From 11b210adcdf6e87aaf2672b2d875e94c4690a000 Mon Sep 17 00:00:00 2001
From: LucaLin <78164141+LucaLin233@users.noreply.github.com>
Date: Tue, 8 Sep 2026 14:35:47 +0000
Subject: [PATCH 1/5] test(push): distinguish transfer failures from lifecycle
 barriers

---
 tests/test-push.sh | 21 +++++++++++++++++++--
 1 file changed, 19 insertions(+), 2 deletions(-)

diff --git a/tests/test-push.sh b/tests/test-push.sh
index b433c16..6adbdbe 100644
--- a/tests/test-push.sh
+++ b/tests/test-push.sh
@@ -1386,12 +1386,29 @@ EOF
     runtime_has_published_worker_state && fail "sliding batch left worker-session state"
     runtime_has_worker_registration_state && fail "sliding batch left registration state"
     : > "$SUCCESS_FILE"; : > "$FAILED_FILE"; : > "$root/state/current"; : > "$root/state/max"
+    diagnose_batch_contract() {
+        printf "DIAG: bash=%s status=%s lifecycle_failed=%s worker_error=%s\n" "$BASH_VERSION" "$1" "$BATCH_WORKER_FAILED" "$BATCH_WORKER_ERROR" >&2
+        declare -p ACTIVE_WORKERS ACTIVE_WORKER_STATE_FILES ACTIVE_WORKER_STATE_STARTS REGISTERING_WORKERS WORKER_REGISTRATION_READY_FILES WORKER_REGISTRATION_RELEASE_FILES WORKER_REGISTRATION_STARTS WORKER_REGISTRATION_STAGE_FILES >&2
+    }
+    assert_transfer_failure_contract() {
+        local label=$1 status=$2
+        if [[ $status != 1 || $BATCH_WORKER_FAILED != false || $BATCH_WORKER_ERROR != false ]]; then
+            diagnose_batch_contract "$status"
+            fail "$label must be ordinary transfer failure, not lifecycle/accounting failure"
+        fi
+        pass "$label preserves ordinary failure status"
+    }
     SERVERS=(good1 bad1 good2)
-    assert_fail "partial server failure returns nonzero" run_transfer source destination
+    partial_status=0
+    run_transfer source destination || partial_status=$?
+    assert_transfer_failure_contract "partial failure" "$partial_status"
+    assert_eq 2 "$(wc -l < "$SUCCESS_FILE")" "partial failure completes both unrelated servers"
     assert_eq 1 "$(wc -l < "$FAILED_FILE")" "partial failure records one failed server"
     : > "$SUCCESS_FILE"; : > "$FAILED_FILE"; : > "$root/state/current"; : > "$root/state/max"
     SERVERS=(bad1 bad2)
-    assert_fail "all server failures return nonzero" run_transfer source destination
+    all_status=0
+    run_transfer source destination || all_status=$?
+    assert_transfer_failure_contract "all failures" "$all_status"
     assert_eq 2 "$(wc -l < "$FAILED_FILE")" "all failures record every server"
     cleanup_runtime
 )

From b7548d5aa9bb76e1e9ca8bec4cbd3065d1068c51 Mon Sep 17 00:00:00 2001
From: LucaLin <78164141+LucaLin233@users.noreply.github.com>
Date: Tue, 8 Sep 2026 14:43:33 +0000
Subject: [PATCH 2/5] test(push): capture first batch lifecycle failure
 boundary

---
 tests/test-push.sh | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

diff --git a/tests/test-push.sh b/tests/test-push.sh
index 6adbdbe..9cc5d3a 100644
--- a/tests/test-push.sh
+++ b/tests/test-push.sh
@@ -1390,6 +1390,32 @@ EOF
         printf "DIAG: bash=%s status=%s lifecycle_failed=%s worker_error=%s\n" "$BASH_VERSION" "$1" "$BATCH_WORKER_FAILED" "$BATCH_WORKER_ERROR" >&2
         declare -p ACTIVE_WORKERS ACTIVE_WORKER_STATE_FILES ACTIVE_WORKER_STATE_STARTS REGISTERING_WORKERS WORKER_REGISTRATION_READY_FILES WORKER_REGISTRATION_RELEASE_FILES WORKER_REGISTRATION_STARTS WORKER_REGISTRATION_STAGE_FILES >&2
     }
+    # Test-local wrappers preserve return values and only log failing boundaries.
+    # No argument values, environment dumps, credentials, or global xtrace.
+    for diagnostic_function in launch_worker prune_active_workers wait_for_worker_slot wait_for_all_workers cleanup_active_failed_worker_sessions; do
+        eval "$(declare -f "$diagnostic_function" | sed "1s/$diagnostic_function/diagnostic_original_$diagnostic_function/")"
+    done
+    diagnostic_boundary() {
+        local name=$1 status=0
+        shift
+        "diagnostic_original_$name" "$@" || status=$?
+        if (( status != 0 )); then
+            printf "DIAG: boundary=%s pid=%s status=%s\n" "$name" "$BASHPID" "$status" >&2
+            diagnose_batch_contract "$status"
+        fi
+        return "$status"
+    }
+    launch_worker() { diagnostic_boundary launch_worker "$@"; }
+    prune_active_workers() { diagnostic_boundary prune_active_workers "$@"; }
+    wait_for_worker_slot() { diagnostic_boundary wait_for_worker_slot "$@"; }
+    wait_for_all_workers() { diagnostic_boundary wait_for_all_workers "$@"; }
+    cleanup_active_failed_worker_sessions() { diagnostic_boundary cleanup_active_failed_worker_sessions "$@"; }
+    eval "$(declare -f stop_batch_after_lifecycle_failure | sed "1s/stop_batch_after_lifecycle_failure/diagnostic_original_stop_batch_after_lifecycle_failure/")"
+    stop_batch_after_lifecycle_failure() {
+        printf "DIAG: barrier observed before lifecycle teardown\n" >&2
+        diagnose_batch_contract "$MANAGED_CLEANUP_FAILURE_STATUS"
+        diagnostic_original_stop_batch_after_lifecycle_failure
+    }
     assert_transfer_failure_contract() {
         local label=$1 status=$2
         if [[ $status != 1 || $BATCH_WORKER_FAILED != false || $BATCH_WORKER_ERROR != false ]]; then

From 4e540889b69703bbd3579743b38f0ef278ccf4fd Mon Sep 17 00:00:00 2001
From: LucaLin <78164141+LucaLin233@users.noreply.github.com>
Date: Tue, 8 Sep 2026 15:06:58 +0000
Subject: [PATCH 3/5] test(push): bound consecutive failure batch probes

---
 tests/test-push.sh | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/tests/test-push.sh b/tests/test-push.sh
index 9cc5d3a..153b478 100644
--- a/tests/test-push.sh
+++ b/tests/test-push.sh
@@ -1424,6 +1424,12 @@ EOF
         fi
         pass "$label preserves ordinary failure status"
     }
+    # Fixed probes: stop on first failure, never reset lifecycle barriers.
+    batch_probe_deadline=$((SECONDS + 90))
+    for batch_probe in 1 2 3 4 5 6 7 8; do
+    (( SECONDS < batch_probe_deadline )) || fail "batch probe budget exhausted"
+    printf "DIAG: consecutive failure batch round=%s/8\n" "$batch_probe"
+    : > "$SUCCESS_FILE"; : > "$FAILED_FILE"; : > "$root/state/current"; : > "$root/state/max"
     SERVERS=(good1 bad1 good2)
     partial_status=0
     run_transfer source destination || partial_status=$?
@@ -1436,6 +1442,7 @@ EOF
     run_transfer source destination || all_status=$?
     assert_transfer_failure_contract "all failures" "$all_status"
     assert_eq 2 "$(wc -l < "$FAILED_FILE")" "all failures record every server"
+    done
     cleanup_runtime
 )
 

From fc9dce8e50972f82664f69efaa2efdb9396ab35c Mon Sep 17 00:00:00 2001
From: LucaLin <78164141+LucaLin233@users.noreply.github.com>
Date: Tue, 8 Sep 2026 15:12:19 +0000
Subject: [PATCH 4/5] docs(test): record batch investigation and validate
 full-suite context

---
 docs/push-batch-investigation.md | 12 ++++++++++++
 1 file changed, 12 insertions(+)
 create mode 100644 docs/push-batch-investigation.md

diff --git a/docs/push-batch-investigation.md b/docs/push-batch-investigation.md
new file mode 100644
index 0000000..aea44fa
--- /dev/null
+++ b/docs/push-batch-investigation.md
@@ -0,0 +1,12 @@
+# Push batch lifecycle investigation
+
+Original failure: main commit 6e24bcbead3d352f305fb92bee9d894de441f554, run 34237080747, Debian full-suite context.
+The partial-transfer case entered lifecycle failure before the following batch failed accounting.
+
+Current probes require exact ordinary failure status and preserve lifecycle barriers. Eight fixed rounds stop on the first failure; no retry-to-green.
+Selective push-only runs have not reproduced the original failure. This document is an unknown dependency under the existing tests/run.sh mapping, intentionally exercising the existing full-suite fallback for this diagnostic PR. No workflow or selection rules are changed.
+
+Capture only Bash version, return codes and lifecycle maps. No credentials, full environment dump or global xtrace.
+A green full run is not proof of a fix. Retain original logs and distinguish fixture instrumentation from production behavior.
+
+Next evidence needed: first failing boundary and pre-teardown state under the full suite. Production code remains unchanged; PR 47 stays blocked pending an explained resolution.

From e40536a7d5790788d3b785e8033b4b5f78658267 Mon Sep 17 00:00:00 2001
From: LucaLin <78164141+LucaLin233@users.noreply.github.com>
Date: Tue, 8 Sep 2026 15:29:51 +0000
Subject: [PATCH 5/5] test(push): remove intrusive function wrappers from batch
 probes

---
 tests/test-push.sh | 27 +--------------------------
 1 file changed, 1 insertion(+), 26 deletions(-)

diff --git a/tests/test-push.sh b/tests/test-push.sh
index 153b478..fbc675b 100644
--- a/tests/test-push.sh
+++ b/tests/test-push.sh
@@ -1390,32 +1390,7 @@ EOF
         printf "DIAG: bash=%s status=%s lifecycle_failed=%s worker_error=%s\n" "$BASH_VERSION" "$1" "$BATCH_WORKER_FAILED" "$BATCH_WORKER_ERROR" >&2
         declare -p ACTIVE_WORKERS ACTIVE_WORKER_STATE_FILES ACTIVE_WORKER_STATE_STARTS REGISTERING_WORKERS WORKER_REGISTRATION_READY_FILES WORKER_REGISTRATION_RELEASE_FILES WORKER_REGISTRATION_STARTS WORKER_REGISTRATION_STAGE_FILES >&2
     }
-    # Test-local wrappers preserve return values and only log failing boundaries.
-    # No argument values, environment dumps, credentials, or global xtrace.
-    for diagnostic_function in launch_worker prune_active_workers wait_for_worker_slot wait_for_all_workers cleanup_active_failed_worker_sessions; do
-        eval "$(declare -f "$diagnostic_function" | sed "1s/$diagnostic_function/diagnostic_original_$diagnostic_function/")"
-    done
-    diagnostic_boundary() {
-        local name=$1 status=0
-        shift
-        "diagnostic_original_$name" "$@" || status=$?
-        if (( status != 0 )); then
-            printf "DIAG: boundary=%s pid=%s status=%s\n" "$name" "$BASHPID" "$status" >&2
-            diagnose_batch_contract "$status"
-        fi
-        return "$status"
-    }
-    launch_worker() { diagnostic_boundary launch_worker "$@"; }
-    prune_active_workers() { diagnostic_boundary prune_active_workers "$@"; }
-    wait_for_worker_slot() { diagnostic_boundary wait_for_worker_slot "$@"; }
-    wait_for_all_workers() { diagnostic_boundary wait_for_all_workers "$@"; }
-    cleanup_active_failed_worker_sessions() { diagnostic_boundary cleanup_active_failed_worker_sessions "$@"; }
-    eval "$(declare -f stop_batch_after_lifecycle_failure | sed "1s/stop_batch_after_lifecycle_failure/diagnostic_original_stop_batch_after_lifecycle_failure/")"
-    stop_batch_after_lifecycle_failure() {
-        printf "DIAG: barrier observed before lifecycle teardown\n" >&2
-        diagnose_batch_contract "$MANAGED_CLEANUP_FAILURE_STATUS"
-        diagnostic_original_stop_batch_after_lifecycle_failure
-    }
+    # Keep production call stacks intact; diagnose only after return.
     assert_transfer_failure_contract() {
         local label=$1 status=$2
         if [[ $status != 1 || $BATCH_WORKER_FAILED != false || $BATCH_WORKER_ERROR != false ]]; then