Fix transient health checks and avoid unnecessary recovery

src/routes/api/servers/[name]/probe-rcon/+server.ts

@@ -2,6 +2,8 @@ import { json } from '@sveltejs/kit';
 import type { RequestHandler } from './$types';
 import { rconStatusResult } from '$lib/server/integrations/rcon.js';
 import { getServer, setServerField } from '$lib/server/fleet/servers.js';
+import { latestGameSubmission } from '$lib/server/fleet/status-evidence.js';
+import { isProvisioning } from '$lib/server/fleet/provisioning.js';
 
 /**
  * On-demand RCON probe for a specific server. Same code path as the periodic
@@ -18,7 +20,14 @@ export const POST: RequestHandler = async ({ params }) => {
     const row = getServer(name);
     if (!row) return json({ ok: false, error: `unknown server: ${name}` }, { status: 404 });
 
+    const epoch = latestGameSubmission(name);
     const result = await rconStatusResult(name);
+    // Keep the diagnostic response, but never persist old-process evidence
+    // across a lifecycle request or replacement registration.
+    const current = getServer(name);
+    if (!current || current.created_at !== row.created_at || epoch !== latestGameSubmission(name) || isProvisioning(name)) {
+        return json(result);
+    }
     const now = new Date().toISOString();
 
     if (result.ok) {