diff --git a/scripts/check-voidbot-operations.ps1 b/scripts/check-voidbot-operations.ps1
index 44bb620..3666e15 100644
--- a/scripts/check-voidbot-operations.ps1
+++ b/scripts/check-voidbot-operations.ps1
@@ -1050,7 +1050,7 @@ try {
           remoteFile = $offsiteStatus.remoteFile
         }
       } else {
-        Add-Check -Name "offsite.status_file" -Status "failed" -Detail "Latest offsite sync status is '$([string]$offsiteStatus.status)' instead of completed." -Data @{
+        Add-Check -Name "offsite.status_file" -Status "warning" -Detail "Latest offsite sync status is '$([string]$offsiteStatus.status)' instead of completed." -Data @{
           error = [string](Get-OptionalPropertyValue -InputObject $offsiteStatus -PropertyName "error")
         }
       }
diff --git a/scripts/start-voidbot-stack.ps1 b/scripts/start-voidbot-stack.ps1
index f4c9307..84e0e3d 100644
--- a/scripts/start-voidbot-stack.ps1
+++ b/scripts/start-voidbot-stack.ps1
@@ -514,17 +514,39 @@ if ($ragBackend -eq "ollama") {
 if ($localLlmEnabled) {
   $status.stage = "local_llm"
   Write-StatusFile -Path $statusPath -Status $status
-  $status.localLlmOllama = Ensure-OllamaEndpoint -BaseUrl $localLlmBaseUrl -Model $localLlmModel -Label "Local LLM Ollama" -LogDir $logDir
+  try {
+    $status.localLlmOllama = Ensure-OllamaEndpoint -BaseUrl $localLlmBaseUrl -Model $localLlmModel -Label "Local LLM Ollama" -LogDir $logDir
+  } catch {
+    $status.localLlmOllama = @{
+      url = $localLlmBaseUrl
+      model = $localLlmModel
+      healthy = $false
+      degradedStartup = $true
+      error = $_.Exception.Message
+    }
+  }
 
   if ($localLlmSocialReadModel -eq $localLlmModel) {
     $status.localLlmSocialReadOllama = @{
       url = $localLlmBaseUrl
       model = $localLlmSocialReadModel
-      healthy = $true
+      healthy = [bool]$status.localLlmOllama.healthy
       mirrorsLocalLlm = $true
+      degradedStartup = if ($status.localLlmOllama.ContainsKey("degradedStartup")) { $status.localLlmOllama.degradedStartup } else { $false }
+      error = if ($status.localLlmOllama.ContainsKey("error")) { $status.localLlmOllama.error } else { $null }
     }
   } else {
-    $status.localLlmSocialReadOllama = Ensure-OllamaEndpoint -BaseUrl $localLlmBaseUrl -Model $localLlmSocialReadModel -Label "Local LLM social-read Ollama" -LogDir $logDir
+    try {
+      $status.localLlmSocialReadOllama = Ensure-OllamaEndpoint -BaseUrl $localLlmBaseUrl -Model $localLlmSocialReadModel -Label "Local LLM social-read Ollama" -LogDir $logDir
+    } catch {
+      $status.localLlmSocialReadOllama = @{
+        url = $localLlmBaseUrl
+        model = $localLlmSocialReadModel
+        healthy = $false
+        degradedStartup = $true
+        error = $_.Exception.Message
+      }
+    }
   }
 
   Write-StatusFile -Path $statusPath -Status $status
diff --git a/state/evidence.jsonl b/state/evidence.jsonl
index e18cf69..0640a12 100644
--- a/state/evidence.jsonl
+++ b/state/evidence.jsonl
@@ -33,3 +33,4 @@
 {"ts":"2026-05-02T21:52:56Z","type":"implementation","status":"ok","note":"Situational sidecar now emits transcript-wide Ghostlight-style participant reads with durable binary label families plus ephemeral situational state; roster-constrained schema and full configured Ollama timeout are needed to keep multi-person reads viable on a busy remote box."}
 {"ts":"2026-05-02T22:58:31Z","type":"research","status":"ok","note":"Gemma3 12B Q4_0 on the 4060 Ti box handles the transcript-wide social-read task once the schema sticks to a simpler subset; Ollama chokes on roster-constrained prefixItems/const schema but both Gemma and Qwen work with local roster reconciliation, and Gemma slightly beats Qwen on latency for the tested slice."}
 {"ts":"2026-05-02T23:40:40Z","type":"implementation","status":"ok","note":"Split room-read sidecar model selection from the main local reply model so Gemma can drive transcript profiling without changing the public/local answer lane."}
+{"ts":"2026-05-03T07:40:36Z","type":"implementation","status":"ok","note":"Offsite backup was stopping bot and worker, then leaving the stack dead if the remote local-LLM box was asleep; startup now records local-LLM degradation without aborting, and watchdog treats a failed offsite status file as a warning while remote freshness is checked separately."}