diff --git a/collectivex/configs/platform_config.json b/collectivex/configs/platform_config.json
index 535cd8fe2f..3a0cd96e84 100644
--- a/collectivex/configs/platform_config.json
+++ b/collectivex/configs/platform_config.json
@@ -133,17 +133,23 @@
     "mi300x": {
       "arch": "gfx942",
       "product": "mi300x",
-      "image": "rocm/sgl-dev:sglang-0.5.14-rocm720-mi35x-mori-0701",
+      "image": "vllm/vllm-openai-rocm:v0.32.0",
       "image_platform": "linux/amd64",
       "gpus_per_node": 8,
       "scale_up_domain": 8,
       "scale_up_transport": "xgmi",
       "launcher": "mi-amds",
-      "backends": {"mori": [8], "uccl-ep": [8]},
-      "ll_backends": {},
+      "backends": {"mori": [8, 16], "uccl-ep": [8, 16]},
+      "ll_backends": {"mori": [16], "uccl-ep": [16]},
       "fabric": {
-        "nic": "n/a (single-node scale-up)",
-        "switch": "n/a (single-node scale-up)"
+        "nic": "Broadcom 400GbE RoCE v2 (bnxt_re), 8 rails",
+        "switch": "routed L3 RoCE fabric"
+      },
+      "network": {
+        "socket_ifname": "ens50f1np1",
+        "rdma_devices": "bnxt_re_bond0,bnxt_re_bond1,bnxt_re_bond2,bnxt_re_bond3,bnxt_re_bond4,bnxt_re_bond5,bnxt_re_bond6,bnxt_re_bond7",
+        "ib_gid_index": 3,
+        "host_verbs_provider": "/usr/local/lib/x86_64-linux-gnu/libbnxt_re-rdmav34.so"
       },
       "operator": {
         "partition": "MI300X-UBUNTU",
@@ -153,18 +159,24 @@
     "mi325x": {
       "arch": "gfx942",
       "product": "mi325x",
-      "image": "rocm/sgl-dev:sglang-0.5.14-rocm720-mi35x-mori-0701",
+      "image": "vllm/vllm-openai-rocm:v0.32.0",
       "image_platform": "linux/amd64",
       "gpus_per_node": 8,
       "scale_up_domain": 8,
       "scale_up_transport": "xgmi",
       "launcher": "mi-amds",
       "runner_label": "cluster:mi325x-amds",
-      "backends": {"mori": [8], "uccl-ep": [8]},
-      "ll_backends": {},
+      "backends": {"mori": [8, 16], "uccl-ep": [8, 16]},
+      "ll_backends": {"mori": [16], "uccl-ep": [16]},
       "fabric": {
-        "nic": "n/a (single-node scale-up)",
-        "switch": "n/a (single-node scale-up)"
+        "nic": "Broadcom 400GbE RoCE v2 (bnxt_re), 8 rails",
+        "switch": "routed L3 RoCE fabric"
+      },
+      "network": {
+        "socket_ifname": "enp193s0f0np0",
+        "rdma_devices": "bnxt_re4,bnxt_re5,bnxt_re1,bnxt_re0,bnxt_re10,bnxt_re11,bnxt_re7,bnxt_re6",
+        "ib_gid_index": 3,
+        "host_verbs_provider": "/usr/local/lib/x86_64-linux-gnu/libbnxt_re-rdmav34.so"
       },
       "operator": {
         "partition": "MI325X-UBUNTU",
diff --git a/collectivex/launchers/launch_mi-amds.sh b/collectivex/launchers/launch_mi-amds.sh
index c82900ccf9..86ad978b19 100644
--- a/collectivex/launchers/launch_mi-amds.sh
+++ b/collectivex/launchers/launch_mi-amds.sh
@@ -103,6 +103,7 @@ for allocation_attempt in 1 2 3; do
 done
 unset COLLX_SALLOC_ATTEMPT COLLX_NETWORK_VALIDATION_ATTEMPT
 CONTAINER_MOUNTS="$MOUNT_SRC:$MOUNT_DIR$DEVICE_MOUNTS"
+collx_add_host_verbs_provider_mount
 # uccl-ep builds from source, so it gets the cross-allocation backend cache single-slurm uses.
 # Its parent is COLLX_STAGE_DIR, not SQUASH_DIR: this SKU's squash dir is node-local and
 # root-owned (/var/lib/squash), so the submit-side mkdir fails and the path is not what compute
diff --git a/collectivex/launchers/launch_single-slurm.sh b/collectivex/launchers/launch_single-slurm.sh
index 2da866f5d9..3188fc6aa0 100644
--- a/collectivex/launchers/launch_single-slurm.sh
+++ b/collectivex/launchers/launch_single-slurm.sh
@@ -67,6 +67,7 @@ collx_select_image "$IMAGE"
 collx_stage_with_backend_cache
 # Multi-node shards on the EFA pools (b300, h100-dgxc) carry the host EFA user space themselves.
 collx_add_efa_host_mounts
+collx_add_host_verbs_provider_mount
 
 command -v salloc >/dev/null || collx_die "salloc not found on this runner"
 allocation=(--partition="$COLLX_PARTITION" --nodes="$NODES" --gres=gpu:"$GPN"
@@ -81,7 +82,13 @@ for allocation_attempt in 1 2 3; do
   [ -z "$excluded_nodes" ] || attempt_allocation+=(--exclude="$excluded_nodes")
   export COLLX_SALLOC_ATTEMPT="$allocation_attempt"
   export COLLX_NETWORK_VALIDATION_ATTEMPT="$allocation_attempt"
-  collx_salloc_jobid "${attempt_allocation[@]}"
+  JOB_ID=""
+  if ! collx_salloc_jobid "${attempt_allocation[@]}"; then
+    # A job that dies before its grant (a failed node prolog) is worth another allocation.
+    [ "$allocation_attempt" != 3 ] || collx_die "scheduler allocation failed on every attempt"
+    collx_log "allocation attempt $allocation_attempt failed; retrying"
+    continue
+  fi
   [ -n "$JOB_ID" ] || collx_die "could not resolve allocated JOB_ID from salloc"
   if ! collx_validate_network_profile_on_job "$JOB_ID" "$NODES" "$COLLX_TRANSPORT"; then
     validation_failure=network
diff --git a/collectivex/results/cases.json b/collectivex/results/cases.json
index d3f2d0fbcb..b8efe5a46c 100644
--- a/collectivex/results/cases.json
+++ b/collectivex/results/cases.json
@@ -5544,28 +5544,636 @@
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
-   "disposition": "unsupported",
-   "reason": "backend-platform-unsupported",
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi300x"
+  },
+  {
+   "case": {
+    "backend": "uccl-ep",
+    "case_id": "mi300x-uccl-ep-deepseek-v4-pro-normal-decode-ep16-uniform-fp8",
+    "ep": 16,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "normal",
+    "nodes": 2,
+    "phase": "decode",
+    "precision": "fp8",
+    "routing": "uniform",
+    "scale_out_transport": "rdma",
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-out",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi300x-xgmi-rdma",
+    "transport": "xgmi-rdma",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi300x"
+  },
+  {
+   "case": {
+    "backend": "uccl-ep",
+    "case_id": "mi300x-uccl-ep-deepseek-v4-pro-low-latency-decode-ep16-uniform-bf16",
+    "ep": 16,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "low-latency",
+    "nodes": 2,
+    "phase": "decode",
+    "precision": "bf16",
+    "routing": "uniform",
+    "scale_out_transport": "rdma",
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-out",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi300x-xgmi-rdma",
+    "transport": "xgmi-rdma",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi300x"
+  },
+  {
+   "case": {
+    "backend": "uccl-ep",
+    "case_id": "mi300x-uccl-ep-deepseek-v4-pro-low-latency-decode-ep16-uniform-fp8",
+    "ep": 16,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "low-latency",
+    "nodes": 2,
+    "phase": "decode",
+    "precision": "fp8",
+    "routing": "uniform",
+    "scale_out_transport": "rdma",
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-out",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi300x-xgmi-rdma",
+    "transport": "xgmi-rdma",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi300x"
+  },
+  {
+   "case": {
+    "backend": "mori",
+    "case_id": "mi300x-mori-deepseek-v4-pro-normal-decode-ep16-uniform-bf16",
+    "ep": 16,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "normal",
+    "nodes": 2,
+    "phase": "decode",
+    "precision": "bf16",
+    "routing": "uniform",
+    "scale_out_transport": "rdma",
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-out",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi300x-xgmi-rdma",
+    "transport": "xgmi-rdma",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi300x"
+  },
+  {
+   "case": {
+    "backend": "mori",
+    "case_id": "mi300x-mori-deepseek-v4-pro-normal-decode-ep16-uniform-fp8",
+    "ep": 16,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "normal",
+    "nodes": 2,
+    "phase": "decode",
+    "precision": "fp8",
+    "routing": "uniform",
+    "scale_out_transport": "rdma",
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-out",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi300x-xgmi-rdma",
+    "transport": "xgmi-rdma",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi300x"
+  },
+  {
+   "case": {
+    "backend": "mori",
+    "case_id": "mi300x-mori-deepseek-v4-pro-low-latency-decode-ep16-uniform-bf16",
+    "ep": 16,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "low-latency",
+    "nodes": 2,
+    "phase": "decode",
+    "precision": "bf16",
+    "routing": "uniform",
+    "scale_out_transport": "rdma",
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-out",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi300x-xgmi-rdma",
+    "transport": "xgmi-rdma",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi300x"
+  },
+  {
+   "case": {
+    "backend": "mori",
+    "case_id": "mi300x-mori-deepseek-v4-pro-low-latency-decode-ep16-uniform-fp8",
+    "ep": 16,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "low-latency",
+    "nodes": 2,
+    "phase": "decode",
+    "precision": "fp8",
+    "routing": "uniform",
+    "scale_out_transport": "rdma",
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-out",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi300x-xgmi-rdma",
+    "transport": "xgmi-rdma",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi300x"
+  },
+  {
+   "case": {
+    "backend": "uccl-ep",
+    "case_id": "mi300x-uccl-ep-deepseek-v4-pro-normal-prefill-ep16-uniform-bf16",
+    "ep": 16,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1024 2048 4096 8192",
+    "mode": "normal",
+    "nodes": 2,
+    "phase": "prefill",
+    "precision": "bf16",
+    "routing": "uniform",
+    "scale_out_transport": "rdma",
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-out",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi300x-xgmi-rdma",
+    "transport": "xgmi-rdma",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi300x"
+  },
+  {
+   "case": {
+    "backend": "uccl-ep",
+    "case_id": "mi300x-uccl-ep-deepseek-v4-pro-normal-prefill-ep16-uniform-fp8",
+    "ep": 16,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1024 2048 4096 8192",
+    "mode": "normal",
+    "nodes": 2,
+    "phase": "prefill",
+    "precision": "fp8",
+    "routing": "uniform",
+    "scale_out_transport": "rdma",
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-out",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi300x-xgmi-rdma",
+    "transport": "xgmi-rdma",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi300x"
+  },
+  {
+   "case": {
+    "backend": "mori",
+    "case_id": "mi300x-mori-deepseek-v4-pro-normal-prefill-ep16-uniform-bf16",
+    "ep": 16,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1024 2048 4096 8192",
+    "mode": "normal",
+    "nodes": 2,
+    "phase": "prefill",
+    "precision": "bf16",
+    "routing": "uniform",
+    "scale_out_transport": "rdma",
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-out",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi300x-xgmi-rdma",
+    "transport": "xgmi-rdma",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi300x"
+  },
+  {
+   "case": {
+    "backend": "mori",
+    "case_id": "mi300x-mori-deepseek-v4-pro-normal-prefill-ep16-uniform-fp8",
+    "ep": 16,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1024 2048 4096 8192",
+    "mode": "normal",
+    "nodes": 2,
+    "phase": "prefill",
+    "precision": "fp8",
+    "routing": "uniform",
+    "scale_out_transport": "rdma",
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-out",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi300x-xgmi-rdma",
+    "transport": "xgmi-rdma",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
    "sku": "mi300x"
   },
   {
    "case": {
-    "backend": "mori",
-    "case_id": "mi300x-mori-deepseek-v4-pro-normal-decode-ep16-uniform-bf16",
-    "ep": 16,
+    "backend": "uccl-ep",
+    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-decode-ep8-uniform-bf16",
+    "ep": 8,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "normal",
+    "nodes": 1,
+    "phase": "decode",
+    "precision": "bf16",
+    "routing": "uniform",
+    "scale_out_transport": null,
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-up",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi325x-xgmi",
+    "transport": "xgmi",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi325x"
+  },
+  {
+   "case": {
+    "backend": "uccl-ep",
+    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-decode-ep8-uniform-fp8",
+    "ep": 8,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "normal",
+    "nodes": 1,
+    "phase": "decode",
+    "precision": "fp8",
+    "routing": "uniform",
+    "scale_out_transport": null,
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-up",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi325x-xgmi",
+    "transport": "xgmi",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi325x"
+  },
+  {
+   "case": {
+    "backend": "mori",
+    "case_id": "mi325x-mori-deepseek-v4-pro-normal-decode-ep8-uniform-bf16",
+    "ep": 8,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "normal",
+    "nodes": 1,
+    "phase": "decode",
+    "precision": "bf16",
+    "routing": "uniform",
+    "scale_out_transport": null,
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-up",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi325x-xgmi",
+    "transport": "xgmi",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi325x"
+  },
+  {
+   "case": {
+    "backend": "mori",
+    "case_id": "mi325x-mori-deepseek-v4-pro-normal-decode-ep8-uniform-fp8",
+    "ep": 8,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "normal",
+    "nodes": 1,
+    "phase": "decode",
+    "precision": "fp8",
+    "routing": "uniform",
+    "scale_out_transport": null,
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-up",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi325x-xgmi",
+    "transport": "xgmi",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi325x"
+  },
+  {
+   "case": {
+    "backend": "uccl-ep",
+    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-prefill-ep8-uniform-bf16",
+    "ep": 8,
+    "experts": 384,
+    "gpus_per_node": 8,
+    "hidden": 7168,
+    "ladder": "1024 2048 4096 8192",
+    "mode": "normal",
+    "nodes": 1,
+    "phase": "prefill",
+    "precision": "bf16",
+    "routing": "uniform",
+    "scale_out_transport": null,
+    "scale_up_domain": 8,
+    "scale_up_transport": "xgmi",
+    "scope": "scale-up",
+    "seed": 67,
+    "suite": "ep-core",
+    "timing": {
+     "chain_drop": 16,
+     "chain_iters_per_trial": 128,
+     "chain_trials_per_point": 4,
+     "iters_per_trial": 8,
+     "trials_per_point": 256,
+     "warmup_iters_per_trial": 32
+    },
+    "topk": 6,
+    "topology_class": "mi325x-xgmi",
+    "transport": "xgmi",
+    "workload": "deepseek-v4-pro"
+   },
+   "detail": null,
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi325x"
+  },
+  {
+   "case": {
+    "backend": "uccl-ep",
+    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-prefill-ep8-uniform-fp8",
+    "ep": 8,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
-    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "ladder": "1024 2048 4096 8192",
     "mode": "normal",
-    "nodes": 2,
-    "phase": "decode",
-    "precision": "bf16",
+    "nodes": 1,
+    "phase": "prefill",
+    "precision": "fp8",
     "routing": "uniform",
-    "scale_out_transport": "rdma",
+    "scale_out_transport": null,
     "scale_up_domain": 8,
     "scale_up_transport": "xgmi",
-    "scope": "scale-out",
+    "scope": "scale-up",
     "seed": 67,
     "suite": "ep-core",
     "timing": {
@@ -5577,33 +6185,33 @@
      "warmup_iters_per_trial": 32
     },
     "topk": 6,
-    "topology_class": "mi300x-xgmi-rdma",
-    "transport": "xgmi-rdma",
+    "topology_class": "mi325x-xgmi",
+    "transport": "xgmi",
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
-   "disposition": "unsupported",
-   "reason": "backend-platform-unsupported",
-   "sku": "mi300x"
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi325x"
   },
   {
    "case": {
-    "backend": "uccl-ep",
-    "case_id": "mi300x-uccl-ep-deepseek-v4-pro-normal-prefill-ep16-uniform-bf16",
-    "ep": 16,
+    "backend": "mori",
+    "case_id": "mi325x-mori-deepseek-v4-pro-normal-prefill-ep8-uniform-bf16",
+    "ep": 8,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
     "ladder": "1024 2048 4096 8192",
     "mode": "normal",
-    "nodes": 2,
+    "nodes": 1,
     "phase": "prefill",
     "precision": "bf16",
     "routing": "uniform",
-    "scale_out_transport": "rdma",
+    "scale_out_transport": null,
     "scale_up_domain": 8,
     "scale_up_transport": "xgmi",
-    "scope": "scale-out",
+    "scope": "scale-up",
     "seed": 67,
     "suite": "ep-core",
     "timing": {
@@ -5615,33 +6223,33 @@
      "warmup_iters_per_trial": 32
     },
     "topk": 6,
-    "topology_class": "mi300x-xgmi-rdma",
-    "transport": "xgmi-rdma",
+    "topology_class": "mi325x-xgmi",
+    "transport": "xgmi",
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
-   "disposition": "unsupported",
-   "reason": "backend-platform-unsupported",
-   "sku": "mi300x"
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi325x"
   },
   {
    "case": {
     "backend": "mori",
-    "case_id": "mi300x-mori-deepseek-v4-pro-normal-prefill-ep16-uniform-bf16",
-    "ep": 16,
+    "case_id": "mi325x-mori-deepseek-v4-pro-normal-prefill-ep8-uniform-fp8",
+    "ep": 8,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
     "ladder": "1024 2048 4096 8192",
     "mode": "normal",
-    "nodes": 2,
+    "nodes": 1,
     "phase": "prefill",
-    "precision": "bf16",
+    "precision": "fp8",
     "routing": "uniform",
-    "scale_out_transport": "rdma",
+    "scale_out_transport": null,
     "scale_up_domain": 8,
     "scale_up_transport": "xgmi",
-    "scope": "scale-out",
+    "scope": "scale-up",
     "seed": 67,
     "suite": "ep-core",
     "timing": {
@@ -5653,33 +6261,33 @@
      "warmup_iters_per_trial": 32
     },
     "topk": 6,
-    "topology_class": "mi300x-xgmi-rdma",
-    "transport": "xgmi-rdma",
+    "topology_class": "mi325x-xgmi",
+    "transport": "xgmi",
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
-   "disposition": "unsupported",
-   "reason": "backend-platform-unsupported",
-   "sku": "mi300x"
+   "disposition": "runnable",
+   "reason": null,
+   "sku": "mi325x"
   },
   {
    "case": {
     "backend": "uccl-ep",
-    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-decode-ep8-uniform-bf16",
-    "ep": 8,
+    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-decode-ep16-uniform-bf16",
+    "ep": 16,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
     "ladder": "1 2 4 8 16 32 64 128 256 512",
     "mode": "normal",
-    "nodes": 1,
+    "nodes": 2,
     "phase": "decode",
     "precision": "bf16",
     "routing": "uniform",
-    "scale_out_transport": null,
+    "scale_out_transport": "rdma",
     "scale_up_domain": 8,
     "scale_up_transport": "xgmi",
-    "scope": "scale-up",
+    "scope": "scale-out",
     "seed": 67,
     "suite": "ep-core",
     "timing": {
@@ -5691,8 +6299,8 @@
      "warmup_iters_per_trial": 32
     },
     "topk": 6,
-    "topology_class": "mi325x-xgmi",
-    "transport": "xgmi",
+    "topology_class": "mi325x-xgmi-rdma",
+    "transport": "xgmi-rdma",
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
@@ -5703,21 +6311,21 @@
   {
    "case": {
     "backend": "uccl-ep",
-    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-decode-ep8-uniform-fp8",
-    "ep": 8,
+    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-decode-ep16-uniform-fp8",
+    "ep": 16,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
     "ladder": "1 2 4 8 16 32 64 128 256 512",
     "mode": "normal",
-    "nodes": 1,
+    "nodes": 2,
     "phase": "decode",
     "precision": "fp8",
     "routing": "uniform",
-    "scale_out_transport": null,
+    "scale_out_transport": "rdma",
     "scale_up_domain": 8,
     "scale_up_transport": "xgmi",
-    "scope": "scale-up",
+    "scope": "scale-out",
     "seed": 67,
     "suite": "ep-core",
     "timing": {
@@ -5729,8 +6337,8 @@
      "warmup_iters_per_trial": 32
     },
     "topk": 6,
-    "topology_class": "mi325x-xgmi",
-    "transport": "xgmi",
+    "topology_class": "mi325x-xgmi-rdma",
+    "transport": "xgmi-rdma",
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
@@ -5740,22 +6348,22 @@
   },
   {
    "case": {
-    "backend": "mori",
-    "case_id": "mi325x-mori-deepseek-v4-pro-normal-decode-ep8-uniform-bf16",
-    "ep": 8,
+    "backend": "uccl-ep",
+    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-low-latency-decode-ep16-uniform-bf16",
+    "ep": 16,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
     "ladder": "1 2 4 8 16 32 64 128 256 512",
-    "mode": "normal",
-    "nodes": 1,
+    "mode": "low-latency",
+    "nodes": 2,
     "phase": "decode",
     "precision": "bf16",
     "routing": "uniform",
-    "scale_out_transport": null,
+    "scale_out_transport": "rdma",
     "scale_up_domain": 8,
     "scale_up_transport": "xgmi",
-    "scope": "scale-up",
+    "scope": "scale-out",
     "seed": 67,
     "suite": "ep-core",
     "timing": {
@@ -5767,8 +6375,8 @@
      "warmup_iters_per_trial": 32
     },
     "topk": 6,
-    "topology_class": "mi325x-xgmi",
-    "transport": "xgmi",
+    "topology_class": "mi325x-xgmi-rdma",
+    "transport": "xgmi-rdma",
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
@@ -5778,22 +6386,22 @@
   },
   {
    "case": {
-    "backend": "mori",
-    "case_id": "mi325x-mori-deepseek-v4-pro-normal-decode-ep8-uniform-fp8",
-    "ep": 8,
+    "backend": "uccl-ep",
+    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-low-latency-decode-ep16-uniform-fp8",
+    "ep": 16,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
     "ladder": "1 2 4 8 16 32 64 128 256 512",
-    "mode": "normal",
-    "nodes": 1,
+    "mode": "low-latency",
+    "nodes": 2,
     "phase": "decode",
     "precision": "fp8",
     "routing": "uniform",
-    "scale_out_transport": null,
+    "scale_out_transport": "rdma",
     "scale_up_domain": 8,
     "scale_up_transport": "xgmi",
-    "scope": "scale-up",
+    "scope": "scale-out",
     "seed": 67,
     "suite": "ep-core",
     "timing": {
@@ -5805,8 +6413,8 @@
      "warmup_iters_per_trial": 32
     },
     "topk": 6,
-    "topology_class": "mi325x-xgmi",
-    "transport": "xgmi",
+    "topology_class": "mi325x-xgmi-rdma",
+    "transport": "xgmi-rdma",
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
@@ -5816,22 +6424,22 @@
   },
   {
    "case": {
-    "backend": "uccl-ep",
-    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-prefill-ep8-uniform-bf16",
-    "ep": 8,
+    "backend": "mori",
+    "case_id": "mi325x-mori-deepseek-v4-pro-normal-decode-ep16-uniform-bf16",
+    "ep": 16,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
-    "ladder": "1024 2048 4096 8192",
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
     "mode": "normal",
-    "nodes": 1,
-    "phase": "prefill",
+    "nodes": 2,
+    "phase": "decode",
     "precision": "bf16",
     "routing": "uniform",
-    "scale_out_transport": null,
+    "scale_out_transport": "rdma",
     "scale_up_domain": 8,
     "scale_up_transport": "xgmi",
-    "scope": "scale-up",
+    "scope": "scale-out",
     "seed": 67,
     "suite": "ep-core",
     "timing": {
@@ -5843,8 +6451,8 @@
      "warmup_iters_per_trial": 32
     },
     "topk": 6,
-    "topology_class": "mi325x-xgmi",
-    "transport": "xgmi",
+    "topology_class": "mi325x-xgmi-rdma",
+    "transport": "xgmi-rdma",
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
@@ -5854,22 +6462,22 @@
   },
   {
    "case": {
-    "backend": "uccl-ep",
-    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-prefill-ep8-uniform-fp8",
-    "ep": 8,
+    "backend": "mori",
+    "case_id": "mi325x-mori-deepseek-v4-pro-normal-decode-ep16-uniform-fp8",
+    "ep": 16,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
-    "ladder": "1024 2048 4096 8192",
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
     "mode": "normal",
-    "nodes": 1,
-    "phase": "prefill",
+    "nodes": 2,
+    "phase": "decode",
     "precision": "fp8",
     "routing": "uniform",
-    "scale_out_transport": null,
+    "scale_out_transport": "rdma",
     "scale_up_domain": 8,
     "scale_up_transport": "xgmi",
-    "scope": "scale-up",
+    "scope": "scale-out",
     "seed": 67,
     "suite": "ep-core",
     "timing": {
@@ -5881,8 +6489,8 @@
      "warmup_iters_per_trial": 32
     },
     "topk": 6,
-    "topology_class": "mi325x-xgmi",
-    "transport": "xgmi",
+    "topology_class": "mi325x-xgmi-rdma",
+    "transport": "xgmi-rdma",
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
@@ -5893,21 +6501,21 @@
   {
    "case": {
     "backend": "mori",
-    "case_id": "mi325x-mori-deepseek-v4-pro-normal-prefill-ep8-uniform-bf16",
-    "ep": 8,
+    "case_id": "mi325x-mori-deepseek-v4-pro-low-latency-decode-ep16-uniform-bf16",
+    "ep": 16,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
-    "ladder": "1024 2048 4096 8192",
-    "mode": "normal",
-    "nodes": 1,
-    "phase": "prefill",
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "low-latency",
+    "nodes": 2,
+    "phase": "decode",
     "precision": "bf16",
     "routing": "uniform",
-    "scale_out_transport": null,
+    "scale_out_transport": "rdma",
     "scale_up_domain": 8,
     "scale_up_transport": "xgmi",
-    "scope": "scale-up",
+    "scope": "scale-out",
     "seed": 67,
     "suite": "ep-core",
     "timing": {
@@ -5919,8 +6527,8 @@
      "warmup_iters_per_trial": 32
     },
     "topk": 6,
-    "topology_class": "mi325x-xgmi",
-    "transport": "xgmi",
+    "topology_class": "mi325x-xgmi-rdma",
+    "transport": "xgmi-rdma",
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
@@ -5931,21 +6539,21 @@
   {
    "case": {
     "backend": "mori",
-    "case_id": "mi325x-mori-deepseek-v4-pro-normal-prefill-ep8-uniform-fp8",
-    "ep": 8,
+    "case_id": "mi325x-mori-deepseek-v4-pro-low-latency-decode-ep16-uniform-fp8",
+    "ep": 16,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
-    "ladder": "1024 2048 4096 8192",
-    "mode": "normal",
-    "nodes": 1,
-    "phase": "prefill",
+    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "mode": "low-latency",
+    "nodes": 2,
+    "phase": "decode",
     "precision": "fp8",
     "routing": "uniform",
-    "scale_out_transport": null,
+    "scale_out_transport": "rdma",
     "scale_up_domain": 8,
     "scale_up_transport": "xgmi",
-    "scope": "scale-up",
+    "scope": "scale-out",
     "seed": 67,
     "suite": "ep-core",
     "timing": {
@@ -5957,8 +6565,8 @@
      "warmup_iters_per_trial": 32
     },
     "topk": 6,
-    "topology_class": "mi325x-xgmi",
-    "transport": "xgmi",
+    "topology_class": "mi325x-xgmi-rdma",
+    "transport": "xgmi-rdma",
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
@@ -5969,15 +6577,15 @@
   {
    "case": {
     "backend": "uccl-ep",
-    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-decode-ep16-uniform-bf16",
+    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-prefill-ep16-uniform-bf16",
     "ep": 16,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
-    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "ladder": "1024 2048 4096 8192",
     "mode": "normal",
     "nodes": 2,
-    "phase": "decode",
+    "phase": "prefill",
     "precision": "bf16",
     "routing": "uniform",
     "scale_out_transport": "rdma",
@@ -6000,23 +6608,23 @@
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
-   "disposition": "unsupported",
-   "reason": "backend-platform-unsupported",
+   "disposition": "runnable",
+   "reason": null,
    "sku": "mi325x"
   },
   {
    "case": {
-    "backend": "mori",
-    "case_id": "mi325x-mori-deepseek-v4-pro-normal-decode-ep16-uniform-bf16",
+    "backend": "uccl-ep",
+    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-prefill-ep16-uniform-fp8",
     "ep": 16,
     "experts": 384,
     "gpus_per_node": 8,
     "hidden": 7168,
-    "ladder": "1 2 4 8 16 32 64 128 256 512",
+    "ladder": "1024 2048 4096 8192",
     "mode": "normal",
     "nodes": 2,
-    "phase": "decode",
-    "precision": "bf16",
+    "phase": "prefill",
+    "precision": "fp8",
     "routing": "uniform",
     "scale_out_transport": "rdma",
     "scale_up_domain": 8,
@@ -6038,14 +6646,14 @@
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
-   "disposition": "unsupported",
-   "reason": "backend-platform-unsupported",
+   "disposition": "runnable",
+   "reason": null,
    "sku": "mi325x"
   },
   {
    "case": {
-    "backend": "uccl-ep",
-    "case_id": "mi325x-uccl-ep-deepseek-v4-pro-normal-prefill-ep16-uniform-bf16",
+    "backend": "mori",
+    "case_id": "mi325x-mori-deepseek-v4-pro-normal-prefill-ep16-uniform-bf16",
     "ep": 16,
     "experts": 384,
     "gpus_per_node": 8,
@@ -6076,14 +6684,14 @@
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
-   "disposition": "unsupported",
-   "reason": "backend-platform-unsupported",
+   "disposition": "runnable",
+   "reason": null,
    "sku": "mi325x"
   },
   {
    "case": {
     "backend": "mori",
-    "case_id": "mi325x-mori-deepseek-v4-pro-normal-prefill-ep16-uniform-bf16",
+    "case_id": "mi325x-mori-deepseek-v4-pro-normal-prefill-ep16-uniform-fp8",
     "ep": 16,
     "experts": 384,
     "gpus_per_node": 8,
@@ -6092,7 +6700,7 @@
     "mode": "normal",
     "nodes": 2,
     "phase": "prefill",
-    "precision": "bf16",
+    "precision": "fp8",
     "routing": "uniform",
     "scale_out_transport": "rdma",
     "scale_up_domain": 8,
@@ -6114,8 +6722,8 @@
     "workload": "deepseek-v4-pro"
    },
    "detail": null,
-   "disposition": "unsupported",
-   "reason": "backend-platform-unsupported",
+   "disposition": "runnable",
+   "reason": null,
    "sku": "mi325x"
   },
   {
diff --git a/collectivex/run.sh b/collectivex/run.sh
index 5a4f61ba58..498f8506e7 100755
--- a/collectivex/run.sh
+++ b/collectivex/run.sh
@@ -52,9 +52,12 @@ done
 
 # Results record the commit they were measured at; uncommitted edits to the benchmark (anything
 # under collectivex/ except results/) would make that a lie, so they block the run by default.
+checkout_state() {
+  git -C "$REPO_ROOT" status --porcelain --untracked-files=no -- \
+    collectivex ':(exclude)collectivex/results'
+}
 source_sha="$(git -C "$REPO_ROOT" rev-parse HEAD)"
-dirty="$(git -C "$REPO_ROOT" status --porcelain --untracked-files=no -- \
-  collectivex ':(exclude)collectivex/results')"
+dirty="$(checkout_state)"
 source_dirty=0
 if [ -n "$dirty" ]; then
   source_dirty=1
@@ -65,8 +68,13 @@ if [ -n "$dirty" ]; then
   fi
 fi
 
-mkdir -p "$HERE/.shards"
-matrix="$HERE/.shards/matrix-$SKU.json"
+# Each invocation stages its matrix and shard controls in its own directory: two runs from one
+# checkout (one per backend, say) would otherwise overwrite each other's files and run the wrong
+# shards.
+shard_dir="$HERE/.shards/run-$$"
+mkdir -p "$shard_dir"
+trap 'rm -rf "$shard_dir"' EXIT
+matrix="$shard_dir/matrix-$SKU.json"
 python3 "$HERE/sweep_matrix.py" "${matrix_args[@]}" --out "$matrix" >/dev/null
 
 # One tab-separated line per selected shard:
@@ -104,7 +112,15 @@ stamp="$(date -u +%Y%m%dT%H%M%SZ)"
 rcs=()
 for row in "${rows[@]}"; do
   IFS=$'\t' read -r id launcher backend nodes gpn domain ncases image staged minutes <<< "$row"
-  control="$HERE/.shards/$id.json"
+  # Every shard stages the checkout as it is now, but its results carry the source recorded at
+  # start: stop rather than stamp a later shard with a commit it did not run.
+  if [ "$dry" = 0 ] && { [ "$(git -C "$REPO_ROOT" rev-parse HEAD)" != "$source_sha" ] ||
+    [ "$(checkout_state)" != "$dirty" ]; }; then
+    echo "collectivex/ changed since this run started at $source_sha; stopping before $id" >&2
+    rcs+=(drift)
+    break
+  fi
+  control="$shard_dir/$id.json"
   python3 "$HERE/sweep_matrix.py" --extract-from "$matrix" --shard-id "$id" --out "$control" \
     >/dev/null 2>&1
   # The pid keeps two runs of one shard started in the same second (graphed and eager, say) out
@@ -143,8 +159,8 @@ done
 echo "=== summary for $SKU (source $source_sha, dirty=$source_dirty)"
 failed=0
 for i in "${!rows[@]}"; do
-  printf '%-34s rc=%s\n' "${rows[$i]%%$'\t'*}" "${rcs[$i]}"
-  [ "${rcs[$i]}" = 0 ] || failed=$((failed + 1))
+  printf '%-34s rc=%s\n' "${rows[$i]%%$'\t'*}" "${rcs[$i]:-not run}"
+  [ "${rcs[$i]:-}" = 0 ] || failed=$((failed + 1))
 done
 python3 "$HERE/summarize.py" --results-dir "$HERE/results/$SKU" || true
 git -C "$REPO_ROOT" status --short -- "collectivex/results/$SKU"
diff --git a/collectivex/runtime/common.sh b/collectivex/runtime/common.sh
index c39c99e17f..2385081f18 100644
--- a/collectivex/runtime/common.sh
+++ b/collectivex/runtime/common.sh
@@ -125,7 +125,7 @@ collx_load_operator_config() {
   unset COLLX_EXCLUDE_NODES COLLX_NODELIST COLLX_LOCK_DIR COLLX_MASTER_PORT
   unset COLLX_SOCKET_IFNAME COLLX_RDMA_DEVICES COLLX_IB_GID_INDEX COLLX_RDMA_SERVICE_LEVEL
   unset COLLX_RDMA_TRAFFIC_CLASS COLLX_RAIL_ISOLATED COLLX_SINGLE_NODE_RDMA_DEVICES COLLX_RDMA_FABRIC
-  unset COLLX_RDMA_RELAXED_ORDERING COLLX_RDMA_GBS
+  unset COLLX_RDMA_RELAXED_ORDERING COLLX_RDMA_GBS COLLX_HOST_VERBS_PROVIDER
   unset MASTER_ADDR MASTER_PORT RANK WORLD_SIZE LOCAL_RANK LOCAL_WORLD_SIZE
   config_path="${COLLECTIVEX_OPERATOR_CONFIG:-${XDG_CONFIG_HOME:-${HOME}/.config}/inferencex/collectivex.json}"
   if [ ! -e "$config_path" ]; then
@@ -243,6 +243,22 @@ collx_add_efa_host_mounts() {
   CONTAINER_MOUNTS="$CONTAINER_MOUNTS,$mounts"
 }
 
+# Where collx_add_host_verbs_provider_mount binds a host verbs provider inside the container.
+COLLX_HOST_VERBS_ROOT=/opt/collectivex/host-verbs
+
+# Some NIC vendors replace the host's rdma-core provider with their own build for an out-of-tree
+# kernel driver (the MI300X/MI325X pools' Broadcom bnxt_re 237.x speaks kernel ABI 8; rdma-core's stock bnxt_re
+# provider, the image's, supports only ABI 1, so libibverbs lists no device). Multi-node shards
+# bind the registry-named host provider read-only; backend preparation installs it into the
+# container over the image's file (verbs_provider_prepare in prepare_backend.sh).
+collx_add_host_verbs_provider_mount() {
+  local provider="${COLLX_HOST_VERBS_PROVIDER:-}"
+  [ -n "$provider" ] && [ "${NODES:-1}" -gt 1 ] && [ "${COLLX_TRANSPORT:-}" != mnnvl ] || return 0
+  [[ "$provider" =~ ^/[A-Za-z0-9_./-]+/lib[a-z0-9_]+-rdmav[0-9]+\.so$ ]] && [[ "$provider" != *..* ]] \
+    || collx_die "invalid host verbs provider path"
+  CONTAINER_MOUNTS="$CONTAINER_MOUNTS,$provider:$COLLX_HOST_VERBS_ROOT/${provider##*/}:ro"
+}
+
 # AWS EFA scale-out (b300: p6-b300.48xlarge, 16 EFA devices per node; h100-dgxc: p5.48xlarge, 32).
 # EFA is not a verbs HCA: its ports report link_layer Unspecified, it has no GID table, service
 # level or traffic class, and NCCL reaches it only through the aws-ofi-nccl libfabric plugin from
@@ -294,7 +310,7 @@ collx_apply_network_profile() {
   unset NVSHMEM_HCA_LIST NVSHMEM_IB_GID_INDEX NVSHMEM_IB_SL
   unset NVSHMEM_IB_ENABLE_IBGDA NVSHMEM_IBGDA_NIC_HANDLER
   unset EP_NIC_NAME EP_OVERRIDE_RDMA_SL
-  unset MORI_RDMA_DEVICES
+  unset MORI_RDMA_DEVICES MORI_SOCKET_IFNAME
   unset MORI_RDMA_TC MORI_IO_TC MORI_RDMA_SL MORI_IO_SL
   unset UCCL_SOCKET_IFNAME UCCL_IB_HCA UCCL_IB_GID_INDEX UCCL_IB_SL UCCL_IB_TC
   unset UCCL_IB_MAX_INFLIGHT_BYTES UCCL_IB_MAX_INFLIGHT_NORMAL UCCL_EP_ENABLE_AGGRESSIVE_ATOMIC
@@ -317,7 +333,10 @@ collx_apply_network_profile() {
   if [ -n "${COLLX_SOCKET_IFNAME:-}" ]; then
     [[ "$COLLX_SOCKET_IFNAME" =~ ^[A-Za-z][A-Za-z0-9_.-]{0,31}$ ]] \
       || collx_die "invalid private socket interface selector"
-    export NCCL_SOCKET_IFNAME="$COLLX_SOCKET_IFNAME" GLOO_SOCKET_IFNAME="$COLLX_SOCKET_IFNAME"
+    # MoRI's shmem bootstrap otherwise picks its own interface, which can be a rail without a
+    # kernel route to the root rank (per-rail /31 links on the Broadcom pools).
+    export NCCL_SOCKET_IFNAME="$COLLX_SOCKET_IFNAME" GLOO_SOCKET_IFNAME="$COLLX_SOCKET_IFNAME" \
+      MORI_SOCKET_IFNAME="$COLLX_SOCKET_IFNAME"
   fi
   [[ "$COLLX_RDMA_DEVICES" =~ ^[A-Za-z][A-Za-z0-9_.-]{0,31}(:[1-9][0-9]*)?(,[A-Za-z][A-Za-z0-9_.-]{0,31}(:[1-9][0-9]*)?)*$ ]] \
     || collx_die "invalid private RDMA device selector"
@@ -553,6 +572,17 @@ collx_slurm_job_name() {
 }
 
 # The job name is prepended so a caller-supplied --job-name wins (salloc takes the last).
+# True only when Slurm says the job is gone: listed as nothing, or purged ("Invalid job id"). A
+# failed query (slurmctld busy or restarting) is not an answer, so it reads as still there.
+collx_job_gone() {
+  local out
+  if out="$(squeue -h -j "$1" -o %A 2>&1)"; then
+    [ -z "$out" ]
+  else
+    [[ "$out" == *"Invalid job id"* ]]
+  fi
+}
+
 collx_salloc_jobid() {
   local log_label log job_id
   set -- --job-name="$(collx_slurm_job_name)" "$@"
@@ -562,7 +592,25 @@ collx_salloc_jobid() {
     return 1
   fi
   collx_log "scheduler-request=submit"
-  if ! (salloc "$@" --no-shell) > "$log" 2>&1; then
+  # salloc --no-shell does not notice a job that ends between the grant and its own poll (a node
+  # prolog that fails the job ten seconds in leaves it waiting forever), so the job is watched
+  # here: once Slurm no longer lists the queued job and no grant was logged, salloc is stopped.
+  (salloc "$@" --no-shell) > "$log" 2>&1 &
+  local salloc_pid=$! queued=""
+  while kill -0 "$salloc_pid" 2>/dev/null; do
+    sleep 10
+    grep -q "Granted job allocation" "$log" && break
+    [ -n "$queued" ] || queued="$(sed -nE 's/.*Pending job allocation ([1-9][0-9]*).*/\1/p' "$log" | head -n1)"
+    if [ -n "$queued" ] && collx_job_gone "$queued" \
+        && ! grep -q "Granted job allocation" "$log"; then
+      collx_log "ERROR: scheduled job $queued ended before its allocation was granted"
+      kill -TERM "$salloc_pid" 2>/dev/null
+      wait "$salloc_pid" 2>/dev/null
+      collx_log_tail "$log"
+      return 1
+    fi
+  done
+  if ! wait "$salloc_pid"; then
     collx_log "ERROR: scheduler allocation failed"
     collx_log_tail "$log"
     return 1
diff --git a/collectivex/runtime/config.py b/collectivex/runtime/config.py
index cdccfc88a3..5aecacbfd7 100644
--- a/collectivex/runtime/config.py
+++ b/collectivex/runtime/config.py
@@ -16,7 +16,7 @@
 NETWORK_FIELDS = {
     "socket_ifname", "rdma_devices", "ib_gid_index", "rdma_service_level",
     "rdma_traffic_class", "rail_isolated", "single_node_rdma_devices", "rdma_fabric",
-    "rdma_relaxed_ordering", "rdma_gbs",
+    "rdma_relaxed_ordering", "rdma_gbs", "host_verbs_provider",
 }
 # Timing knobs, in the order the legacy colon-string encoded them, paired with the run_ep flag
 # each one drives. The names match configs/sweep.json so a case's timing block is readable
diff --git a/collectivex/runtime/elf_vernaux.py b/collectivex/runtime/elf_vernaux.py
new file mode 100644
index 0000000000..08b3368013
--- /dev/null
+++ b/collectivex/runtime/elf_vernaux.py
@@ -0,0 +1,76 @@
+"""Retarget one version requirement of a shared library to another version of the same library.
+
+Broadcom's host libbnxt_re is linked on Ubuntu 24.04 and needs GLIBC_2.38 for one symbol,
+__isoc23_strtol; the ROCm images ship glibc 2.35. After that symbol is unversioned and supplied by a
+shim, the GLIBC_2.38 entry in .gnu.version_r still makes the loader refuse the library. This rewrites
+that entry's name and hash to a version libc does provide, in place; nothing else in the file moves.
+
+usage: elf_vernaux.py LIBRARY FROM_VERSION TO_VERSION
+"""
+import struct
+import sys
+
+SHT_GNU_VERNEED = 0x6FFFFFFE
+
+
+def elf_hash(name: bytes) -> int:
+    h = 0
+    for byte in name:
+        h = ((h << 4) + byte) & 0xFFFFFFFF
+        g = h & 0xF0000000
+        if g:
+            h ^= g >> 24
+        h &= ~g & 0xFFFFFFFF
+    return h
+
+
+def retarget(path: str, old: str, new: str) -> None:
+    with open(path, "r+b") as stream:
+        data = bytearray(stream.read())
+        if data[:4] != b"\x7fELF" or data[4] != 2 or data[5] != 1:
+            raise SystemExit("not a little-endian ELF64 file")
+        shoff, = struct.unpack_from("<Q", data, 0x28)
+        shentsize, shnum = struct.unpack_from("<HH", data, 0x3A)
+        sections = [struct.unpack_from("<IIQQQQIIQQ", data, shoff + i * shentsize) for i in range(shnum)]
+        verneed = [s for s in sections if s[1] == SHT_GNU_VERNEED]
+        if len(verneed) != 1:
+            raise SystemExit("expected one .gnu.version_r section")
+        _, _, _, _, offset, _, link, info, _, _ = verneed[0]
+        strtab = sections[link][4]
+
+        def name_at(index: int) -> bytes:
+            start = strtab + index
+            return bytes(data[start:data.index(0, start)])
+
+        targets, replacement = [], None
+        entry = offset
+        for _ in range(info):
+            _, _, _, aux, following = struct.unpack_from("<HHIII", data, entry)
+            cursor = entry + aux
+            while True:
+                _, _, _, name, step = struct.unpack_from("<IHHII", data, cursor)
+                if name_at(name) == old.encode():
+                    targets.append(cursor)
+                if name_at(name) == new.encode():
+                    replacement = name
+                if not step:
+                    break
+                cursor += step
+            if not following:
+                break
+            entry += following
+        if not targets:
+            raise SystemExit(f"{old} is not required")
+        if replacement is None:
+            raise SystemExit(f"{new} is not already required, so its name is not in the string table")
+        for cursor in targets:
+            struct.pack_into("<I", data, cursor, elf_hash(new.encode()))
+            struct.pack_into("<I", data, cursor + 8, replacement)
+        stream.seek(0)
+        stream.write(data)
+
+
+if __name__ == "__main__":
+    if len(sys.argv) != 4:
+        raise SystemExit(__doc__)
+    retarget(*sys.argv[1:])
diff --git a/collectivex/runtime/prepare_backend.sh b/collectivex/runtime/prepare_backend.sh
index 9450e9c0c5..76079c1e87 100644
--- a/collectivex/runtime/prepare_backend.sh
+++ b/collectivex/runtime/prepare_backend.sh
@@ -809,8 +809,83 @@ flashinfer_ep_prepare() {
   flashinfer_probe || { collx_log "ERROR: FlashInfer EP one-sided A2A unavailable"; return 1; }
 }
 
+# Host verbs provider (collx_add_host_verbs_provider_mount). The vendor's build targets the host's
+# distro: Broadcom's libbnxt_re 237.1.137.0 is linked on Ubuntu 24.04 and needs GLIBC_2.38 for one
+# symbol, __isoc23_strtol (C23 strtol, which only adds 0b prefixes), while the ROCm images ship glibc
+# 2.35. That symbol is unversioned and served by a forwarding shim, and the now-unused GLIBC_2.38
+# requirement is retargeted to GLIBC_2.2.5 (elf_vernaux.py). The result must load and must list the
+# registry's RDMA devices, or preparation fails. It replaces the image's provider of the same name
+# in this node's writable container, which every rank of the shard then loads.
+readonly VERBS_PATCHELF_SPEC="patchelf==0.17.2.1"
+
+# The RDMA devices libibverbs lists, space-separated. ibv_devices is not in every image.
+verbs_device_names() {
+  python3 - <<'PY'
+import ctypes
+lib = ctypes.CDLL("libibverbs.so.1")
+lib.ibv_get_device_list.restype = ctypes.POINTER(ctypes.c_void_p)
+lib.ibv_get_device_name.restype = ctypes.c_char_p
+count = ctypes.c_int(0)
+devices = lib.ibv_get_device_list(ctypes.byref(count))
+print(" ".join(lib.ibv_get_device_name(ctypes.c_void_p(devices[i])).decode()
+               for i in range(count.value)) if devices else "")
+PY
+}
+
+verbs_lists_registry_devices() {
+  local listed device
+  listed=" $(verbs_device_names 2>/dev/null) "
+  for device in ${COLLX_RDMA_DEVICES//,/ }; do
+    [[ "$listed" == *" ${device%%:*} "* ]] || return 1
+  done
+}
+
+verbs_provider_prepare() {
+  local provider="${COLLX_HOST_VERBS_PROVIDER:-}" name source providers target work
+  [ -n "$provider" ] && [ "${COLLX_NODES:-1}" -gt 1 ] && [ "${COLLX_TRANSPORT:-}" != mnnvl ] || return 0
+  # An image that ships a provider for this driver (the vLLM ROCm images carry Broadcom's) needs
+  # nothing from the host.
+  if verbs_lists_registry_devices; then
+    collx_log "verbs provider: the image's provider lists every registry device; keeping it"
+    return 0
+  fi
+  name="${provider##*/}"
+  source="$COLLX_HOST_VERBS_ROOT/$name"
+  [ -r "$source" ] || { collx_log "ERROR: host verbs provider $name is not bound"; return 1; }
+  # rdma-core tries its provider directory before the loader path, so a copy there wins over one
+  # an image installs elsewhere (/usr/local/lib for Broadcom's packages).
+  providers="$(ls -d /usr/lib/*-linux-gnu/libibverbs 2>/dev/null | head -n 1)"
+  [ -n "$providers" ] || { collx_log "ERROR: the image has no rdma-core provider directory"; return 1; }
+  target="$providers/$name"
+  work="$(mktemp -d /tmp/collectivex-verbs.XXXXXX)" || return 1
+  cp "$source" "$work/$name" || return 1
+  if ! python3 -c 'import ctypes, sys; ctypes.CDLL(sys.argv[1])' "$work/$name" 2>/dev/null; then
+    collx_log "verbs provider: $name needs a newer glibc; shimming __isoc23_strtol"
+    printf '%s\n' '#include <stdlib.h>' \
+      'long __isoc23_strtol(const char *s, char **end, int base) { return strtol(s, end, base); }' \
+      > "$work/cx_c23.c"
+    cc -shared -fPIC -O2 -o "$work/libcx_c23.so" "$work/cx_c23.c" \
+      && python3 -m pip install -q --disable-pip-version-check --no-input \
+        --target "$work/patchelf" "$VERBS_PATCHELF_SPEC" >&2 \
+      && "$work/patchelf/bin/patchelf" --clear-symbol-version __isoc23_strtol \
+        --add-needed libcx_c23.so --set-rpath '$ORIGIN' "$work/$name" \
+      && python3 "$COLLX_RUNTIME_DIR/elf_vernaux.py" "$work/$name" GLIBC_2.38 GLIBC_2.2.5 \
+      || { collx_log "ERROR: verbs provider patch failed"; return 1; }
+    python3 -c 'import ctypes, sys; ctypes.CDLL(sys.argv[1])' "$work/$name" \
+      || { collx_log "ERROR: patched verbs provider does not load"; return 1; }
+    cp "$work/libcx_c23.so" "$providers/libcx_c23.so" || return 1
+  fi
+  cp "$work/$name" "$target.collectivex" && mv -f "$target.collectivex" "$target" \
+    || { collx_log "ERROR: verbs provider install failed"; return 1; }
+  rm -rf "$work"
+  verbs_lists_registry_devices \
+    || { collx_log "ERROR: verbs provider $name does not list every registry device"; return 1; }
+  collx_log "verbs provider: installed host $name"
+}
+
 main() {
   collx_apply_network_profile "${COLLX_NODES:-1}" "${COLLX_TRANSPORT:-}" || return 1
+  verbs_provider_prepare || return 1
   validate_container_network || return 1
   ofi_nccl_prepare || return 1
   case "$COLLX_BENCH" in
diff --git a/collectivex/tests/test_elf_vernaux.py b/collectivex/tests/test_elf_vernaux.py
new file mode 100644
index 0000000000..62be70c746
--- /dev/null
+++ b/collectivex/tests/test_elf_vernaux.py
@@ -0,0 +1,68 @@
+"""elf_vernaux.py retargets one version requirement in place and refuses anything else."""
+import struct
+import sys
+import tempfile
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "runtime"))
+import elf_vernaux  # noqa: E402
+
+
+def _library(versions) -> bytes:
+    """A minimal ELF64 with a string table and one .gnu.version_r entry (libc.so.6) whose aux
+    records require `versions`, in order."""
+    strings = b"\0libc.so.6\0" + b"".join(v.encode() + b"\0" for v in versions)
+    offset_of = lambda text: strings.index(text.encode() + b"\0")
+    verneed = struct.pack("<HHIII", 1, len(versions), offset_of("libc.so.6"), 16, 0)
+    for index, version in enumerate(versions):
+        step = 16 if index + 1 < len(versions) else 0
+        verneed += struct.pack("<IHHII", elf_vernaux.elf_hash(version.encode()), 0, index + 2,
+                               offset_of(version), step)
+    strtab_offset = 64
+    verneed_offset = strtab_offset + len(strings)
+    shoff = verneed_offset + len(verneed)
+    header = b"\x7fELF" + bytes([2, 1, 1]) + bytes(9)
+    header += struct.pack("<HHIQQQIHHHHHH", 3, 62, 1, 0, 0, shoff, 0, 64, 0, 0, 64, 3, 0)
+    sections = struct.pack("<IIQQQQIIQQ", 0, 0, 0, 0, 0, 0, 0, 0, 0, 0)
+    sections += struct.pack("<IIQQQQIIQQ", 0, 3, 0, 0, strtab_offset, len(strings), 0, 0, 1, 0)
+    sections += struct.pack("<IIQQQQIIQQ", 0, elf_vernaux.SHT_GNU_VERNEED, 0, 0, verneed_offset,
+                            len(verneed), 1, 1, 8, 0)
+    return header + strings + verneed + sections
+
+
+class Retarget(unittest.TestCase):
+    def _run(self, versions, old, new):
+        with tempfile.TemporaryDirectory() as directory:
+            path = Path(directory) / "lib.so"
+            path.write_bytes(_library(versions))
+            before = path.read_bytes()
+            elf_vernaux.retarget(str(path), old, new)
+            return before, path.read_bytes()
+
+    def test_the_old_requirement_takes_the_new_name_and_hash(self) -> None:
+        before, after = self._run(["GLIBC_2.2.5", "GLIBC_2.38"], "GLIBC_2.38", "GLIBC_2.2.5")
+        self.assertEqual(len(before), len(after))
+        strtab = 64
+        aux = before.index(struct.pack("<I", elf_vernaux.elf_hash(b"GLIBC_2.38")))
+        new_hash, _, _, new_name, _ = struct.unpack_from("<IHHII", after, aux)
+        self.assertEqual(new_hash, elf_vernaux.elf_hash(b"GLIBC_2.2.5"))
+        self.assertEqual(new_name, before.index(b"GLIBC_2.2.5\0") - strtab)
+        # Only those eight bytes change.
+        changed = [i for i in range(len(before)) if before[i] != after[i]]
+        self.assertTrue(changed and min(changed) >= aux and max(changed) < aux + 12)
+
+    def test_a_missing_requirement_or_replacement_fails_closed(self) -> None:
+        with self.assertRaises(SystemExit):
+            self._run(["GLIBC_2.2.5"], "GLIBC_2.38", "GLIBC_2.2.5")
+        with self.assertRaises(SystemExit):
+            self._run(["GLIBC_2.38"], "GLIBC_2.38", "GLIBC_2.2.5")
+
+    def test_the_hash_is_the_sysv_elf_hash(self) -> None:
+        # Values glibc's own verneed entries carry.
+        self.assertEqual(elf_vernaux.elf_hash(b"GLIBC_2.2.5"), 0x09691A75)
+        self.assertEqual(elf_vernaux.elf_hash(b"GLIBC_2.38"), 0x069691B8)
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/collectivex/tests/test_runtime.py b/collectivex/tests/test_runtime.py
index 310d5b4e27..ae6abf2312 100644
--- a/collectivex/tests/test_runtime.py
+++ b/collectivex/tests/test_runtime.py
@@ -228,6 +228,47 @@ def test_nvshmem_hca_list_per_placement(self) -> None:
                 self.assertEqual(stdout.strip().splitlines()[-1], expected)
 
 
+class HostVerbsProviderMountTests(unittest.TestCase):
+    PROVIDER = "/usr/local/lib/x86_64-linux-gnu/libbnxt_re-rdmav34.so"
+
+    @staticmethod
+    def _mounts(provider: str, nodes: int):
+        return run_common(
+            f"COLLX_HOST_VERBS_PROVIDER='{provider}' NODES={nodes} COLLX_TRANSPORT=xgmi-rdma"
+            " CONTAINER_MOUNTS=/stage:/ix;"
+            ' collx_add_host_verbs_provider_mount && echo "$CONTAINER_MOUNTS"')
+
+    def test_multi_node_shards_bind_the_host_provider_read_only(self) -> None:
+        # Backend preparation installs it over the image's provider; the bind only delivers it.
+        result = self._mounts(self.PROVIDER, 2)
+        self.assertEqual(result.stdout.strip(), "/stage:/ix,"
+                         f"{self.PROVIDER}:/opt/collectivex/host-verbs/libbnxt_re-rdmav34.so:ro")
+
+    def test_single_node_and_unset_keep_their_mounts(self) -> None:
+        for provider, nodes in ((self.PROVIDER, 1), ("", 2)):
+            with self.subTest(provider=provider, nodes=nodes):
+                self.assertEqual(self._mounts(provider, nodes).stdout.strip(), "/stage:/ix")
+
+    def test_a_path_that_is_not_a_verbs_provider_fails_closed(self) -> None:
+        for provider in ("/usr/lib/libc.so.6", "relative/libbnxt_re-rdmav34.so",
+                         "/usr/../etc/libbnxt_re-rdmav34.so"):
+            with self.subTest(provider=provider):
+                self.assertNotEqual(self._mounts(provider, 2).returncode, 0)
+
+
+class SocketInterfaceProfileTests(unittest.TestCase):
+    def test_every_bootstrap_uses_the_registry_interface(self) -> None:
+        # MoRI's shmem bootstrap picks its own interface otherwise; on per-rail /31 links that can
+        # be one without a kernel route to the root rank.
+        stdout = run_common(
+            "export COLLX_RDMA_DEVICES=bnxt_re0 COLLX_SOCKET_IFNAME=mgmt0 MORI_SOCKET_IFNAME=stale;"
+            " collx_apply_network_profile 2 xgmi-rdma;"
+            ' echo "$NCCL_SOCKET_IFNAME $GLOO_SOCKET_IFNAME $MORI_SOCKET_IFNAME $UCCL_SOCKET_IFNAME"',
+            check=True,
+        ).stdout
+        self.assertEqual(stdout.strip().splitlines()[-1], "mgmt0 mgmt0 mgmt0 mgmt0")
+
+
 class RelaxedOrderingProfileTests(unittest.TestCase):
     def test_only_the_registry_flag_relaxes_the_deepep_window(self) -> None:
         for flag, expected in (("1", "1"), ("0", "unset"), ("", "unset")):