diff --git a/AGENTS.md b/AGENTS.md
index 8694f97..f8f5ca3 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -40,7 +40,7 @@ Indonesian (`karung`=sack, `truk`=truck); YOLO class names are English
   office `:5721` only (403 on `:5000`); **`ocr_engine` POST both ports** (synced
   via `GET /api/do/settings`). Discard batch only if **both** `net_sack==0` and
   `net_box==0`. DO helpers: `src/do_batch.py`, OCR: `src/do_ocr.py`;
-  ERD: `docs/do-erd.md`.
+  ERD: `ERD.md` (repo root).
 - **Counting filters by class name, not ID**: `SackDetector`/`BoxDetector`/
   `TruckDetector` filter via `BaseDetector(class_filter)` (`src/detection.py`);
   tracker keeps `("sack", "truck", "box")` (`src/tracking.py`); counting uses
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 407acc8..4be3159 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -22,7 +22,7 @@ version tags are cut in this repo — POC stage, `main` is the release line).
   (`extract_do_fields` tesseract|paddle|none).
 - Operator DO panel + OCR toggle; monitoring 3-way mode switch + model mode +
   OCR/require-plate controls; history/export plate, DO, expected, net columns.
-- `docs/do-erd.md` (Mermaid ERD); tests `tests/test_do_batch.py`.
+- `ERD.md` (Mermaid ERD, repo root; superseded `docs/do-erd.md`); tests `tests/test_do_batch.py`.
 
 ### Changed
 - Default batch mode seed **`auto`** (was hard-coded `manual` when
diff --git a/ERD.md b/ERD.md
new file mode 100644
index 0000000..bff5089
--- /dev/null
+++ b/ERD.md
@@ -0,0 +1,115 @@
+# ERD — karung counter (SQLite)
+
+Data model for the karung feedmill counter. SQLite DB at
+`$OUTPUT_DIR/jetson_counter.db` (`config.yaml` `output.dir`,
+default `/opt/jetson-counter/`; env override `DB_PATH`).
+
+**No DB foreign keys enforced** (zero `FOREIGN KEY` in code) — all relationships
+are soft, by naming/best-effort writes.
+
+```mermaid
+erDiagram
+    batches ||--o{ delivery_orders : "soft via batch_id / do_numbers"
+    batches ||--o{ daily_summaries : "rollup by counting_date"
+
+    batches {
+        INTEGER id PK
+        TEXT counting_date "NOT NULL, YYYY-MM-DD, cutoff-shifted"
+        INTEGER batch_number "NOT NULL, per date+camera+label"
+        TEXT camera_name "NOT NULL, default CC1"
+        TEXT object_label "NOT NULL, default karung-pakan"
+        INTEGER count "NOT NULL, gross sack loading"
+        TEXT start_time "NOT NULL, ISO datetime"
+        TEXT end_time "NOT NULL, ISO datetime"
+        TIMESTAMP created_at "default CURRENT_TIMESTAMP"
+        INTEGER box_loading "NOT NULL, default 0"
+        INTEGER box_unloading "NOT NULL, default 0"
+        TEXT model_mode "NOT NULL, default A"
+        TEXT plate "NOT NULL, default ''"
+        TEXT do_numbers "NOT NULL, JSON array of No. DO"
+        INTEGER expected_sack "NOT NULL, default 0"
+        INTEGER expected_box "NOT NULL, default 0"
+        INTEGER net_sack "NOT NULL, default 0"
+        INTEGER net_box "NOT NULL, default 0"
+    }
+
+    delivery_orders {
+        INTEGER id PK
+        TEXT counting_date "NOT NULL"
+        TEXT photo_path "NOT NULL, relative under do_photos/"
+        TEXT no_do "default ''"
+        TEXT plate "default ''"
+        INTEGER expected_sack "default 0"
+        INTEGER expected_box "default 0"
+        TEXT ocr_raw "nullable JSON: no_do, plate, ocr_text, engine, ocr_ok, ocr_error"
+        TEXT status "default 'draft'"
+        INTEGER batch_id "soft ref to batches.id, nullable"
+        TIMESTAMP created_at "default CURRENT_TIMESTAMP"
+        TIMESTAMP updated_at "default CURRENT_TIMESTAMP"
+    }
+
+    daily_summaries {
+        INTEGER id PK
+        TEXT counting_date "NOT NULL"
+        TEXT camera_name "NOT NULL"
+        TEXT object_label "NOT NULL"
+        INTEGER total_count "NOT NULL, default 0, SUM(batches.count)"
+        INTEGER total_batches "NOT NULL, default 0, COUNT(batches.id)"
+        TIMESTAMP updated_at "default CURRENT_TIMESTAMP"
+    }
+```
+
+## Keys
+
+- `batches`: `UNIQUE(counting_date, batch_number, camera_name, object_label)`
+- `daily_summaries`: `UNIQUE(counting_date, camera_name, object_label)`
+- `delivery_orders`: `idx_do_date_status` on `(counting_date, status)`
+
+## Relationships (logical)
+
+| From | To | How |
+|---|---|---|
+| `delivery_orders` → `batches` | N—1 | On stop: `batches.do_numbers` JSON array of `No. DO`; `delivery_orders.batch_id` set best-effort. Active batch: `current_batch.json.do_ids`. |
+| `batches` → `daily_summaries` | N—1 | Same `(counting_date, camera_name, object_label)`; stop recompute `SUM(count)`, `COUNT(id)`. |
+
+## Derived columns
+
+- `net_sack = count − sack unloading`, `net_box = box_loading − box_unloading`
+  (`src/do_batch.py:37`).
+- **Sack unloading has no column** — it only survives inside `net_sack`
+  (legacy rows with `net_sack=0` from the `ALTER TABLE ... DEFAULT 0`
+  migration are not backfillable).
+- History UI shows gross `count` / `box_loading` only (no Net).
+
+## Status (`delivery_orders.status`)
+
+`draft` (photo upload) → `staged` (saved/edited) → `attached` (in
+active/stopped batch). Discard stop with both nets 0 returns DOs to
+`staged`. `discarded` appears in older notes but **no code writes it** —
+discard is a `DELETE`.
+
+## Hazards
+
+- `INSERT OR REPLACE` on `batches` (UNIQUE match = delete+insert) changes
+  `id` on re-stop → can orphan `delivery_orders.batch_id`.
+- Schema divergence: `predict.py init_db()` creates only `batches` +
+  `daily_summaries` (3 extra cols, no `delivery_orders`); full 3-table
+  schema (9 extra `batches` cols) comes from `counter_dashboard.py`
+  `_ensure_db()` at import. Fresh DB touched only by `predict.py` is
+  incomplete; its DO updates fail silently (`try/except: pass`).
+
+## JSON sidecars + photo tree (not in SQLite)
+
+| Path | Role |
+|---|---|
+| `$OUTPUT_DIR/batch_mode.json` | `{mode: auto\|do_manual\|manual}` |
+| `$OUTPUT_DIR/current_batch.json` | active batch + `do_ids`, plate, expected_*, nets inputs |
+| `$OUTPUT_DIR/do_settings.json` | `require_plate`, `require_do`, `ocr_engine` |
+| `$OUTPUT_DIR/do_photos/YYYY-MM-DD/do_<uuid>.jpg` | source photos (7-day retention) |
+
+## Sources
+
+- DDL: `counter_dashboard.py:123-198` (`_ensure_db`, superset),
+  `predict.py:117-166` (`init_db`, partial).
+- Writes: `predict.py:249-306`, `counter_dashboard.py:610-668`.
+- Doc flow: `docs/do-erd.md` (merged into this file).
diff --git a/README.md b/README.md
index 90857dd..3bb6b2f 100644
--- a/README.md
+++ b/README.md
@@ -101,7 +101,7 @@ Office (`:5721`) switches mode / model mode / require_plate (403 on operator
 POST). **DO manual flow:** open `http://<jetson>:5000/operator` on a phone →
 Ambil Foto DO → review/edit OCR fields → Mulai Batch → count → Selesai
 (soft-warn if zone busy). Discard only when both sack and box nets are 0.
-ERD: [`docs/do-erd.md`](docs/do-erd.md).
+ERD: [`ERD.md`](ERD.md).
 
 ## Repository layout
 
@@ -131,7 +131,7 @@ Script-by-script reference: [`docs/scripts.md`](docs/scripts.md).
 - [`docs/models.md`](docs/models.md) — model inventory & detection-mode support matrix
 - [`docs/configuration.md`](docs/configuration.md) — all config files/variables
 - [`docs/deployment.md`](docs/deployment.md) — Jetson services, TensorRT, deploy flow
-- [`docs/do-erd.md`](docs/do-erd.md) — delivery-order / batch ERD (Mermaid)
+- [`ERD.md`](ERD.md) — SQLite ERD (Mermaid): batches / delivery_orders / daily_summaries
 - [`docs/operator-tiles-history-plan.md`](docs/operator-tiles-history-plan.md) — operator tiles + history cleanup plan
 - [`docs/scripts.md`](docs/scripts.md) — entry points & utility scripts
 - [`CHANGELOG.md`](CHANGELOG.md) — release history
diff --git a/docs/architecture.md b/docs/architecture.md
index c8cfe0c..2b5226e 100644
--- a/docs/architecture.md
+++ b/docs/architecture.md
@@ -147,7 +147,7 @@ dashboard integration is file-based.)
 
 DO pipeline (dashboard): smartphone upload → OCR draft → edit/stage → start
 gates → active batch → stop-preview soft-warn → stop → `batches` + DO columns.
-ERD: [`docs/do-erd.md`](do-erd.md). UI plan: [`docs/operator-tiles-history-plan.md`](operator-tiles-history-plan.md).
+ERD: [`../ERD.md`](../ERD.md). UI plan: [`docs/operator-tiles-history-plan.md`](operator-tiles-history-plan.md).
 
 ## Retired (`archive/`, not imported)
 
diff --git a/docs/do-erd.md b/docs/do-erd.md
deleted file mode 100644
index 47cde62..0000000
--- a/docs/do-erd.md
+++ /dev/null
@@ -1,81 +0,0 @@
-# ERD — DO-gated manual batch (delivery orders)
-
-Mermaid ER for karung feedmill counter. Soft FK: no DB foreign keys enforced;
-`delivery_orders.batch_id` is informational once a batch row exists. While a
-batch is active, membership lives in `current_batch.json` (`do_ids`).
-
-```mermaid
-erDiagram
-    batches ||--o{ delivery_orders : "soft via do_numbers / batch_id"
-    batches ||--o{ daily_summaries : "rolls up by counting_date"
-
-    batches {
-        INTEGER id PK
-        TEXT counting_date
-        INTEGER batch_number
-        TEXT camera_name
-        TEXT object_label
-        INTEGER count
-        TEXT start_time
-        TEXT end_time
-        INTEGER box_loading
-        INTEGER box_unloading
-        TEXT model_mode
-        TEXT plate
-        TEXT do_numbers
-        INTEGER expected_sack
-        INTEGER expected_box
-        INTEGER net_sack
-        INTEGER net_box
-        TIMESTAMP created_at
-    }
-
-    delivery_orders {
-        INTEGER id PK
-        TEXT counting_date
-        TEXT photo_path
-        TEXT no_do
-        TEXT plate
-        INTEGER expected_sack
-        INTEGER expected_box
-        TEXT ocr_raw
-        TEXT status
-        INTEGER batch_id
-        TIMESTAMP created_at
-        TIMESTAMP updated_at
-    }
-
-    daily_summaries {
-        INTEGER id PK
-        TEXT counting_date
-        TEXT camera_name
-        TEXT object_label
-        INTEGER total_count
-        INTEGER total_batches
-        TIMESTAMP updated_at
-    }
-```
-
-## Relationships (logical)
-
-| From | To | How |
-|---|---|---|
-| `delivery_orders` → `batches` | N—1 | On stop: `batches.do_numbers` JSON array of `No. DO`; `delivery_orders.batch_id` set best-effort. Active batch: `current_batch.json.do_ids`. |
-| `batches` → `daily_summaries` | N—1 | Same counting_date/camera/label; stop recompute `SUM(count)`. |
-
-## Status (`delivery_orders.status`)
-
-`draft` → `staged` (saved/edited) → `attached` (in active/stopped batch) |
-`discarded` (deleted). Discarded stop (both nets 0) returns DOs to `staged`.
-
-## JSON sidecars + photo tree (not in SQLite)
-
-| Path | Role |
-|---|---|
-| `$OUTPUT_DIR/batch_mode.json` | `{mode: auto\|do_manual\|manual}` |
-| `$OUTPUT_DIR/current_batch.json` | active batch + `do_ids`, plate, expected_*, nets inputs |
-| `$OUTPUT_DIR/do_settings.json` | `require_plate`, `require_do`, `ocr_engine` |
-| `$OUTPUT_DIR/do_photos/YYYY-MM-DD/do_<uuid>.jpg` | source photos (7-day retention) |
-
-Schema DDL: `counter_dashboard.py` `_ensure_db()` (additive PRAGMA migrations
-on `batches` + `CREATE TABLE delivery_orders`).