diff --git a/docs/reviews/history_delete_http500_fix_v1_evidence.md b/docs/reviews/history_delete_http500_fix_v1_evidence.md
index 36a1f3b..2d5c8ed 100644
--- a/docs/reviews/history_delete_http500_fix_v1_evidence.md
+++ b/docs/reviews/history_delete_http500_fix_v1_evidence.md
@@ -53,3 +53,7 @@ production full-generation consumer now filters only live facts already
 dominated by causal state applied from that complete envelope; standalone
 snapshot imports retain their strict rejection. A regression exercises this
 ordering and proves the tombstone and absent session remain unchanged.
+The recovery envelope also republishes both consumed and rejected correlated
+ACKs. This lets Android close an interrupted generation with the real durable
+desktop rejection evidence and reclaim normal bounded capacity; no generation,
+ACK, tombstone or archive is manually purged or synthesized.
diff --git a/tests/test_sync_error_classification.py b/tests/test_sync_error_classification.py
index 21bc875..1645edd 100644
--- a/tests/test_sync_error_classification.py
+++ b/tests/test_sync_error_classification.py
@@ -1,4 +1,5 @@
 import json
+import sqlite3
 import subprocess
 import sys
 import tempfile
@@ -15,6 +16,32 @@ import sync_peer_worker  # noqa: E402
 
 
 class SyncErrorClassificationTest(unittest.TestCase):
+    def test_recovery_envelope_includes_consumed_and_rejected_acknowledgements(self):
+        with sqlite3.connect(":memory:") as database:
+            database.execute(
+                "CREATE TABLE sync_consumed_generations("
+                "generation_id TEXT,producer_peer_id TEXT,consumer_peer_id TEXT,"
+                "consumed_at TEXT,result TEXT,ack_json TEXT)"
+            )
+            for generation, result in (("gen_consumed", "consumed"),
+                                       ("gen_rejected", "rejected"),
+                                       ("gen_other", "consumed")):
+                database.execute(
+                    "INSERT INTO sync_consumed_generations VALUES(?,?,?,?,?,?)",
+                    (generation, "android" if generation != "gen_other" else "other",
+                     "desktop", generation, result,
+                     json.dumps({"generation_id": generation, "result": result})),
+                )
+
+            acknowledgements = sync_peer_worker.recovery_acknowledgements(
+                database, "android", "desktop"
+            )
+
+        self.assertEqual(
+            [(value["generation_id"], value["result"]) for value in acknowledgements],
+            [("gen_consumed", "consumed"), ("gen_rejected", "rejected")],
+        )
+
     def test_mtp_refreshes_a_retained_peer_advertisement_before_validation(self):
         expected_peer = "peer_00000000-0000-4000-8000-000000000001"
         with tempfile.TemporaryDirectory() as directory:
diff --git a/tools/sync_peer_worker.py b/tools/sync_peer_worker.py
index 1c2ec0a..4f63522 100644
--- a/tools/sync_peer_worker.py
+++ b/tools/sync_peer_worker.py
@@ -67,6 +67,17 @@ def publish_json(path: Path, value: dict) -> None:
     os.replace(temporary, path)
 
 
+def recovery_acknowledgements(db, producer_peer_id: str, consumer_peer_id: str) -> list[dict]:
+    """Load bounded terminal ACK evidence for interrupted peer conversations."""
+    rows = db.execute(
+        "SELECT ack_json FROM sync_consumed_generations WHERE producer_peer_id=? "
+        "AND consumer_peer_id=? AND result IN('consumed','rejected') "
+        "ORDER BY consumed_at,generation_id LIMIT 32",
+        (producer_peer_id, consumer_peer_id),
+    ).fetchall()
+    return [json.loads(row[0]) for row in rows]
+
+
 def run_adapter(
     adapter: Path,
     operation: str,
@@ -309,18 +320,14 @@ def main() -> int:
     request_path = args.transport_root / "request-v1.json"
     publish_json(request_path, request)
     with closing(generation.connect_database(args.database)) as db:
-        acknowledgement_rows = db.execute(
-            "SELECT ack_json FROM sync_consumed_generations WHERE producer_peer_id=? "
-            "AND consumer_peer_id=? AND result='consumed' ORDER BY consumed_at,generation_id LIMIT 32",
-            (peer["peer_id"], desktop_peer),
-        ).fetchall()
+        acknowledgements = recovery_acknowledgements(db, peer["peer_id"], desktop_peer)
     archive_acknowledgements = {
         "format": "trainlog-sync-archive-acknowledgements",
         "version": 1,
         "run_id": args.run_id,
         "android_peer_id": peer["peer_id"],
         "desktop_peer_id": desktop_peer,
-        "acknowledgements": [json.loads(row[0]) for row in acknowledgement_rows],
+        "acknowledgements": acknowledgements,
     }
     archive_ack_path = args.transport_root / "desktop-archive-acknowledgements-v1.json"
     publish_json(archive_ack_path, archive_acknowledgements)