Fix: Adjust LLM token budget and navigation flow

Increase `max_completion_tokens` for `gpt_oss_120b` from 8192 to 16384
to prevent incomplete responses.

Navigate to the detail screen before the recording screen when starting
a new case, ensuring correct back navigation.

clients/wearos/app/src/main/java/com/doctate/watch/presentation/AppNav.kt

@@ -42,6 +42,9 @@ fun AppNav(pendingCommand: StateFlow<NavCommand.Pending>) {
             }
             NavCommand.NewCase -> {
                 val id = app.caseStore.createLocal()
+                // Push DETAIL under RECORDING so popBackStack on stop lands
+                // on the case page, ready for another recording.
+                navController.navigate(Routes.detail(id))
                 navController.navigate(Routes.recording(id))
             }
             is NavCommand.ContinueCase -> {
@@ -64,6 +67,10 @@ fun AppNav(pendingCommand: StateFlow<NavCommand.Pending>) {
                         onNewCase = {
                             coroutineScope.launch {
                                 val id = app.caseStore.createLocal()
+                                // Same backstack shape as NavCommand.NewCase:
+                                // [LIST, DETAIL, RECORDING] so stop returns to
+                                // the case page rather than the list.
+                                navController.navigate(Routes.detail(id))
                                 navController.navigate(Routes.recording(id))
                             }
                         },

clients/wearos/run.sh

@@ -384,8 +384,12 @@ start_emulator_and_wait() {
     local emulator_bin="$1"
     local avd="$2"
 
-    info "Starting AVD '$avd' (background)..."
-    nohup "$emulator_bin" -avd "$avd" >/dev/null 2>&1 &
+    # -no-snapshot-load forces a cold boot. Quick Boot's RAM snapshot freezes
+    # the wallclock from the last session, and Wear OS in-AVD NTP often does
+    # not catch up — case timestamps end up days in the past. Cold boot
+    # picks up the host RTC at startup, no in-app correction needed.
+    info "Starting AVD '$avd' (cold boot, background)..."
+    nohup "$emulator_bin" -avd "$avd" -no-snapshot-load >/dev/null 2>&1 &
     disown
 
     info "Waiting for ADB to see the device..."

server/src/analyze/backend.rs

@@ -157,14 +157,15 @@ pub fn backends() -> &'static [LlmBackend] {
             // gpt-oss-120b: schema OFF on purpose. Reasoning tokens count
             // against `max_completion_tokens`; with `high` reasoning the
             // CoT alone can eat several thousand tokens, so the budget is
-            // doubled to 8192 to leave room for the actual answer. Without
-            // headroom the model returns `choices[0].message.content == null`
+            // raised to 4x the default to leave room for the actual answer.
+            // Without headroom the model returns
+            // `choices[0].message.content == null` (finish_reason="length")
             // — the LlmError::Parse path. Pre-multi-backend (commit bf6464d)
             // gpt-oss ran without schema and worked fine; this restores
             // that body byte-for-byte except for the larger budget.
             ionos_backend("gpt_oss_120b", "GPT OSS 120b", "openai/gpt-oss-120b")
                 .with_reasoning_effort("high")
-                .with_max_completion_tokens(8192)
+                .with_max_completion_tokens(16384)
                 .with_use_json_schema(false),
             ionos_backend(
                 "llama_3_1_405b",