diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index f601bf9..6c26a43 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -14,7 +14,7 @@ jobs:
 
       # The curriculum gate comes first, before anything else can run.
       # validate.mjs reads only data/ and lib/ and exits non-zero on a
-      # blocking failure. Baseline: PASS — 0 blocking, 167 advisory.
+      # blocking failure. Baseline: PASS — 0 blocking, 0 advisory.
       - name: Curriculum validation
         run: node validate.mjs
 
@@ -28,7 +28,35 @@ jobs:
       - name: Typecheck
         run: npm run typecheck
 
+      # The sync round-trip needs a real Postgres and a running server, so a
+      # fake would not exercise the change_seq trigger, the last-write-wins
+      # upsert, or cursor paging — which is where sync actually goes wrong.
+      - name: Start Postgres
+        run: |
+          docker run -d --name hankan-pg \
+            -e POSTGRES_PASSWORD=test -e POSTGRES_DB=hankan \
+            -p 55432:5432 postgres:16-alpine
+          for i in $(seq 1 30); do
+            docker exec hankan-pg pg_isready -U postgres -d hankan && break
+            sleep 1
+          done
+
+      - name: Start the sync server
+        env:
+          DATABASE_URL: postgres://postgres:test@localhost:55432/hankan
+          HANKAN_TOKEN: test-token
+          HANKAN_TEST_MODE: "1"
+          PORT: "8788"
+        run: |
+          node --experimental-strip-types server/src/main.ts &
+          for i in $(seq 1 30); do
+            curl -sf http://localhost:8788/health && break
+            sleep 1
+          done
+
       - name: Tests
+        env:
+          HANKAN_TEST_SERVER: http://localhost:8788
         run: npm test
 
       # Every word in every unit's words[] must resolve in lemma or surface.
diff --git a/README.md b/README.md
index 7ba0e9f..69e92c0 100644
--- a/README.md
+++ b/README.md
@@ -18,11 +18,13 @@ types/                                 TypeScript declarations for lib/ and shar
 tools/dict/                            the dictionary build pipeline
 app/                                   Vite + React + TypeScript, and the Capacitor shell
   src/db/                              one storage interface, two SQLite drivers
-  src/domain/                          the gate, the lexicon, SRS, the stub tutor
+  src/domain/                          the gate, the lexicon, SRS, the tutor clients
+  src/sync/                            push/pull against the Pi
   src/ui/                              six tabs, the review overlay, the 한글 keyboard
   public/dict/                         generated dictionary — committed, shipped
-test/                                  golden tests for lib/, driver conformance, domain logic
-server/                                placeholder; PORT.md steps 4-6
+server/                                sync + the tutor endpoint (optional)
+tools/icons/                           the 한칸 mark, generated
+test/                                  lib goldens, driver conformance, domain, sync, SSE
 ```
 
 `data/`, `lib/`, `prompt/` and `validate.mjs` are byte-identical to the export
@@ -144,11 +146,32 @@ that file.
 
 CI runs them in that order, `validate.mjs` first.
 
+## Sync and the tutor — optional, and genuinely optional
+
+The app is complete without a server: its own SQLite, the shipped dictionary,
+and a local stand-in tutor. Pointing it at a Pi adds two things — the real
+선생님, and syncing between devices. Everything degrades to the offline
+behaviour when the server is unreachable, and a sync failure is recorded in
+settings rather than surfaced as an interruption.
+
+Setup, the Caddy config, and how sync resolves conflicts: [server/README.md](server/README.md).
+
+Three details worth knowing here:
+
+- **Seeded rows carry `updated_at = 0`**, so a fresh device can neither push
+  its empty defaults nor win a conflict with them. The artifact's clobbering
+  bug is unrepresentable rather than merely avoided, and
+  `test/sync/roundtrip.test.ts` asserts it against a real Postgres.
+- **An allowlist decides what leaves the device.** `meta` mixes preferences
+  with per-install bookkeeping; `dict.loadedBands` crossing the wire would
+  tell a phone it holds rows it never downloaded.
+- **The system prompt is cached.** It is ~12k characters of gate, identical
+  for as long as the learner stays in one unit, so every turn after the first
+  reads it at a fraction of the input price.
+
 ## Not in this pass
 
-No sync layer, no tutor server endpoint, no Play Store packaging — PORT.md steps
-4–6. The schema carries `updated_at` on every syncable row and the dictionary is
-fetched by URL, so both land without rework.
+Android beyond the existing Capacitor scaffold, and Play Store packaging.
 
 ## Known limitation