Say that the rsync count file is a cache the live-PID scan overwrites, so a stale value stops reading as a leak

common.sh

@@ -1807,6 +1807,11 @@ if [[ -n "${STATE_DIR:-}" ]] && [[ ! -d "$STATE_DIR" ]]; then
     mkdir -p "$STATE_DIR" 2>/dev/null || true
 fi
 
+# A cache, not a ledger. Nothing decrements it — acquire_rsync_lock() recounts live PIDs from the
+# rsync_*.lock files and overwrites this with the truth before enforcing the limit, so the file
+# routinely sits at a stale non-zero value with no rsync running. That is expected, not a leak:
+# killed transfers cannot wedge the limit, because a lock whose PID is dead is not counted.
+# Read it as "what the last acquire saw", never as "how many are running now".
 RSYNC_COUNT_FILE="$LOCK_DIR/rsync_active_count"
 RSYNC_MAX_CONCURRENT=3
 LOCK_WARN_AGE=300       # seconds — warn if lock older than this (5min default)