diff --git a/Plugin/unraid/include/watchdog.php b/Plugin/unraid/include/watchdog.php
index bbafb05..724fe3b 100644
--- a/Plugin/unraid/include/watchdog.php
+++ b/Plugin/unraid/include/watchdog.php
@@ -222,7 +222,20 @@ function vv_wd_local_system(): array {
     $loadRaw  = @file_get_contents('/proc/loadavg') ?: '0';
     $daemonOk = (trim(shell_exec('docker info >/dev/null 2>&1; echo $?') ?: '1') === '0');
     $oomCount = (int)trim(@file_get_contents(STATE_DIR . '/system_watchdog_oom.db') ?: '0');
-    $cores    = (int)($sys['cpu_cores'] ?: (int)(trim(shell_exec('nproc 2>/dev/null') ?: '1')));
+    // nproc first, and cpu_cores only as a fallback — the reverse of what this used to do.
+    //
+    // The page draws its load bar and colours against this number multiplied by the same
+    // RW_LOAD_* multipliers resource_watchdog.sh uses, and that script computes its thresholds
+    // from `nproc` (resource_watchdog.sh:194). cpu_cores is the API's physical core count: 16 here
+    // against 32 threads. Preferring it made the page turn amber at load 32 and red at 48 while
+    // the watchdog did not reach soft pressure until 64 and medium until 96 — the page reporting
+    // a crisis about a subsystem that considered the machine idle, which is the one thing a page
+    // whose entire job is showing what the watchdogs think must not do.
+    //
+    // Whether load should be judged against threads or cores is a real argument; it is not this
+    // file's argument to have. The number here has to be the number the actor used, or the page
+    // is describing a different machine.
+    $cores    = (int)(trim(shell_exec('nproc 2>/dev/null') ?: '') ?: ($sys['cpu_cores'] ?: 1));
     return [
         'mem_total'  => (int)($mem['total_kb'] * 1024),
         'mem_avail'  => (int)($mem['free_kb']  * 1024),