT p_limit;
+END;
+$$;
+
+-- 6c. search_memories_cross_project — v2: with p_latest_only + v2 columns
+CREATE OR REPLACE FUNCTION search_memories_cross_project(
+  p_query_embedding vector(1536),
+  p_project_id UUID DEFAULT NULL,
+  p_source_project VARCHAR DEFAULT NULL,
+  p_category VARCHAR DEFAULT NULL,
+  p_tags VARCHAR[] DEFAULT NULL,
+  p_include_archived BOOLEAN DEFAULT FALSE,
+  p_include_portable BOOLEAN DEFAULT TRUE,
+  p_agent_type VARCHAR DEFAULT NULL,
+  p_limit INTEGER DEFAULT 10,
+  p_similarity_threshold FLOAT DEFAULT 0.35,
+  p_latest_only BOOLEAN DEFAULT TRUE
+)
+RETURNS TABLE (
+  id UUID,
+  content TEXT,
+  summary VARCHAR,
+  category VARCHAR,
+  tags VARCHAR[],
+  context_tags VARCHAR[],
+  source_type VARCHAR,
+  source_ref VARCHAR,
+  source_project VARCHAR,
+  original_confidence FLOAT,
+  current_confidence FLOAT,
+  similarity FLOAT,
+  relevance_score FLOAT,
+  created_at TIMESTAMP WITH TIME ZONE,
+  updated_at TIMESTAMP WITH TIME ZONE,
+  durability VARCHAR,
+  is_universal BOOLEAN,
+  times_returned INTEGER,
+  times_cited INTEGER,
+  last_returned_at TIMESTAMP WITH TIME ZONE,
+  last_cited_at TIMESTAMP WITH TIME ZONE,
+  memory_type VARCHAR,
+  valid_at TIMESTAMP WITH TIME ZONE,
+  invalid_at TIMESTAMP WITH TIME ZONE,
+  is_latest BOOLEAN,
+  source_tool VARCHAR
+)
+LANGUAGE plpgsql AS $$
+BEGIN
+  RETURN QUERY
+  SELECT
+    m.id, m.content, m.summary, m.category, m.tags, m.context_tags,
+    m.source_type, m.source_ref, m.source_project,
+    m.original_confidence,
+    calculate_current_confidence(
+      m.original_confidence, m.created_at, m.times_cited, m.times_returned, m.memory_type
+    ) AS current_confidence,
+    1 - (m.embedding <=> p_query_embedding) AS similarity,
+    (1 - (m.embedding <=> p_query_embedding))
+      * calculate_current_confidence(
+          m.original_confidence, m.created_at, m.times_cited, m.times_returned, m.memory_type
+        )
+      * (1 + ln(1 + m.times_cited) * 0.1) AS relevance_score,
+    m.created_at, m.updated_at, m.durability, m.is_universal,
+    m.times_returned, m.times_cited, m.last_returned_at, m.last_cited_at,
+    m.memory_type, m.valid_at, m.invalid_at, m.is_latest, m.source_tool
+  FROM traqr_memories m
+  WHERE (p_include_archived OR m.is_archived = FALSE)
+    AND m.is_forgotten = FALSE
+    AND (NOT p_latest_only OR m.is_latest = TRUE)
+    AND (m.invalid_at IS NULL OR m.invalid_at > NOW())
+    AND (
+      (p_project_id IS NOT NULL AND m.project_id = p_project_id)
+      OR (p_source_project IS NOT NULL AND m.source_project = p_source_project)
+      OR (p_include_portable AND m.is_universal = TRUE)
+    )
+    AND (p_category IS NULL OR m.category = p_category)
+    AND (p_tags IS NULL OR m.tags && p_tags)
+    AND (p_agent_type IS NULL OR m.agent_type = p_agent_type)
+    AND 1 - (m.embedding <=> p_query_embedding) >= p_similarity_threshold
+  ORDER BY relevance_score DESC
+  LIMIT p_limit;
+END;
+$$;
+
+-- 6d. archive_decayed_memories — v2: uses 5-arg confidence
+CREATE OR REPLACE FUNCTION archive_decayed_memories()
+RETURNS TABLE (
+  archived_id UUID,
+  content_preview TEXT,
+  final_confidence FLOAT,
+  times_cited INTEGER,
+  times_returned INTEGER,
+  reason TEXT
+)
+LANGUAGE plpgsql AS $$
+BEGIN
+  RETURN QUERY
+  WITH decayed AS (
+    SELECT
+      m.id,
+      LEFT(m.content, 100) AS content_preview,
+      calculate_current_confidence(
+        m.original_confidence, m.created_at, m.times_cited, m.times_returned, m.memory_type
+      ) AS conf,
+      m.times_cited,
+      m.times_returned,
+      CASE
+        WHEN m.times_returned > 5 AND m.times_cited = 0 THEN 'noise'
+        WHEN m.times_cited = 0 THEN 'uncited'
+        ELSE 'low-confidence'
+      END AS reason
+    FROM traqr_memories m
+    WHERE m.is_archived = FALSE
+      AND m.is_forgotten = FALSE
+      AND calculate_current_confidence(
+        m.original_confidence, m.created_at, m.times_cited, m.times_returned, m.memory_type
+      ) < 0.3
+  ),
+  archived AS (
+    UPDATE traqr_memories
+    SET is_archived = TRUE,
+        archived_at = NOW(),
+        archive_reason = d.reason,
+        updated_at = NOW()
+    FROM decayed d
+    WHERE traqr_memories.id = d.id
+    RETURNING traqr_memories.id
+  )
+  SELECT d.id, d.content_preview, d.conf, d.times_cited, d.times_returned, d.reason
+  FROM decayed d
+  JOIN archived a ON a.id = d.id;
+END;
+$$;
+
+-- 6e. bm25_search — keyword search over dual tsvectors
+-- NOTE: ts_rank_cd returns REAL; cast to ::FLOAT. VARCHAR columns need ::TEXT.
+CREATE OR REPLACE FUNCTION bm25_search(
+  p_query_text TEXT,
+  p_project_id UUID DEFAULT NULL,
+  p_domain TEXT DEFAULT NULL,
+  p_category TEXT DEFAULT NULL,
+  p_limit INTEGER DEFAULT 20,
+  p_min_score FLOAT DEFAULT 0.01
+)
+RETURNS TABLE (
+  id UUID,
+  content TEXT,
+  summary TEXT,
+  bm25_score FLOAT,
+  domain TEXT,
+  category TEXT,
+  memory_type TEXT
+)
+LANGUAGE plpgsql AS $$
+DECLARE
+  tsquery_en tsquery;
+  tsquery_simple tsquery;
+BEGIN
+  tsquery_en := plainto_tsquery('english', p_query_text);
+  tsquery_simple := plainto_tsquery('simple', p_query_text);
+
+  RETURN QUERY
+  SELECT
+    m.id,
+    m.content,
+    m.summary::TEXT,
+    GREATEST(
+      ts_rank_cd(m.search_vector_en, tsquery_en),
+      ts_rank_cd(m.search_vector_simple, tsquery_simple)
+    )::FLOAT AS bm25_score,
+    m.domain::TEXT,
+    m.category::TEXT,
+    m.memory_type::TEXT
+  FROM traqr_memories m
+  WHERE (m.search_vector_en @@ tsquery_en
+         OR m.search_vector_simple @@ tsquery_simple)
+    AND m.is_archived = FALSE
+    AND m.is_forgotten = FALSE
+    AND (m.invalid_at IS NULL OR m.invalid_at > NOW())
+    AND (p_project_id IS NULL OR m.project_id = p_project_id)
+    AND (p_domain IS NULL OR m.domain = p_domain)
+    AND (p_category IS NULL OR m.category = p_category)
+  ORDER BY bm25_score DESC
+  LIMIT p_limit;
+END;
+$$;
+
+-- 6f. temporal_search — valid_at range + embedding similarity
+CREATE OR REPLACE FUNCTION temporal_search(
+  p_query_embedding vector(1536),
+  p_date_start TIMESTAMPTZ,
+  p_date_end TIMESTAMPTZ,
+  p_project_id UUID DEFAULT NULL,
+  p_similarity_threshold FLOAT DEFAULT 0.3,
+  p_limit INTEGER DEFAULT 20
+)
+RETURNS TABLE (
+  id UUID,
+  content TEXT,
+  summary TEXT,
+  similarity FLOAT,
+  temporal_proximity FLOAT,
+  valid_at TIMESTAMPTZ
+)
+LANGUAGE plpgsql AS $$
+DECLARE
+  date_mid TIMESTAMPTZ;
+  total_days FLOAT;
+BEGIN
+  date_mid := p_date_start + (p_date_end - p_date_start) / 2;
+  total_days := GREATEST(EXTRACT(EPOCH FROM (p_date_end - p_date_start)) / 86400.0, 1.0);
+
+  RETURN QUERY
+  SELECT
+    m.id,
+    m.content,
+    m.summary,
+    1 - (m.embedding <=> p_query_embedding) AS similarity,
+    GREATEST(0.0, 1.0 - (
+      ABS(EXTRACT(EPOCH FROM (m.valid_at - date_mid)) / 86400.0) / (total_days / 2.0)
+    )) AS temporal_proximity,
+    m.valid_at
+  FROM traqr_memories m
+  WHERE m.valid_at BETWEEN p_date_start AND p_date_end
+    AND m.is_archived = FALSE
+    AND m.is_forgotten = FALSE
+    AND (p_project_id IS NULL OR m.project_id = p_project_id)
+    AND 1 - (m.embedding <=> p_query_embedding) >= p_similarity_threshold
+  ORDER BY similarity DESC
+  LIMIT p_limit;
+END;
+$$;
+
+-- 6g. graph_search — link expansion CTE traversing memory_relationships
+CREATE OR REPLACE FUNCTION graph_search(
+  p_seed_ids UUID[],
+  p_edge_types TEXT[] DEFAULT ARRAY['updates', 'extends', 'derives', 'related'],
+  p_max_depth INTEGER DEFAULT 2,
+  p_limit INTEGER DEFAULT 20
+)
+RETURNS TABLE (
+  id UUID,
+  content TEXT,
+  summary TEXT,
+  graph_score FLOAT,
+  edge_type TEXT,
+  depth INTEGER
+)
+LANGUAGE plpgsql AS $$
+BEGIN
+  RETURN QUERY
+  WITH RECURSIVE graph_walk AS (
+    -- Seed: direct neighbors of seed memories
+    SELECT
+      mr.target_memory_id AS memory_id,
+      mr.edge_type,
+      mr.confidence AS score,
+      1 AS depth
+    FROM memory_relationships mr
+    WHERE mr.source_memory_id = ANY(p_seed_ids)
+      AND mr.edge_type = ANY(p_edge_types)
+
+    UNION ALL
+
+    -- Expand: neighbors of neighbors (0.7 decay per hop)
+    SELECT
+      mr.target_memory_id,
+      mr.edge_type,
+      gw.score * mr.confidence * 0.7 AS score,
+      gw.depth + 1
+    FROM graph_walk gw
+    JOIN memory_relationships mr ON mr.source_memory_id = gw.memory_id
+    WHERE gw.depth < p_max_depth
+      AND mr.edge_type = ANY(p_edge_types)
+  )
+  SELECT
+    m.id,
+    m.content,
+    m.summary,
+    MAX(gw.score) AS graph_score,
+    (ARRAY_AGG(gw.edge_type ORDER BY gw.score DESC))[1] AS edge_type,
+    MIN(gw.depth) AS depth
+  FROM graph_walk gw
+  JOIN traqr_memories m ON m.id = gw.memory_id
+  WHERE m.is_archived = FALSE AND m.is_forgotten = FALSE
+  GROUP BY m.id, m.content, m.summary
+  ORDER BY graph_score DESC
+  LIMIT p_limit;
+END;
+$$;
+
+-- 6h. search_entities — embedding-based entity lookup
+CREATE OR REPLACE FUNCTION search_entities(
+  p_user_id UUID,
+  p_embedding vector(1536),
+  p_entity_type VARCHAR DEFAULT NULL,
+  p_threshold FLOAT DEFAULT 0.85,
+  p_limit INTEGER DEFAULT 5
+)
+RETURNS TABLE (
+  id UUID,
+  name VARCHAR,
+  entity_type VARCHAR,
+  similarity FLOAT,
+  mentions_count INTEGER
+)
+LANGUAGE plpgsql AS $$
+BEGIN
+  RETURN QUERY
+  SELECT
+    e.id, e.name, e.entity_type,
+    1 - (e.embedding <=> p_embedding) AS similarity,
+    e.mentions_count
+  FROM memory_entities e
+  WHERE e.user_id = p_user_id
+    AND e.is_archived = FALSE
+    AND (p_entity_type IS NULL OR e.entity_type = p_entity_type)
+    AND 1 - (e.embedding <=> p_embedding) >= p_threshold
+  ORDER BY similarity DESC
+  LIMIT p_limit;
+END;
+$$;
+
+-- 6i. count_entity_mentions — 3+ threshold check
+CREATE OR REPLACE FUNCTION count_entity_mentions(
+  p_user_id UUID,
+  p_name VARCHAR
+)
+RETURNS INTEGER
+LANGUAGE plpgsql AS $$
+DECLARE
+  mention_count INTEGER;
+BEGIN
+  SELECT COUNT(*) INTO mention_count
+  FROM traqr_memories m
+  WHERE m.user_id = p_user_id
+    AND m.is_archived = FALSE
+    AND m.is_forgotten = FALSE
+    AND m.content ILIKE '%' || p_name || '%';
+  RETURN mention_count;
+END;
+$$;
+
+-- 6j. forget_expired_memories — daily cron function
+CREATE OR REPLACE FUNCTION forget_expired_memories()
+RETURNS INTEGER
+LANGUAGE plpgsql AS $$
+DECLARE
+  forgotten_count INTEGER;
+BEGIN
+  WITH to_forget AS (
+    UPDATE traqr_memories
+    SET is_forgotten = TRUE,
+        forgotten_at = NOW(),
+        updated_at = NOW()
+    WHERE is_forgotten = FALSE
+      AND forget_after IS NOT NULL
+      AND forget_after < NOW()
+    RETURNING id
+  )
+  SELECT COUNT(*) INTO forgotten_count FROM to_forget;
+  RETURN forgotten_count;
+END;
+$$;
+
+-- 6k. increment_memory_returns — batch fire-and-forget from context assembly
+CREATE OR REPLACE FUNCTION increment_memory_returns(p_memory_ids UUID[])
+RETURNS void AS $$
+BEGIN
+  UPDATE traqr_memories
+  SET
+    times_returned = COALESCE(times_returned, 0) + 1,
+    last_returned_at = NOW()
+  WHERE id = ANY(p_memory_ids)
+    AND is_archived = FALSE;
+END;
+$$ LANGUAGE plpgsql;
+
+-- 6l. cite_memory — record a citation (resets decay timer)
+CREATE OR REPLACE FUNCTION cite_memory(p_memory_id UUID)
+RETURNS void AS $$
+BEGIN
+  UPDATE traqr_memories
+  SET
+    times_cited = COALESCE(times_cited, 0) + 1,
+    last_cited_at = NOW(),
+    last_validated = NOW()
+  WHERE id = p_memory_id
+    AND is_archived = FALSE;
+END;
+$$ LANGUAGE plpgsql;
+
+
+-- ============================================================================
+-- 7. SCHEMA VERSION TRACKING + MIGRATION BOOTSTRAP
+-- ============================================================================
+
+INSERT INTO schema_version (version, description)
+VALUES (1, 'Memory Engine v1 -- base schema (migrations 001-010)')
+ON CONFLICT (version) DO NOTHING;
+
+INSERT INTO schema_version (version, description)
+VALUES (2, 'Memory Engine v2 -- pipeline, retrieval, temporal, entities (M5-M9)')
+ON CONFLICT (version) DO NOTHING;
+
+-- Migration tracking table (so migrate.ts runner works for future migrations)
+CREATE TABLE IF NOT EXISTS _traqr_migrations (
+  name TEXT PRIMARY KEY,
+  applied_at TIMESTAMPTZ DEFAULT NOW()
+);
+
+INSERT INTO _traqr_migrations (name) VALUES
+  ('001_memory_schema.sql'),
+  ('002_slack_task_queue.sql'),
+  ('003_cross_project_memory.sql'),
+  ('004_memory_durability.sql'),
+  ('005_memory_citation_tracking.sql'),
+  ('006_memory_curation_functions.sql'),
+  ('007_accelerated_decay.sql'),
+  ('008_audit_cleanup.sql'),
+  ('009_quality_audit_cleanup.sql'),
+  ('010_citation_boosted_ranking.sql'),
+  ('011_memory_engine_v2_schema.sql')
+ON CONFLICT DO NOTHING;
+
+
+-- ============================================================================
+-- DONE. Verify with:
+-- ============================================================================
+--
+-- SELECT tablename FROM pg_tables WHERE schemaname = 'public'
+--   AND tablename IN ('memory_users','memory_domains','traqr_memories',
+--     'traqr_memory_history','memory_relationships','memory_entities',
+--     'memory_entity_links','schema_version','_traqr_migrations');
+--
+-- SELECT column_name, data_type FROM information_schema.columns
+--   WHERE table_name = 'traqr_memories' ORDER BY ordinal_position;
+--
+-- SELECT indexname FROM pg_indexes WHERE tablename = 'traqr_memories';
+--
+-- SELECT * FROM bm25_search('test query', NULL, NULL, NULL, 5, 0.01);