diff --git a/drb-c2-core/app/config.py b/drb-c2-core/app/config.py
index d4e0269..bfdd844 100644
--- a/drb-c2-core/app/config.py
+++ b/drb-c2-core/app/config.py
@@ -38,8 +38,14 @@ class Settings(BaseSettings):
     # Correlation consensus models
     # corr_cheap_model  — first-pass LLM correlator (runs on every call)
     # corr_smart_model  — tiebreaker (only fires when rules and cheap LLM disagree)
-    corr_cheap_model: str = "gemini-2.0-flash"
-    corr_smart_model: str = "gemini-1.5-pro"
+    # Both IDs below were retired by Google and returned 404 on every call from
+    # some point before 2026-08-18 until they were corrected. Because a failed
+    # LLM call falls back to the rules decision, nothing broke loudly -- the
+    # entire LLM tier and the consensus tiebreak were simply dead in production
+    # while correlation behaviour was being tuned against rules-only output.
+    # Verify against https://ai.google.dev/gemini-api/docs/models before changing.
+    corr_cheap_model: str = "gemini-3.6-flash"   # was gemini-2.0-flash (shut down)
+    corr_smart_model: str = "gemini-2.5-pro"     # was gemini-1.5-pro (shut down)
     summary_interval_minutes: int = 2          # how often the summary loop runs
     correlation_window_hours: int = 2          # slow/location path: max hours since last call
     embedding_similarity_threshold: float = 0.93  # slow-path: requires location corroboration
diff --git a/drb-c2-core/app/internal/llm_correlator.py b/drb-c2-core/app/internal/llm_correlator.py
index 9dfb6e4..3f2862b 100644
--- a/drb-c2-core/app/internal/llm_correlator.py
+++ b/drb-c2-core/app/internal/llm_correlator.py
@@ -241,10 +241,37 @@ async def decide(call_id: str, ctx: dict) -> Optional[dict]:
         )
         return decision
     except Exception as e:
-        logger.warning(f"LLM correlator failed for call {call_id}: {e}")
+        _log_llm_failure("LLM correlator", call_id, settings.corr_cheap_model, e)
         return None
 
 
+_dead_models: set[str] = set()
+
+
+def _log_llm_failure(where: str, call_id: str, model: str, exc: Exception) -> None:
+    """
+    Log an LLM failure, escalating a dead model ID to ERROR once per model.
+
+    A per-call WARNING was the only signal that gemini-2.0-flash had been shut
+    down, and since every failure falls back to the rules decision the pipeline
+    kept running normally -- the LLM tier was dead for an unknown number of days
+    while correlation was being tuned against rules-only output. A transient API
+    error is genuinely a warning; a model that does not exist is a config bug
+    that will never fix itself, so it gets ERROR and says what to do.
+    """
+    text = str(exc)
+    if "404" in text or "not found" in text.lower() or "no longer available" in text.lower():
+        if model not in _dead_models:
+            _dead_models.add(model)
+            logger.error(
+                f"{where}: model {model!r} is unavailable -- the LLM correlation tier "
+                f"is DISABLED and every call is falling back to rules-only. Update "
+                f"CORR_CHEAP_MODEL/CORR_SMART_MODEL in config.py. Google said: {text}"
+            )
+        return
+    logger.warning(f"{where} failed for call {call_id}: {text}")
+
+
 async def tiebreak(rules_decision: dict, llm_decision: dict, ctx: dict) -> dict:
     """
     Run the smart tiebreaker (corr_smart_model) when rules and LLM disagree.
@@ -263,7 +290,7 @@ async def tiebreak(rules_decision: dict, llm_decision: dict, ctx: dict) -> dict:
         )
         return decision
     except Exception as e:
-        logger.warning(f"LLM tiebreak failed for call {call_id}: {e} — using rules decision")
+        _log_llm_failure("LLM tiebreak", call_id, settings.corr_smart_model, e)
         return rules_decision