“isolated follow-up checks confirm the actual failing component” addresses diagnostic result has not been mapped to the relevant hardware or software subsystem, but the current hardware or infrastructure issue is component-specific test outcomes require targeted root-cause isolation. The option “follow-up network tests identify and correct the low-bandwidth connection” is more directly tied to the required outcome.

“ClusterKit report shows passing configured node and link diagnostics” addresses node-only GPU checks do not cover cluster-wide network readiness, but the current hardware or infrastructure issue is component-specific test outcomes require targeted root-cause isolation. The option “follow-up network tests identify and correct the low-bandwidth connection” is more directly tied to the required outcome.

“all required nodes exceed minimum throughput and outliers have resolutions” addresses aggregate scores conceal outlier node and fabric-path failures, but the current hardware or infrastructure issue is component-specific test outcomes require targeted root-cause isolation. The option “follow-up network tests identify and correct the low-bandwidth connection” is more directly tied to the required outcome.

Useful for node-only GPU checks do not cover cluster-wide network readiness, but it does not remove the specific condition: concurrent tenant workload interferes with baseline cluster testing.

“inspect failed test metadata and supporting device telemetry first” addresses diagnostic result has not been mapped to the relevant hardware or software subsystem, but the current hardware or infrastructure issue is concurrent tenant workload interferes with baseline cluster testing. The option “reserve an approved maintenance allocation and standardize test background load” is more directly tied to the required outcome.

“capture test configuration node class and expected baseline with each run” addresses benchmark configuration and cohort definitions are not comparable, but the current hardware or infrastructure issue is concurrent tenant workload interferes with baseline cluster testing. The option “reserve an approved maintenance allocation and standardize test background load” is more directly tied to the required outcome.

“ClusterKit report shows passing configured node and link diagnostics” addresses node-only GPU checks do not cover cluster-wide network readiness, but the current hardware or infrastructure issue is concurrent tenant workload interferes with baseline cluster testing. The option “repeated ClusterKit measures converge within specified variance” is more directly tied to the required outcome.

“every planned node produces a complete result entry in the report” addresses cluster test inventory or SSH reachability excludes required workers, but the current hardware or infrastructure issue is concurrent tenant workload interferes with baseline cluster testing. The option “repeated ClusterKit measures converge within specified variance” is more directly tied to the required outcome.

“reported performance deltas are attributable to equivalent selected tests” addresses benchmark configuration and cohort definitions are not comparable, but the current hardware or infrastructure issue is concurrent tenant workload interferes with baseline cluster testing. The option “repeated ClusterKit measures converge within specified variance” is more directly tied to the required outcome.

“inspect per-node and per-link ClusterKit distributions with rejection thresholds” addresses aggregate scores conceal outlier node and fabric-path failures, but the current hardware or infrastructure issue is diagnostic result has not been mapped to the relevant hardware or software subsystem. The option “inspect failed test metadata and supporting device telemetry first” is more directly tied to the required outcome.

Useful for node-only GPU checks do not cover cluster-wide network readiness, but it does not remove the specific condition: diagnostic result has not been mapped to the relevant hardware or software subsystem.

“reserve an approved maintenance allocation and standardize test background load” addresses concurrent tenant workload interferes with baseline cluster testing, but the current hardware or infrastructure issue is diagnostic result has not been mapped to the relevant hardware or software subsystem. The option “inspect failed test metadata and supporting device telemetry first” is more directly tied to the required outcome.

“follow-up network tests identify and correct the low-bandwidth connection” addresses component-specific test outcomes require targeted root-cause isolation, but the current hardware or infrastructure issue is diagnostic result has not been mapped to the relevant hardware or software subsystem. The option “isolated follow-up checks confirm the actual failing component” is more directly tied to the required outcome.

“every planned node produces a complete result entry in the report” addresses cluster test inventory or SSH reachability excludes required workers, but the current hardware or infrastructure issue is diagnostic result has not been mapped to the relevant hardware or software subsystem. The option “isolated follow-up checks confirm the actual failing component” is more directly tied to the required outcome.

“reported performance deltas are attributable to equivalent selected tests” addresses benchmark configuration and cohort definitions are not comparable, but the current hardware or infrastructure issue is diagnostic result has not been mapped to the relevant hardware or software subsystem. The option “isolated follow-up checks confirm the actual failing component” is more directly tied to the required outcome.

“run a supported ClusterKit assessment spanning required node and fabric tests” addresses node-only GPU checks do not cover cluster-wide network readiness, but the current hardware or infrastructure issue is incomplete suite execution was mistaken for a complete pass. The option “require every mandatory ClusterKit stage and record explicit pass or exception” is more directly tied to the required outcome.

“inspect per-node and per-link ClusterKit distributions with rejection thresholds” addresses aggregate scores conceal outlier node and fabric-path failures, but the current hardware or infrastructure issue is incomplete suite execution was mistaken for a complete pass. The option “require every mandatory ClusterKit stage and record explicit pass or exception” is more directly tied to the required outcome.

“inspect failed test metadata and supporting device telemetry first” addresses diagnostic result has not been mapped to the relevant hardware or software subsystem, but the current hardware or infrastructure issue is incomplete suite execution was mistaken for a complete pass. The option “require every mandatory ClusterKit stage and record explicit pass or exception” is more directly tied to the required outcome.

“all required nodes exceed minimum throughput and outliers have resolutions” addresses aggregate scores conceal outlier node and fabric-path failures, but the current hardware or infrastructure issue is incomplete suite execution was mistaken for a complete pass. The option “output manifest contains all required completed tests with no skipped acceptance stages” is more directly tied to the required outcome.

“ClusterKit report shows passing configured node and link diagnostics” addresses node-only GPU checks do not cover cluster-wide network readiness, but the current hardware or infrastructure issue is incomplete suite execution was mistaken for a complete pass. The option “output manifest contains all required completed tests with no skipped acceptance stages” is more directly tied to the required outcome.

“every planned node produces a complete result entry in the report” addresses cluster test inventory or SSH reachability excludes required workers, but the current hardware or infrastructure issue is incomplete suite execution was mistaken for a complete pass. The option “output manifest contains all required completed tests with no skipped acceptance stages” is more directly tied to the required outcome.

“bisect NCCL participants and correlate per-node NIC and fabric counters” addresses specific nodes or network paths degrade the larger collective, but the current hardware or infrastructure issue is end-to-end collective communication path has an unverified performance bottleneck. The option “run NCCL tests across the involved racks with per-link congestion monitoring” is more directly tied to the required outcome.

Useful for intended fabric rail balancing is not realized in collective traffic, but it does not remove the specific condition: end-to-end collective communication path has an unverified performance bottleneck.

“map ranks to the intended GPU pod and fabric topology before benchmarking” addresses process placement and communication topology are misaligned, but the current hardware or infrastructure issue is end-to-end collective communication path has an unverified performance bottleneck. The option “run NCCL tests across the involved racks with per-link congestion monitoring” is more directly tied to the required outcome.

“tuned network path meets accepted latency and NCCL bandwidth without persistent congestion” addresses microburst congestion and flow control effects are hidden by averages, but the current hardware or infrastructure issue is end-to-end collective communication path has an unverified performance bottleneck. The option “cross-rack NCCL bus bandwidth meets the approved topology baseline” is more directly tied to the required outcome.

“repaired node set achieves comparable bandwidth across all test subsets” addresses specific nodes or network paths degrade the larger collective, but the current hardware or infrastructure issue is end-to-end collective communication path has an unverified performance bottleneck. The option “cross-rack NCCL bus bandwidth meets the approved topology baseline” is more directly tied to the required outcome.

“NCCL logs show correct HCA selection and full-speed all-reduce” addresses communication ranks are bound to nonfunctional or slow fabric adapters, but the current hardware or infrastructure issue is end-to-end collective communication path has an unverified performance bottleneck. The option “cross-rack NCCL bus bandwidth meets the approved topology baseline” is more directly tied to the required outcome.

Useful for intended fabric rail balancing is not realized in collective traffic, but it does not remove the specific condition: local NVLink measurements do not establish multi-node fabric throughput.

“bisect NCCL participants and correlate per-node NIC and fabric counters” addresses specific nodes or network paths degrade the larger collective, but the current hardware or infrastructure issue is local NVLink measurements do not establish multi-node fabric throughput. The option “execute NCCL collective tests that span physical nodes and switches” is more directly tied to the required outcome.

“choose and verify intended network interfaces for every NCCL process” addresses communication ranks are bound to nonfunctional or slow fabric adapters, but the current hardware or infrastructure issue is local NVLink measurements do not establish multi-node fabric throughput. The option “execute NCCL collective tests that span physical nodes and switches” is more directly tied to the required outcome.

“cross-rack NCCL bus bandwidth meets the approved topology baseline” addresses end-to-end collective communication path has an unverified performance bottleneck, but the current hardware or infrastructure issue is local NVLink measurements do not establish multi-node fabric throughput. The option “multi-node benchmark results verify the intended fabric data path” is more directly tied to the required outcome.

“results show matched parameters and explain expected scaling behavior” addresses collective-size and participant-count effects are ignored in baseline comparisons, but the current hardware or infrastructure issue is local NVLink measurements do not establish multi-node fabric throughput. The option “multi-node benchmark results verify the intended fabric data path” is more directly tied to the required outcome.

“tuned network path meets accepted latency and NCCL bandwidth without persistent congestion” addresses microburst congestion and flow control effects are hidden by averages, but the current hardware or infrastructure issue is local NVLink measurements do not establish multi-node fabric throughput. The option “multi-node benchmark results verify the intended fabric data path” is more directly tied to the required outcome.