diff --git a/migrations/211_md_canonical_layer.sql b/migrations/211_md_canonical_layer.sql
new file mode 100644
index 0000000..dd2c24b
--- /dev/null
+++ b/migrations/211_md_canonical_layer.sql
@@ -0,0 +1,58 @@
+-- 211_md_canonical_layer.sql
+-- Phase 1A: Markdown canonical layer (schema only).
+-- Plan: ~/.claude/plans/plan-idempotent-sundae.md (round 3 approved)
+--
+-- 어떤 워커도 이 컬럼을 채우지 않는다 (Phase 1A scope).
+-- 신규 컬럼은 모두 nullable 또는 default 보장 → 기존 코드 경로 영향 0.
+-- 상태값은 TEXT+CHECK (확장 시 enum drop/rebuild 비용 회피).
+-- Postgres 16.13: ADD COLUMN ... DEFAULT <constant> 는 fast path (table rewrite 없음).
+
+ALTER TABLE documents
+    ADD COLUMN md_content                   TEXT,
+    ADD COLUMN md_frontmatter               JSONB        NOT NULL DEFAULT '{}'::jsonb,
+    ADD COLUMN md_format_version            TEXT         NOT NULL DEFAULT '1.0',
+    ADD COLUMN md_status                    TEXT         NOT NULL DEFAULT 'pending'
+        CHECK (md_status IN ('pending','processing','success','partial','failed','skipped')),
+    ADD COLUMN md_extraction_engine         TEXT,
+    ADD COLUMN md_extraction_engine_version TEXT,
+    ADD COLUMN md_extraction_quality        JSONB,
+    ADD COLUMN md_extraction_error          TEXT,
+    ADD COLUMN md_content_hash              TEXT,
+    ADD COLUMN md_source_hash               TEXT,
+    ADD COLUMN md_generated_at              TIMESTAMPTZ,
+    -- content_origin: canonical markdown 생성 경로 분류.
+    --   'extracted' = 업로드 원본에서 추출되어야 하는 계열 (완료 여부는 md_status가 결정).
+    --   'manual'    = 사람이 UI에서 직접 작성한 markdown.
+    --   'ai_drafted'= LLM이 검색 evidence 기반으로 작성 (Phase 4).
+    --   'imported'  = 외부 시스템/기존 markdown에서 가져온 문서 (DEVONthink 등).
+    ADD COLUMN content_origin               TEXT         NOT NULL DEFAULT 'extracted'
+        CHECK (content_origin IN ('extracted','manual','ai_drafted','imported')),
+    -- md_draft_status: AI-generated Markdown review lifecycle only.
+    -- 사용 조건은 별도 CHECK (documents_md_draft_status_only_ai)에서 강제.
+    ADD COLUMN md_draft_status              TEXT
+        CHECK (md_draft_status IS NULL OR md_draft_status IN ('draft','pending_review','approved','revised','rejected'));
+
+ALTER TABLE documents
+    ADD CONSTRAINT documents_md_draft_status_only_ai
+    CHECK (md_draft_status IS NULL OR content_origin = 'ai_drafted');
+
+-- 큐 폴링용 (pending/processing만)
+CREATE INDEX idx_documents_md_status_pending
+    ON documents (md_status)
+    WHERE md_status IN ('pending', 'processing');
+
+-- content_origin 기반 필터 (AI 생성문서 분리 등)
+CREATE INDEX idx_documents_content_origin
+    ON documents (content_origin);
+
+-- frontmatter 기반 검색 (project_code, doc_type 등 GIN).
+-- Phase 1A 시점엔 모든 행이 '{}' 이므로 즉시 활용 X.
+-- Phase 1B 이후 md_frontmatter 필터링/조회를 위한 선제 인덱스.
+-- INSERT latency 문제 발생 시 1순위 rollback 대상.
+CREATE INDEX idx_documents_md_frontmatter_gin
+    ON documents USING GIN (md_frontmatter);
+
+-- 검토 큐 (Phase 4 이후)
+CREATE INDEX idx_documents_md_draft_status
+    ON documents (md_draft_status)
+    WHERE md_draft_status IS NOT NULL;
diff --git a/migrations/212_document_lineage.sql b/migrations/212_document_lineage.sql
new file mode 100644
index 0000000..7d61565
--- /dev/null
+++ b/migrations/212_document_lineage.sql
@@ -0,0 +1,29 @@
+-- 212_document_lineage.sql
+-- Phase 1A: AI 생성 문서가 어느 원본에서 파생됐는지 추적.
+-- Plan: ~/.claude/plans/plan-idempotent-sundae.md (round 3 approved)
+--
+-- 이력 테이블 FK = ON DELETE RESTRICT (feedback_history_table_fk_restrict).
+-- 부모 문서 hard delete 차단, soft delete (documents.deleted_at) 만 허용.
+-- relation_type 도 TEXT+CHECK (확장 시 enum drop/rebuild 비용 회피).
+
+CREATE TABLE document_lineage (
+    id                  BIGSERIAL PRIMARY KEY,
+    source_document_id  BIGINT NOT NULL REFERENCES documents(id) ON DELETE RESTRICT,
+    derived_document_id BIGINT NOT NULL REFERENCES documents(id) ON DELETE RESTRICT,
+    relation_type       TEXT NOT NULL
+        CHECK (relation_type IN ('cited','summarized_from','generated_from','revised_from')),
+    metadata            JSONB NOT NULL DEFAULT '{}'::jsonb,
+    created_at          TIMESTAMPTZ NOT NULL DEFAULT now(),
+    CONSTRAINT document_lineage_no_self
+        CHECK (source_document_id <> derived_document_id),
+    CONSTRAINT document_lineage_uq
+        UNIQUE (source_document_id, derived_document_id, relation_type)
+);
+
+-- reverse-lookup: '이 문서가 어느 생성문서의 source 였나'
+CREATE INDEX idx_document_lineage_source
+    ON document_lineage (source_document_id);
+
+-- forward-lookup: '이 생성문서의 source 는 무엇이었나'
+CREATE INDEX idx_document_lineage_derived
+    ON document_lineage (derived_document_id);