apiVersion: "catalog.confighub.com/v1alpha1"
kind: "ConfigReviewRecord"
metadata:
  name: "aicr-l40-node-snapshot-review"
spec:
  question: "What differs between these GPU nodes, and does the difference matter for each node's intended role?"
  source:
    format: "aicr-snapshot"
    inspectionCommands:
      - "aicr snapshot --output baseline.yaml"
      - "aicr snapshot --output current.yaml"
      - "aicr diff --baseline baseline.yaml --target current.yaml"
    note: "Snapshot and diff inspect observed node state. They do not require a recipe or deploy a bundle."
  snapshots:
    baseline:
      path: "tests/fixtures/config-assessment/aicr-snapshot-baseline.yaml"
      sha256: "27172d5c9ca2a59341ff403b075d765334690b79365209d8111beb4e47da0a22"
      measurementCount: 4
    target:
      path: "tests/fixtures/config-assessment/aicr-snapshot-target.yaml"
      sha256: "094a1b5764bdcc380461ea4e1eddb9c6bab271fb4bed7ff578b072ffbd8a641c"
      measurementCount: 4
  observedDifferences:
    -
      path: "OS/kernel-cmdline/cmdline"
      baseline: "quiet splash iommu=pt"
      target: "quiet splash"
      classification: "observed-difference"
    -
      path: "OS/kernel-modules/nvidia_peermem"
      baseline: "loaded"
      target: "absent"
      classification: "observed-difference"
  unboundInterpretation:
    result: "finding"
    decision: "The snapshots differ. Select the intended profile for each node before deciding whether either node should change."
  selectedIntent:
    profileCatalog:
      path: "config-catalog/aicr-snapshot-profiles.yaml"
      sha256: "23193c03ad70e179a93e02fadd03d91a44f8b2a692372f6066a5e6092418f888"
      provider:
        name: "Config Workshop demonstration"
        role: "deployment-profile-curator"
      version: "2026-08-25"
      use: "demonstration-only"
    upstream:
      project: "NVIDIA AICR"
      version: "v0.20.0"
      sourceCatalogRecord: "examples/aicr/eks-h100-training-kubeflow-v0-20-0/source-catalog/source-catalog-record.yaml"
      sourceCatalogRecordSha256: "b136f9e9c0614d7ef86728c1723237325c4a96dc4e73b58d8f6d3d7ef88c7238"
      retainedCatalogInventorySha256: "676f2d59eacd79ae1b72e5cbe00216b577def1da412dbdabb032f317a62dc1d8"
    bindings:
      baseline:
        profile: "l40-mellanox-rdma"
        dimensions:
          accelerator: "l40"
          network: "mellanox"
          rdma: true
      target:
        profile: "l40-standard-networking"
        dimensions:
          accelerator: "l40"
          network: "standard"
          rdma: false
  variantAssessment:
    baseline:
      profile: "l40-mellanox-rdma"
      provider:
        name: "Config Workshop demonstration"
        role: "deployment-profile-curator"
      dimensions:
        accelerator: "l40"
        network: "mellanox"
        rdma: true
      targetAssumptions:
        - "The node has a Mellanox network role that requires RDMA."
        - "The operating system and driver stack are expected to expose peer-memory support."
      result: "pass"
      requirements:
        -
          path: "GPU/hardware/model"
          required: true
          expected:
            operator: "equals"
            value: "L40"
          observed: "L40"
          result: "pass"
          explanation: "The profile is intended for an NVIDIA L40 node."
        -
          path: "OS/kernel-cmdline/cmdline"
          required: true
          expected:
            operator: "contains"
            value: "iommu=pt"
          observed: "quiet splash iommu=pt"
          result: "pass"
          explanation: "The RDMA profile requires this kernel command-line setting."
        -
          path: "OS/kernel-modules/nvidia_peermem"
          required: true
          expected:
            operator: "equals"
            value: "loaded"
          observed: "loaded"
          result: "pass"
          explanation: "The RDMA profile requires the peer-memory module."
      decision: "The recorded fields satisfy this profile; checks outside the snapshot remain separate."
    target:
      profile: "l40-standard-networking"
      provider:
        name: "Config Workshop demonstration"
        role: "deployment-profile-curator"
      dimensions:
        accelerator: "l40"
        network: "standard"
        rdma: false
      targetAssumptions:
        - "No Mellanox or other RDMA network role is assigned to this node."
        - "Peer-memory support is not required for the intended workload."
      result: "pass"
      requirements:
        -
          path: "GPU/hardware/model"
          required: true
          expected:
            operator: "equals"
            value: "L40"
          observed: "L40"
          result: "pass"
          explanation: "The profile is intended for an NVIDIA L40 node."
        -
          path: "OS/kernel-cmdline/cmdline"
          required: false
          observed: "quiet splash"
          result: "not-applicable"
          explanation: "This profile does not require the RDMA kernel setting."
        -
          path: "OS/kernel-modules/nvidia_peermem"
          required: false
          observed: "absent"
          result: "not-applicable"
          explanation: "This profile does not require the peer-memory module."
      decision: "The recorded fields satisfy this profile; checks outside the snapshot remain separate."
    targetUsingBaselineProfile:
      purpose: "Shows why the same observed state can be correct for one variant and wrong for another."
      profile: "l40-mellanox-rdma"
      provider:
        name: "Config Workshop demonstration"
        role: "deployment-profile-curator"
      dimensions:
        accelerator: "l40"
        network: "mellanox"
        rdma: true
      targetAssumptions:
        - "The node has a Mellanox network role that requires RDMA."
        - "The operating system and driver stack are expected to expose peer-memory support."
      result: "finding"
      requirements:
        -
          path: "GPU/hardware/model"
          required: true
          expected:
            operator: "equals"
            value: "L40"
          observed: "L40"
          result: "pass"
          explanation: "The profile is intended for an NVIDIA L40 node."
        -
          path: "OS/kernel-cmdline/cmdline"
          required: true
          expected:
            operator: "contains"
            value: "iommu=pt"
          observed: "quiet splash"
          result: "finding"
          explanation: "The RDMA profile requires this kernel command-line setting."
        -
          path: "OS/kernel-modules/nvidia_peermem"
          required: true
          expected:
            operator: "equals"
            value: "loaded"
          observed: "absent"
          result: "finding"
          explanation: "The RDMA profile requires the peer-memory module."
      decision: "2 required settings do not match this profile."
  assessmentClasses:
    nodeInspection:
      evidenceState: "completed"
      resultState: "available"
      note: "Two exact snapshot files were parsed, hashed, and compared."
    configurationValidation:
      evidenceState: "not-applicable"
      resultState: "not-applicable"
      note: "A node snapshot is observed state, not a deployable configuration candidate."
    postDeploymentValidation:
      evidenceState: "blocked"
      resultState: "not-run"
      executionOutcome: "missing-deployment-timeout"
      note: "The recipe-declared components are absent, so expected resource health cannot be evaluated."
    hardwareRuntimeProof:
      evidenceState: "not-run"
      resultState: "not-run"
      note: "No RDMA transfer, GPU workload, model request, or performance test ran for these sanitized fixtures."
  retention:
    acceptedContextSha256: "528316f6386420c9d591d1a8e0e5a092055f3fee512d2fde9796499007be5c54"
    localFiles:
      - "tests/fixtures/config-assessment/aicr-snapshot-baseline.yaml"
      - "tests/fixtures/config-assessment/aicr-snapshot-target.yaml"
      - "data/aicr-snapshot-review/review.yaml"
    optionalOci:
      artifactType: "application/vnd.confighub.config-review.v1"
      contents:
        - "baseline.yaml"
        - "target.yaml"
        - "profiles.yaml"
        - "review.yaml"
      note: "The optional local OCI keeps the observations and interpretation together. It is a review artifact, not deployable Kubernetes configuration."
    configHub:
      status: "not-run"
      purpose: "Keep the two observations and their accepted interpretation as non-deployable Units for later comparison."
      commands:
        - "cub space create aicr-node-review --component aicr-node-state --variant reviewed --stage Review"
        - "cub unit create --space aicr-node-review --provider None baseline-snapshot tests/fixtures/config-assessment/aicr-snapshot-baseline.yaml"
        - "cub unit create --space aicr-node-review --provider None target-snapshot tests/fixtures/config-assessment/aicr-snapshot-target.yaml"
        - "cub unit create --space aicr-node-review --provider None snapshot-review data/aicr-snapshot-review/review.yaml"
status:
  result: "reviewed"
  decision: "The observed differences are compatible with the selected profiles. The target would have 2 findings if it were assigned the baseline profile."
  limits:
    - "The snapshots are sanitized fixtures, not fresh measurements from a live GPU cluster."
    - "The deployment profiles demonstrate provider-owned intent and are not NVIDIA AICR leaves or production hardware advice."
    - "No configuration was generated, deployed, validated after deployment, or exercised on GPU hardware."