diff --git a/aborist/cli.py b/aborist/cli.py
index c97fb9c..2d6b142 100644
--- a/aborist/cli.py
+++ b/aborist/cli.py
@@ -794,6 +794,32 @@ def _render_query_human(result: dict, question: str) -> str:
             f"answer {result.get('answer_chars', 0):,} chars"
         )
 
+    # Per-phase timings — surfaces where the per-call cost lands.
+    # Hermes-bound queries should show `llm_ms` dominating; if
+    # search_ms or persist_ms creeps up that's a retrieval / WAL
+    # signal the operator wants visible. Cache-hit rows skip llm
+    # entirely so the breakdown also tells you which path you're
+    # paying for.
+    timings = result.get("timings") or {}
+    if timings:
+        parts: list[str] = []
+        order = (
+            ("cache_lookup_ms", "cache"),
+            ("search_ms", "search"),
+            ("context_ms", "context"),
+            ("llm_ms", "llm"),
+            ("persist_ms", "persist"),
+        )
+        for key, label in order:
+            v = timings.get(key)
+            if isinstance(v, (int, float)) and v > 0:
+                parts.append(f"{label} {v / 1000:.2f}s")
+        total = timings.get("total_ms")
+        if isinstance(total, (int, float)):
+            parts.append(f"**total {total / 1000:.2f}s**")
+        if parts:
+            lines.append("timings: " + " · ".join(parts))
+
     # Reference-frame notes (Ticket #000002). When detect_frame
     # classified the query as `reference`, surface the named work
     # so an operator knows the substrate routed to a fictional