diff --git a/.github/scripts/run_seekdb_migration_test.py b/.github/scripts/run_seekdb_migration_test.py
new file mode 100644
index 0000000000..7f3e81ccd2
--- /dev/null
+++ b/.github/scripts/run_seekdb_migration_test.py
@@ -0,0 +1,192 @@
+# Copyright (c) 2026 OceanBase.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Bound one native migration test and capture lock evidence before terminating it."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import signal
+import subprocess
+import sys
+import tarfile
+import time
+from contextlib import suppress
+from pathlib import Path
+
+_QUERIES = {
+    "sessions": "SELECT ID, COMMAND, TIME, STATE FROM information_schema.PROCESSLIST LIMIT 100",
+    "locks": "SELECT * FROM oceanbase.V$OB_LOCKS LIMIT 100",
+    "transactions": "SELECT * FROM oceanbase.V$OB_TRANSACTION_SCHEDULERS LIMIT 100",
+    "startup_jobs": "SELECT * FROM oceanbase.DBA_OB_ROOTSERVICE_JOBS LIMIT 100",
+}
+
+
+def archive_native_logs(root: Path, output: Path) -> None:
+    # Preserve the complete available history, including rotated logs, rather
+    # than dropping the transaction origin when lock-wait messages fill a tail.
+    # Only archive logs from disposable test instances, never database files.
+    with tarfile.open(output, "w:gz", compresslevel=1) as archive:
+        for native in sorted(root.glob("pc-*/seekdb/log/*.log*")):
+            if native.is_file() and not native.is_symlink():
+                try:
+                    archive.add(native, arcname=str(native.relative_to(root)), recursive=False)
+                except FileNotFoundError:
+                    # Fixture cleanup or log rotation can race collection.
+                    print(f"Log disappeared during collection: {native}", flush=True)
+
+
+def snapshot(root: Path) -> None:
+    import pymysql
+
+    # Connect to the existing socket directly. Opening a pylibseekdb lifecycle
+    # handle here could restart the server or change the lifetime under test.
+    for socket in sorted(root.glob("pc-*/seekdb/run/sql.sock"))[:4]:
+        print(json.dumps({"socket": str(socket)}), flush=True)
+        try:
+            connection = pymysql.connect(
+                unix_socket=str(socket),
+                user="root",
+                database="test",
+                autocommit=True,
+                connect_timeout=2,
+                read_timeout=2,
+                write_timeout=2,
+            )
+            try:
+                with connection.cursor() as cursor:
+                    cursor.execute("SET SESSION ob_query_timeout = 2000000")
+                    for name, query in _QUERIES.items():
+                        try:
+                            cursor.execute(query)
+                        except pymysql.MySQLError as error:
+                            print(json.dumps({"query": name, "error": type(error).__name__}), flush=True)
+                            continue
+                        print(
+                            json.dumps(
+                                {
+                                    "query": name,
+                                    "columns": [column[0] for column in cursor.description],
+                                    "rows": cursor.fetchall(),
+                                },
+                                default=str,
+                                ensure_ascii=False,
+                            ),
+                            flush=True,
+                        )
+            finally:
+                connection.close()
+        except pymysql.MySQLError as error:
+            print(json.dumps({"error": type(error).__name__, "code": error.args[0]}), flush=True)
+
+
+def stop(process: subprocess.Popen[bytes]) -> None:
+    # Include native children in the test's process group when pytest cannot
+    # reach fixture cleanup. Never kill other tests or an existing user server.
+    with suppress(ProcessLookupError):
+        os.killpg(process.pid, signal.SIGTERM)
+    try:
+        process.wait(timeout=5)
+    except subprocess.TimeoutExpired:
+        with suppress(ProcessLookupError):
+            os.killpg(process.pid, signal.SIGKILL)
+        process.wait()
+
+
+def diagnostics(root: Path, output: Path) -> None:
+    with output.open("a", encoding="utf-8") as log:
+        probe = subprocess.Popen(  # noqa: S603 -- fixed interpreter and this checked-in diagnostic script
+            [sys.executable, "-u", __file__, "--snapshot", str(root)],
+            stdout=log,
+            stderr=subprocess.STDOUT,
+            start_new_session=True,
+        )
+        try:
+            probe.wait(timeout=10)
+        except subprocess.TimeoutExpired:
+            stop(probe)
+            log.write("Database snapshot exceeded its diagnostic deadline.\n")
+        # The holder may have acquired its lock during startup. A tail alone
+        # loses that origin when the waiter emits minutes of repetitive logs.
+        for native in sorted(root.glob("pc-*/seekdb/log/*.log*"))[:4]:
+            if not native.is_file() or native.is_symlink():
+                continue
+            # The test may finish and remove its temporary database while the
+            # snapshot is running; that must not turn a passing test red.
+            with suppress(OSError):
+                log.write(f"\nNative log: {native}\n")
+                with native.open("rb") as source:
+                    log.write(source.read(2 * 1024 * 1024).decode("utf-8", errors="replace"))
+                    if native.stat().st_size > 2 * 1024 * 1024:
+                        log.write("\n--- tail ---\n")
+                        source.seek(max(source.tell(), native.stat().st_size - 2 * 1024 * 1024))
+                        log.write(source.read(2 * 1024 * 1024).decode("utf-8", errors="replace"))
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--snapshot", type=Path)
+    parser.add_argument("--archive-native-logs", type=Path)
+    parser.add_argument("--diagnose-after", type=float, default=90)
+    parser.add_argument("--timeout", type=float, default=180)
+    parser.add_argument("--diagnostics", type=Path, default=Path("seekdb-live-diagnostics.log"))
+    parser.add_argument("node", nargs="?")
+    args = parser.parse_args()
+    if args.snapshot is not None:
+        snapshot(args.snapshot)
+        return 0
+    if args.archive_native_logs is not None:
+        archive_native_logs(Path(os.environ["TMPDIR"]), args.archive_native_logs)
+        return 0
+    if not args.node or not 0 < args.diagnose_after < args.timeout:
+        parser.error("provide a pytest node and 0 < diagnose-after < timeout")
+    root = Path(os.environ["TMPDIR"])
+    started = time.monotonic()
+    process = subprocess.Popen(  # noqa: S603 -- pytest node comes from CI collection, without a shell
+        [
+            sys.executable,
+            "-m",
+            "pytest",
+            "-vv",
+            "-o",
+            "faulthandler_timeout=120",
+            "--durations=20",
+            "--log-cli-level=INFO",
+            args.node,
+        ],
+        start_new_session=True,
+    )
+    try:
+        try:
+            return process.wait(timeout=args.diagnose_after)
+        except subprocess.TimeoutExpired:
+            print(f"Capturing seekdb lock diagnostics for {args.node}", flush=True)
+            try:
+                diagnostics(root, args.diagnostics)
+            except OSError as error:
+                print(f"Diagnostic capture failed: {type(error).__name__}", flush=True)
+        try:
+            return process.wait(timeout=max(0, args.timeout - (time.monotonic() - started)))
+        except subprocess.TimeoutExpired:
+            print(f"seekdb test exceeded {args.timeout:g}s: {args.node}", flush=True)
+            return 124
+    finally:
+        if process.poll() is None:
+            stop(process)
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/.github/workflows/master.yml b/.github/workflows/master.yml
index 88c3c900d3..b85b6275b8 100644
--- a/.github/workflows/master.yml
+++ b/.github/workflows/master.yml
@@ -181,11 +181,13 @@ jobs:
           mkdir -p "$TMPDIR"
           collect_diagnostics() {
             df -h . /tmp
-            # A timed-out pytest process cannot run fixture cleanup. Preserve
-            # native startup/recovery evidence too: a tail alone can omit the
-            # transaction that acquired the lock blocking the final SQL query.
-            find "$TMPDIR" -type f \( -name '*.log' -o -name '*.log.wf' \) \
-              -printf '%P\0' | tar -czf seekdb-native.tar.gz -C "$TMPDIR" --null -T -
+            # A timed-out pytest process cannot run fixture cleanup. Keep a
+            # short text tail and the complete available native log history,
+            # including rotated files, to retain the original lock holder.
+            find "$TMPDIR" -type f \( -name '*.log' -o -name '*.log.*' \) \
+              -print -exec tail -c 8388608 {} \; > seekdb-native.log
+            .venv/bin/python .github/scripts/run_seekdb_migration_test.py \
+              --archive-native-logs seekdb-native.tar.gz
           }
           trap collect_diagnostics EXIT
           df -h . /tmp
@@ -208,10 +210,16 @@ jobs:
             exit 1
           fi
           for seekdb_runner_node in "${seekdb_runner_nodes[@]}"; do
-            timeout --signal=TERM --kill-after=10s 180s \
-              uv run --locked --extra seekdb pytest -vv \
-              -o faulthandler_timeout=120 --durations=20 --log-cli-level=INFO \
-              "$seekdb_runner_node" -k seekdb \
+            uv run --locked --extra seekdb python .github/scripts/run_seekdb_migration_test.py \
+              "$seekdb_runner_node" \
+              2>&1 | tee -a seekdb-migrations.log
+          done
+          # These are required independent runs, not retries: any failure stops
+          # the job. Exercise the lifecycle race on both supported matrix ends.
+          for repetition in 2 3; do
+            echo "Baseline adoption stability run $repetition/3" | tee -a seekdb-migrations.log
+            uv run --locked --extra seekdb python .github/scripts/run_seekdb_migration_test.py \
+              'tests/builtin/persistence/test_mysql_migration_runner.py::test_mysql_baseline_adoption_recovers_an_empty_committed_version_table[seekdb]' \
               2>&1 | tee -a seekdb-migrations.log
           done
 
@@ -222,7 +230,9 @@ jobs:
           name: seekdb-migrations-${{ matrix.python-version }}
           path: |
             seekdb-migrations.log
+            seekdb-native.log
             seekdb-native.tar.gz
+            seekdb-live-diagnostics.log
           if-no-files-found: ignore
 
   atomic-memory-clean-seekdb:
diff --git a/.github/workflows/seekdb-lock-observation.yml b/.github/workflows/seekdb-lock-observation.yml
new file mode 100644
index 0000000000..30df53f1a6
--- /dev/null
+++ b/.github/workflows/seekdb-lock-observation.yml
@@ -0,0 +1,50 @@
+name: SeekDB extended lock observation
+
+on:
+  pull_request:
+    paths:
+      - .github/workflows/seekdb-lock-observation.yml
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: seekdb-lock-observation-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: false
+
+jobs:
+  observe:
+    runs-on: ubuntu-latest
+    timeout-minutes: 35
+    steps:
+      - name: Check out the documented reproduction snapshot
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
+        with:
+          ref: 0b4dd71c0e5f8189096f2958e79007440c438da2
+          persist-credentials: false
+
+      - name: Set up the environment
+        uses: ./.github/actions/setup-python-env
+        with:
+          python-version: "3.11"
+
+      - name: Observe lock recovery beyond the normal test deadline
+        timeout-minutes: 30
+        run: |
+          git rev-parse HEAD
+          # Keep the ordinary migration CI deadline unchanged. This standalone
+          # experiment records snapshots every 45 seconds for up to 20 minutes
+          # per attempt, including beyond the observed 1000-second SQL timeout.
+          uv run --locked --extra seekdb python scripts/probe_seekdb_migration_lock.py \
+            --attempts 12 --timeout 1200 --output .powercontext/seekdb-lock-observation
+
+      - name: Upload observation evidence
+        if: always()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
+        with:
+          name: seekdb-lock-observation-${{ github.run_id }}-${{ github.run_attempt }}
+          path: .powercontext/seekdb-lock-observation
+          include-hidden-files: true
+          if-no-files-found: error
+          retention-days: 7