fix(agentx): re-enable weka live assistant replay

Author: cquil11

benchmarks/benchmark_lib.sh

@@ -949,13 +949,13 @@ install_agentic_deps() {
 build_replay_cmd() {
     # aiperf invocation for the inferencex-agentx-mvp scenario.
     #
-    # Weka pre-canned assistant mode is the default
-    # (AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES=0): the loader emits
-    # reconstructed assistant segments from the trace and AIPerf discards
-    # the server's live response for future prompt construction. This keeps
-    # reset_context payloads self-contained and preserves hash-id fidelity.
-    # Set AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES=1 explicitly to test
-    # live-response threading.
+    # Live-assistant mode is the default
+    # (AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES=1): the loader emits
+    # user-only deltas and the worker threads the server's live assistant
+    # response back into the session. Set
+    # AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES=0 explicitly to compare
+    # against pre-canned assistant replay, where the server response is
+    # discarded for future prompt construction.
     #
     # The scenario plugin locks: --cache-bust first_turn_prefix and
     # --trace-idle-gap-cap-seconds 60 (per-trace idle-gap compression
@@ -966,7 +966,7 @@ build_replay_cmd() {
     local result_dir="$1"
     local duration="${DURATION:-1800}"
 
-    export AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES="${AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES:-0}"
+    export AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES="${AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES:-1}"
     # Dataset configuration (load + reconstruct + inputs.json + mmap)
     # routinely takes 4-5 min for the 949-trace weka corpus on fast /tmp
     # (B300) but can stretch to 14 min on slower /tmp + parallel contention