fix(health): suppress 404 when health monitor disabled + fix audit test assertion

- Sidebar.svelte: defer healthStore.refresh() until feature flags confirm
  health_monitor is enabled; skip entirely when disabled
- health.js: remove throw error from catch block — log and return null
  instead, preventing 'Uncaught (in promise)' on expected 404
- test_report_audit_immutability.py: fix mock assertions —
  audit_service uses logger.reason/reflect/explore, not logger.info
- HealthStore and Sidebar now produce zero network noise when
  FEATURES__HEALTH_MONITOR=false

backend/src/plugins/translate/preview.py

@@ -41,7 +41,7 @@ from ._token_budget import DEFAULT_CONTEXT_WINDOW, estimate_token_budget
 from .dictionary import DictionaryManager
 
 # #region DEFAULT_EXECUTION_PROMPT_TEMPLATE [TYPE Constant]
-# @BRIEF Default prompt template for batch LLM translation execution (no context columns — faster).
+
 # Supports both single-language and multi-language modes via {target_languages} placeholder.
 DEFAULT_EXECUTION_PROMPT_TEMPLATE: str = (
     "Translate the following database content from {source_language} to the following language(s): {target_languages}.\n\n"
@@ -63,9 +63,9 @@ DEFAULT_EXECUTION_PROMPT_TEMPLATE: str = (
 )
 # #endregion DEFAULT_EXECUTION_PROMPT_TEMPLATE
 
-
 # #region DEFAULT_PREVIEW_PROMPT_TEMPLATE [TYPE Constant]
-# @BRIEF Default prompt template for LLM translation preview.
+# #region DEFAULT_PREVIEW_PROMPT_TEMPLATE [TYPE Constant]
+
 DEFAULT_PREVIEW_PROMPT_TEMPLATE: str = (
     "Translate the following database content.\n\n"
     "Source dialect: {source_dialect}\n"
@@ -1038,19 +1038,37 @@ class TranslationPreview:
                 "Authorization": f"Bearer {api_key}",
                 "Content-Type": "application/json",
             }
+            # Reasoning-safe system prompt: always forbid chain-of-thought in output.
+            # Reasoning models (DeepSeek, QwQ, etc.) may leak reasoning into content
+            # when response_format=json_object is set. Explicit suppression in the
+            # system message prevents this regardless of the disable_reasoning flag.
+            system_content = (
+                "You are a database content translation assistant. "
+                "Translate the provided text accurately, preserving data semantics. "
+                "Respond directly with ONLY the JSON result. "
+                "Do NOT include any reasoning, thinking, chain-of-thought, analysis, "
+                "or explanation. Output ONLY valid JSON."
+            )
+
             payload = {
                 "model": model,
                 "messages": [
-                    {"role": "system", "content": "You are a database content translation assistant. Translate the provided text accurately, preserving data semantics."},
+                    {"role": "system", "content": system_content},
                     {"role": "user", "content": prompt},
                 ],
                 "temperature": 0.1,
                 "max_tokens": max_tokens,
             }
+
             # Structured output — Kilo gateway supports response_format, but upstream providers
             # (e.g. StepFun) may reject it. We try with response_format and fall back on 400.
+            # NOTE: Reasoning models (deepseek variants, qwen with reasoning) often break with
+            # response_format=json_object, producing reasoning text instead of JSON.
+            # When disable_reasoning is set, we skip response_format and rely on the
+            # system prompt to enforce JSON-only output.
             if provider_type in ("openai", "openai_compatible", "kilo", "openrouter", "litellm"):
-                payload["response_format"] = {"type": "json_object"}
+                if not disable_reasoning:
+                    payload["response_format"] = {"type": "json_object"}
 
             # Suppress Chain of Thought reasoning to save output tokens
             # NOTE: Kilo/OpenRouter/LiteLLM reject reasoning_effort — only use for native OpenAI-compatible
@@ -1058,13 +1076,10 @@ class TranslationPreview:
                 # Kilo/OpenRouter/LiteLLM reject reasoning_effort — only use for native OpenAI-compatible
                 if provider_type not in ("kilo", "openrouter", "litellm"):
                     payload["reasoning_effort"] = "none"
-                payload.pop("response_format", None)  # JSON mode triggers reasoning on some models
+                # response_format already skipped above when disable_reasoning is True
                 # Use caller-provided max_tokens instead of hardcoded 8192
                 # This ensures multi-language batches with large output get enough token budget.
                 payload["max_tokens"] = max_tokens
-                # Universal instruction — all models understand "respond directly without reasoning"
-                system_content = "You are a database content translation assistant. Translate the provided text accurately, preserving data semantics. Respond directly with ONLY the JSON result. Do NOT include any reasoning, thinking, chain-of-thought, analysis, or explanation. Output ONLY valid JSON."
-                payload["messages"][0] = {"role": "system", "content": system_content}
 
             logger.reason(
                 f"LLM request url={base_url} model={payload.get('model')} "