feat(net): add retry with backoff to network calls

Sources/App/CareSyncController.swift

@@ -23,6 +23,7 @@ final class CareSyncController: ObservableObject {
     static let base = URL(string: "https://agentpet.thenightwatcher.online")!
 
     private var debounce: Timer?
+    private var failCount = 0
 
     init() {
         linked = UserDefaults.standard.string(forKey: Self.tokenKey) != nil
@@ -127,6 +128,7 @@ final class CareSyncController: ObservableObject {
 
         var request = URLRequest(url: Self.base.appendingPathComponent("api/care/sync"))
         request.httpMethod = "POST"
+        request.timeoutInterval = 15
         request.setValue("application/json", forHTTPHeaderField: "Content-Type")
         request.setValue("Bearer \(token)", forHTTPHeaderField: "Authorization")
         request.httpBody = try? JSONSerialization.data(withJSONObject: ["pets": pets])
@@ -137,16 +139,27 @@ final class CareSyncController: ObservableObject {
             if status == 200 {
                 lastSyncAt = Date()
                 lastError = nil
+                failCount = 0
             } else if status == 401 {
                 // Token revoked from the web side: unlink quietly.
                 disconnect()
             } else {
-                lastError = NSLocalizedString("Sync failed, will retry.", comment: "")
-                scheduleSync(after: 300)
+                retryWithBackoff()
             }
         } catch {
-            lastError = NSLocalizedString("Sync failed, will retry.", comment: "")
-            scheduleSync(after: 300)
+            retryWithBackoff()
         }
     }
+
+    private func retryWithBackoff() {
+        failCount += 1
+        if failCount >= 5 {
+            lastError = NSLocalizedString("Sync failed repeatedly. Re-link to retry.", comment: "")
+            return
+        }
+        let delays: [TimeInterval] = [30, 120, 300, 600]
+        let delay = delays[min(failCount - 1, delays.count - 1)]
+        lastError = NSLocalizedString("Sync failed, will retry.", comment: "")
+        scheduleSync(after: delay)
+    }
 }

Sources/App/NativeUsageProbe.swift

@@ -58,9 +58,20 @@ final class NativeUsageProbe: ObservableObject {
         request.setValue("oauth-2025-04-20", forHTTPHeaderField: "anthropic-beta")
         request.setValue("claude-code/2.1.69", forHTTPHeaderField: "User-Agent")
 
-        guard let (data, response) = try? await URLSession.shared.data(for: request),
-              (response as? HTTPURLResponse)?.statusCode == 200,
-              let json = try? JSONSerialization.jsonObject(with: data) as? [String: Any]
+        var data: Data?
+        for attempt in 0..<2 {
+            guard let (d, response) = try? await URLSession.shared.data(for: request),
+                  let http = response as? HTTPURLResponse else {
+                if attempt < 1 { try? await Task.sleep(nanoseconds: 1_000_000_000); continue }
+                return nil
+            }
+            if http.statusCode == 200 { data = d; break }
+            if http.statusCode >= 500, attempt < 1 {
+                try? await Task.sleep(nanoseconds: 1_000_000_000); continue
+            }
+            return nil
+        }
+        guard let data, let json = try? JSONSerialization.jsonObject(with: data) as? [String: Any]
         else { return nil }
 
         // Windows are {"utilization": <percent used>, "resets_at": …}.
@@ -135,9 +146,21 @@ final class NativeUsageProbe: ObservableObject {
             request.setValue(account, forHTTPHeaderField: "ChatGPT-Account-Id")
         }
 
-        guard let (data, response) = try? await URLSession.shared.data(for: request),
-              let http = response as? HTTPURLResponse, http.statusCode == 200
-        else { return nil }
+        var data: Data?
+        var httpResponse: HTTPURLResponse?
+        for attempt in 0..<2 {
+            guard let (d, response) = try? await URLSession.shared.data(for: request),
+                  let http = response as? HTTPURLResponse else {
+                if attempt < 1 { try? await Task.sleep(nanoseconds: 1_000_000_000); continue }
+                return nil
+            }
+            if http.statusCode == 200 { data = d; httpResponse = http; break }
+            if http.statusCode >= 500, attempt < 1 {
+                try? await Task.sleep(nanoseconds: 1_000_000_000); continue
+            }
+            return nil
+        }
+        guard let data, let http = httpResponse else { return nil }
 
         // used-percent comes in response headers; the body's rate_limit
         // windows carry the same plus reset timing.

Sources/App/PetBrowser.swift

@@ -94,16 +94,27 @@ final class PetBrowser: ObservableObject {
         }
     }
 
-    /// Loads one manifest; returns `[]` on any failure so the other source still shows.
+    /// Loads one manifest; retries once on transient failures. Returns `[]` on
+    /// permanent failure so the other source still shows.
     private static func fetch(_ url: URL, isCommunity: Bool) async -> [RemotePet] {
-        do {
-            let (data, _) = try await URLSession.shared.data(from: url)
-            var list = try JSONDecoder().decode(Manifest.self, from: data).pets
-            if isCommunity { for i in list.indices { list[i].isCommunity = true } }
-            return list
-        } catch {
-            return []
+        for attempt in 0..<2 {
+            do {
+                var request = URLRequest(url: url)
+                request.timeoutInterval = 10
+                let (data, response) = try await URLSession.shared.data(for: request)
+                let code = (response as? HTTPURLResponse)?.statusCode ?? 200
+                if (200..<300).contains(code) {
+                    var list = try JSONDecoder().decode(Manifest.self, from: data).pets
+                    if isCommunity { for i in list.indices { list[i].isCommunity = true } }
+                    return list
+                }
+                guard code == 429 || code >= 500, attempt < 1 else { return [] }
+            } catch {
+                guard attempt < 1 else { return [] }
+            }
+            try? await Task.sleep(nanoseconds: 1_000_000_000)
         }
+        return []
     }
 
     /// Keeps the first occurrence of each slug (community entries come first).