diff --git a/run_live_paper.py b/run_live_paper.py
index a020ef4..954f7f0 100644
--- a/run_live_paper.py
+++ b/run_live_paper.py
@@ -923,7 +923,16 @@ def main(argv: list[str] | None = None) -> int:
             from research.ingestion.news_enrichment import catalyst_score as _cscore
             adv = cat_map.get((sym or "").upper())
             if not adv:
-                return 1.0
+                # No LLM advisory for THIS symbol -> fall back to the binary
+                # headline boost instead of dropping news from the ranking.
+                # Returning a flat 1.0 here meant that whenever the enrichment
+                # was down, cat_map was empty for EVERY symbol and news stopped
+                # influencing the rank at all — silently. That is exactly what
+                # happened from 2026-08-03 (Ollama host/model misconfigured)
+                # through 2026-09-13: 40 days of live trading, two days of which
+                # preceded the first real-money fill, with zero news input to the
+                # ranking and no error anywhere to show for it.
+                return news_boost if has_headline else 1.0
             if adv.get("is_dilutive") and float(adv.get("conviction") or 0.0) >= (
                     ollama_cfg.dilution_veto_min_conviction):
                 return dilutive_penalty
@@ -1340,11 +1349,52 @@ def main(argv: list[str] | None = None) -> int:
         except Exception as exc:  # noqa: BLE001
             print(f"[enrich] no DB connection, enrichment off: {exc}", flush=True)
             return
+        # enrich_recent_news NEVER raises — it reports via its counts dict. The
+        # old loop discarded that return value, so a totally dead enrichment
+        # (Ollama unreachable, model absent) looked identical to a healthy idle
+        # one and ran unnoticed for 40 days. Emit a module_tick so the failure is
+        # visible to the monitoring board and queryable in the event store.
+        consecutive_bad = 0
         while not stop_event.wait(float(ollama_cfg.enrichment_interval_seconds)):
             try:
-                enrich_recent_news(econ, ollama_cfg)
+                counts = enrich_recent_news(econ, ollama_cfg) or {}
             except Exception as exc:  # noqa: BLE001
+                counts = {"enriched": 0, "skipped": 0, "errors": -1}
                 print(f"[enrich] error: {exc}", flush=True)
+            errors = int(counts.get("errors") or 0)
+            enriched = int(counts.get("enriched") or 0)
+            # "bad" = the pass tried to classify and every attempt failed.
+            bad = errors != 0 and enriched == 0
+            consecutive_bad = consecutive_bad + 1 if bad else 0
+            if bad:
+                print(f"[enrich] {errors} failures, 0 enriched "
+                      f"(host={ollama_cfg.host} model={ollama_cfg.model})", flush=True)
+            # Report on the first bad pass and then hourly, so a persistent
+            # outage stays visible without flooding the store every interval.
+            period = max(1, int(3600 / max(1.0, float(ollama_cfg.enrichment_interval_seconds))))
+            if enriched or (bad and (consecutive_bad == 1 or consecutive_bad % period == 0)):
+                rt["store"].emit(ModuleTickEvent(
+                    timestamp=datetime.now(),
+                    mode=EventMode.PAPER,
+                    correlation_id=rt["session_id"],
+                    message=(f"news enrichment DOWN: {errors} failures, 0 enriched "
+                             f"({consecutive_bad} passes)" if bad
+                             else f"news enrichment: +{enriched} classified"),
+                    module="enrichment",
+                    stage="classify",
+                    duration_ms=0.0,
+                    input_count=enriched + errors,
+                    output_count=enriched,
+                    metrics={"enriched": enriched, "errors": errors,
+                             "skipped": int(counts.get("skipped") or 0),
+                             "host": ollama_cfg.host, "model": ollama_cfg.model,
+                             "consecutive_failed_passes": consecutive_bad},
+                    errors=([{"error": "news catalyst enrichment produced nothing — "
+                              f"check Ollama at {ollama_cfg.host} and that model "
+                              f"'{ollama_cfg.model}' is pulled. News ranking is "
+                              "running on the binary headline boost until it recovers."}]
+                            if bad else []),
+                ))
 
     scheduler = Scheduler()
     scheduler.add("news", step_news, float(os.environ.get("ALPACA_NEWS_INTERVAL_SECONDS", "120")),