diff --git a/packages/core/src/agent_memory/core/ledger.py b/packages/core/src/agent_memory/core/ledger.py
index f7d1c37b..fbafba3c 100644
--- a/packages/core/src/agent_memory/core/ledger.py
+++ b/packages/core/src/agent_memory/core/ledger.py
@@ -8,9 +8,11 @@
 from __future__ import annotations
 
 import dataclasses
-import pathlib
 import re
 
+from .locking import store_lock
+from .paths import StoreLayout
+
 LEDGER_FILENAME = "decisions.md"
 VERDICT_ACCEPTED = "accepted"
 VERDICT_REJECTED = "rejected"
@@ -41,8 +43,9 @@ def render(self) -> str:
 
 
 class DecisionLedger:
-    def __init__(self, path: pathlib.Path):
-        self._path = path
+    def __init__(self, layout: StoreLayout):
+        self._layout = layout
+        self._path = layout.dream_reports / LEDGER_FILENAME
 
     def decided(self) -> dict[str, Decision]:
         if not self._path.exists():
@@ -61,7 +64,14 @@ def decided(self) -> dict[str, Decision]:
         return found
 
     def append(self, decision: Decision) -> Decision:
-        self._path.parent.mkdir(parents=True, exist_ok=True)
-        head = self._path.read_text(encoding="utf-8") if self._path.exists() else HEADING + "\n\n"
-        self._path.write_text(head + decision.render() + "\n", encoding="utf-8")
+        with store_lock(self._layout):
+            self._path.parent.mkdir(parents=True, exist_ok=True)
+            head = (
+                self._path.read_text(encoding="utf-8")
+                if self._path.exists()
+                else HEADING + "\n\n"
+            )
+            staged = self._path.with_name(self._path.name + ".pending")
+            staged.write_text(head + decision.render() + "\n", encoding="utf-8")
+            staged.replace(self._path)
         return decision
diff --git a/packages/core/src/agent_memory/core/manage.py b/packages/core/src/agent_memory/core/manage.py
index 0f444ebd..2684cc43 100644
--- a/packages/core/src/agent_memory/core/manage.py
+++ b/packages/core/src/agent_memory/core/manage.py
@@ -23,7 +23,7 @@
 from .clock import Clock
 from .database import Database
 from .errors import FieldError, MemoryStoreError, NotFoundError, ValidationError
-from .ledger import LEDGER_FILENAME, VERDICT_ACCEPTED, VERDICT_REJECTED, Decision, DecisionLedger
+from .ledger import VERDICT_ACCEPTED, VERDICT_REJECTED, Decision, DecisionLedger
 from .pending import Pending
 from .record import DATE_FIELDS, MemoryRecord
 from .sessions import Pointer, parse_pointer
@@ -351,7 +351,7 @@ def _entry(self, name: str) -> MemoryRecord:
         return record
 
     def _ledger(self) -> DecisionLedger:
-        return DecisionLedger(self._store.layout.dream_reports / LEDGER_FILENAME)
+        return DecisionLedger(self._store.layout)
 
     def _usage(self) -> tuple[dict[str, int], dict[str, int], dict[str, str]]:
         """Lifetime counts decide what was never useful; only new reads earn weight, so one
diff --git a/tests/system/test_concurrency.py b/tests/system/test_concurrency.py
index a2db2e26..a2698d79 100644
--- a/tests/system/test_concurrency.py
+++ b/tests/system/test_concurrency.py
@@ -2,8 +2,11 @@
 
 import multiprocessing as mp
 
+from agent_memory.core.ledger import Decision, DecisionLedger
 from agent_memory.core.store import Store
 
+DECISIONS_PER_WRITER = 6
+
 
 def _write(root: str, name: str) -> None:
     store = Store(root, agent=name)
@@ -34,3 +37,31 @@ def test_two_processes_recording_at_once_lose_nothing(tmp_path):
     assert {"writer-alpha", "writer-beta"} <= names
     index_lines = store.layout.memory_index.read_text(encoding="utf-8")
     assert "writer-alpha" in index_lines and "writer-beta" in index_lines
+
+
+def _append(root: str, name: str) -> None:
+    ledger = DecisionLedger(Store(root).layout)
+    for step in range(DECISIONS_PER_WRITER):
+        ledger.append(
+            Decision(proposal_id=f"{name}-{step}", verdict="rejected", at="2026-01-01T00:00:00Z")
+        )
+
+
+def test_two_processes_appending_decisions_lose_nothing(tmp_path):
+    root = tmp_path / "store"
+    Store(root).init()
+    context = mp.get_context("spawn")
+    writers = ("writer-alpha", "writer-beta", "writer-gamma")
+    workers = [
+        context.Process(target=_append, args=(str(root), name))
+        for name in writers
+    ]
+    for worker in workers:
+        worker.start()
+    for worker in workers:
+        worker.join()
+    assert [worker.exitcode for worker in workers] == [0, 0, 0]
+
+    decided = DecisionLedger(Store(root).layout).decided()
+    expected = {f"{name}-{step}" for name in writers for step in range(DECISIONS_PER_WRITER)}
+    assert expected <= set(decided)