fix: normalize LLM provider key lookup and skip disabled HyDE ping (final review)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

backend/app/api/routes/status.py

@@ -8,7 +8,7 @@ from fastapi import APIRouter
 
 from app.config.settings import settings
 from app.domain.retrieval import RetrievedChunk
-from app.services.llm.llm_factory import get_llm_client
+from app.services.llm.llm_factory import get_llm_client, get_llm_factory
 from app.shared.bootstrap import (
     get_bm25_retriever,
     get_binary_store,
@@ -129,6 +129,27 @@ async def get_health():
     }
 
 
+def _normalize_llm_provider(raw_provider: str) -> str:
+    """Normalize a raw LLM_PROVIDER/HYDE_LLM_PROVIDER settings string to the
+    canonical LLMProvider enum value, the SAME way LLMFactory.create() does.
+
+    TrackedLLMClient.chat() (tracked_client.py) always records usage under
+    `self._inner.config.provider.value` — the NORMALIZED enum value produced by
+    LLMFactory._parse_provider() — never the raw string a caller passed to
+    get_llm_client(). Reusing that same normalization here (instead of
+    duplicating the alias table) guarantees the tracker key this route reads
+    always agrees with the key TrackedLLMClient wrote, even when the raw
+    settings value is a non-canonical alias (e.g. "deepseek-v3") or different
+    casing. Falls back to the raw string, unchanged, if it does not match any
+    known provider/alias, so this passive status endpoint still renders
+    (as "never_called") instead of raising on a misconfigured provider string.
+    """
+    try:
+        return get_llm_factory()._parse_provider(raw_provider).value
+    except ValueError:
+        return raw_provider
+
+
 def _resolve_role_provider_model(role: str) -> tuple[str, str]:
     """Return the (provider, model) pair currently configured for one AI model role.
 
@@ -138,10 +159,10 @@ def _resolve_role_provider_model(role: str) -> tuple[str, str]:
     recorded when HyDE actually ran.
     """
     if role == "main_llm":
-        return settings.llm_provider, settings.llm_model
+        return _normalize_llm_provider(settings.llm_provider), settings.llm_model
     if role == "hyde_llm":
         return (
-            settings.hyde_llm_provider or settings.llm_provider,
+            _normalize_llm_provider(settings.hyde_llm_provider or settings.llm_provider),
             settings.hyde_llm_model or settings.llm_model,
         )
     if role == "embedding":
@@ -204,7 +225,16 @@ async def get_model_statuses():
 
 
 async def _ping_main_or_hyde(role: str) -> None:
-    """Send one minimal chat completion to the LLM configured for `role`."""
+    """Send one minimal chat completion to the LLM configured for `role`.
+
+    Skipped entirely for "hyde_llm" when settings.hyde_enabled is False,
+    mirroring _ping_reranker()'s disabled-skip pattern: when HyDE is turned
+    off (or reuses the main LLM, the default), issuing this ping would just be
+    a redundant duplicate chat call against the same model for no benefit.
+    "main_llm" is always pinged regardless of this check.
+    """
+    if role == "hyde_llm" and not settings.hyde_enabled:
+        return
     provider, model = _resolve_role_provider_model(role)
     try:
         client = get_llm_client(provider=provider, model=model)