diff --git a/CHANGELOG.md b/CHANGELOG.md
index b11b9af3..574762cd 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,6 +4,7 @@
 
 ### Fixed
 - `tear_down_after_script` runs when `set_up_before_script` fails, so it can release file-scoped resources acquired before the failure (#1318)
+- Under `--parallel`, `tear_down_after_script` runs after the file's own tests instead of alongside them, so a fixture it releases stays alive for the tests that read it. The same file no longer passed sequentially and failed in parallel (#1320)
 
 ## [0.50.0](https://github.com/TypedDevs/bashunit/compare/0.49.0...0.50.0) - 2026-08-18
 
diff --git a/docs/ai-agents.md b/docs/ai-agents.md
index ea9c9fa7..e339a93a 100644
--- a/docs/ai-agents.md
+++ b/docs/ai-agents.md
@@ -141,8 +141,6 @@ actually make against this API:
   tell `cmd "a b"` from `cmd a b` — a quoting bug passes it. Use
   `assert_have_been_called_with_args <spy> <arg>...` (no `call_index`) to compare
   argument by argument whenever an argument may contain a space.
-- Do not delete a shared fixture in `tear_down_after_script`: under `--parallel` the
-  file's tests may still be running, and they will vanish from the totals silently.
 - Re-record snapshots with `--snapshot-update`, scoped by `--filter`; do not delete
   snapshot files. A missing snapshot is written silently and never fails, so a wrong
   `rm` turns a real assertion into a rubber stamp. Read `git diff` afterwards either way.
diff --git a/src/runner/discovery.sh b/src/runner/discovery.sh
index 1b232702..9d980ab2 100644
--- a/src/runner/discovery.sh
+++ b/src/runner/discovery.sh
@@ -228,11 +228,22 @@ function bashunit::runner::load_test_files() {
       worker_stderr_paths[worker_stderr_count]="$_worker_stderr"
       worker_stderr_owners[worker_stderr_count]="$test_file"
       worker_stderr_count=$((worker_stderr_count + 1))
-      bashunit::runner::call_test_functions "$test_file" "$_cached_fns" 2>"$_worker_stderr" &
+      # The file's teardown belongs inside the worker. Run from this shell it
+      # released the fixture while the worker's tests were still reading it, so
+      # the same file passed sequentially and failed under --parallel (#1320).
+      # call_test_functions waits for its own per-test workers before it
+      # returns, which is what makes this ordering hold.
+      {
+        bashunit::runner::call_test_functions "$test_file" "$_cached_fns"
+        bashunit::runner::run_tear_down_after_script "$test_file"
+        # A hook failure recorded in here dies with the subshell (#1147), so
+        # publish it the way a test publishes its result.
+        bashunit::runner::publish_file_hook_failure "$?" "$test_file"
+      } 2>"$_worker_stderr" &
     else
       bashunit::runner::call_test_functions "$test_file" "$_cached_fns"
+      bashunit::runner::run_tear_down_after_script "$test_file"
     fi
-    bashunit::runner::run_tear_down_after_script "$test_file"
     bashunit::runner::clean_script_test_functions "$_script_fns_to_clean"
     bashunit::runner::clean_set_up_and_tear_down_after_script
     if ! bashunit::parallel::is_enabled; then
diff --git a/src/runner/result.sh b/src/runner/result.sh
index 2474adf2..900c7d40 100644
--- a/src/runner/result.sh
+++ b/src/runner/result.sh
@@ -62,6 +62,48 @@ function bashunit::runner::parse_result_parallel() {
   echo "$execution_result" >"$unique_test_result_file"
 }
 
+##
+# Publishes a file-scoped hook failure from a --parallel worker, so the parent's
+# summary counts it.
+#
+# tear_down_after_script runs inside the file's worker (#1320) and
+# record_file_hook_failure's counter lives in that subshell, so without this the
+# hook failed, said so on the terminal, and left the run green. The aggregator
+# reads one `.result` per test out of the file's suite dir, so the hook publishes
+# its own. A fixed basename: the per-test ordinals are scoped inside
+# call_test_functions, and the aggregator globs `*.result`.
+#
+# Written only on failure. A zero-assertion, exit-0 payload is what the
+# aggregator reads as a risky test, which would add a phantom risky test to every
+# file that defines the hook.
+#
+# Arguments: $1 - the hook's exit status, $2 - the test file
+##
+function bashunit::runner::publish_file_hook_failure() {
+  local status=$1
+  local test_file=$2
+
+  [ "$status" -ne 0 ] || return 0
+  bashunit::parallel::is_enabled || return 0
+
+  bashunit::runner::parallel_suite_dir_to_slot "$test_file"
+  local test_suite_dir=$_BASHUNIT_RUNNER_SUITE_DIR_OUT
+  [ -d "$test_suite_dir" ] || mkdir -p "$test_suite_dir"
+
+  # Hand-built rather than exported: export_subshell_context would carry the
+  # assertion totals this file's own tests already published, and the aggregator
+  # would count every one of them twice. The hook contributes a failed test, no
+  # assertions.
+  local payload="\
+##ASSERTIONS_FAILED=0\
+##ASSERTIONS_PASSED=0\
+##ASSERTIONS_SKIPPED=0\
+##ASSERTIONS_INCOMPLETE=0\
+##ASSERTIONS_SNAPSHOT=0\
+##TEST_EXIT_CODE=$status##"
+  printf '%s\n' "$payload" >"$test_suite_dir/tear_down_after_script.result"
+}
+
 function bashunit::runner::parse_result_sync() {
   local fn_name=$1
   local execution_result=$2
diff --git a/tests/acceptance/bashunit_parallel_consistency_test.sh b/tests/acceptance/bashunit_parallel_consistency_test.sh
index eb4e7f4e..9a03b3ac 100644
--- a/tests/acceptance/bashunit_parallel_consistency_test.sh
+++ b/tests/acceptance/bashunit_parallel_consistency_test.sh
@@ -104,3 +104,39 @@ Assertions: 5 passed, 5 total"
 
   assert_same "3 passed 3 total 5 passed 5 total " "$(summary_counts "$capture")"
 }
+
+# A file-scoped teardown has to run after the file's tests, not alongside them.
+# Under --parallel the runner dispatched the worker and ran the hook on the very
+# next line, so a resource acquired in set_up_before_script vanished mid-test and
+# the file failed in parallel while passing sequentially (#1320).
+function test_parallel_runs_tear_down_after_script_after_the_files_tests() {
+  local dir fixture
+  dir="$(bashunit::temp_dir parallel_file_teardown)"
+  fixture="$dir/shared_fixture_test.sh"
+  {
+    printf 'function set_up_before_script() {\n'
+    printf '  : >"$SHARED_FIXTURE"\n'
+    printf '}\n'
+    printf 'function tear_down_after_script() {\n'
+    printf '  rm -f "$SHARED_FIXTURE"\n'
+    printf '}\n'
+    printf 'function test_reads_the_shared_fixture() {\n'
+    printf '  sleep 0.5\n'
+    printf '  assert_file_exists "$SHARED_FIXTURE"\n'
+    printf '}\n'
+  } >"$fixture"
+
+  local sequential_output parallel_output
+  sequential_output=$(SHARED_FIXTURE="$dir/sequential.fixture" NO_COLOR=1 \
+    ./bashunit --no-parallel --env "$TEST_ENV_FILE" "$fixture" 2>&1) || true
+  parallel_output=$(SHARED_FIXTURE="$dir/parallel.fixture" NO_COLOR=1 \
+    ./bashunit --parallel --env "$TEST_ENV_FILE" "$fixture" 2>&1) || true
+
+  local sequential_counts parallel_counts
+  sequential_counts=$(summary_counts "$sequential_output")
+  parallel_counts=$(summary_counts "$parallel_output")
+
+  assert_same "$sequential_counts" "$parallel_counts"
+  # Guard against both sides failing the same way and matching vacuously.
+  assert_contains "1 passed 1 total" "$sequential_counts"
+}
diff --git a/tests/acceptance/bashunit_teardown_after_script_error_test.sh b/tests/acceptance/bashunit_teardown_after_script_error_test.sh
index 30dfb2e6..15a51dc2 100644
--- a/tests/acceptance/bashunit_teardown_after_script_error_test.sh
+++ b/tests/acceptance/bashunit_teardown_after_script_error_test.sh
@@ -84,3 +84,30 @@ test_bashunit_when_teardown_after_script_with_intermediate_failing_command.sh
   assert_contains "$assertions_summary" "$actual"
   assert_general_error "$(./bashunit --no-parallel --env "$TEST_ENV_FILE" "$test_file")"
 }
+
+# Under --parallel the hook runs inside the file's worker (#1320), where its own
+# counter dies with the subshell the way #1147 describes. The count reaches the
+# parent through publish_file_hook_failure. Drop that and the hook error still
+# prints while the run reports "All tests passed", so pin the count here.
+function test_bashunit_when_tear_down_after_script_errors_in_parallel() {
+  local test_file=./tests/acceptance/fixtures/test_bashunit_when_teardown_after_script_errors.sh
+
+  local error_line="✗ Error: Tear down after script"
+  local message_line="missing_cleanup_command"
+  local tests_summary="Tests:      1 passed, 1 failed, 2 total"
+  local assertions_summary="Assertions: 1 passed, 0 failed, 1 total"
+
+  local actual_raw
+  set +e
+  actual_raw="$(./bashunit --parallel --detailed --env "$TEST_ENV_FILE" "$test_file")"
+  set -e
+
+  local actual
+  actual="$(printf "%s" "$actual_raw" | strip_ansi)"
+
+  assert_contains "$error_line" "$actual"
+  assert_contains "$message_line" "$actual"
+  assert_contains "$tests_summary" "$actual"
+  assert_contains "$assertions_summary" "$actual"
+  assert_general_error "$(./bashunit --parallel --env "$TEST_ENV_FILE" "$test_file")"
+}