{
  "schema_version": "operator-workflow/v1",
  "request": {
    "action": "review corrected C1 64K context qualification plan",
    "model": "bottlecapai/ThinkingCap-Qwen3.8-27B-NVFP4A4-AWQ",
    "hardware": "RTX 5090",
    "execution_scope": "planning and source-only validation; no endpoint, GPU, lifecycle, route, or promotion action"
  },
  "gate_state": "ready_for_authorized_preflight_and_submit",
  "targets": [
    "C1 64K BF16-KV text-context qualification",
    "full-vision qualification under the same 64K serve configuration"
  ],
  "tools_used": [
    "native benchmark job-spec validator",
    "native context selection validator",
    "read-only source inspection"
  ],
  "artifacts": {
    "private_job_spec": "private operator evidence/context-64k-plan.private.json",
    "parameter_validation": {
      "profile": "smoke",
      "advertised_context": 65536,
      "token_buckets": [
        8192,
        62464
      ],
      "output_headroom_tokens": 2048,
      "template_and_transport_margin_tokens": 1024,
      "near_limit_accounting": "62464 prompt target + 2048 output reserve + 1024 margin = 65536",
      "case_ids": [
        "native-needle",
        "native-order",
        "native-distractor",
        "native-identifiers",
        "native-crosslink"
      ],
      "positions": [
        0.1,
        0.5,
        0.9
      ],
      "repetitions": 2,
      "planned_observations": 60,
      "timeout_s": 7200
    },
    "source_provenance": [
      {
        "path": "anvil_serving/benchmarking/jobs.py",
        "purpose": "portable job-spec schema and bounded JSON validation"
      },
      {
        "path": "anvil_serving/benchmarking/suite_runner.py",
        "purpose": "supported context parameter names, capacity check, request execution, and result scoring"
      },
      {
        "path": "anvil_serving/benchmarking/context.py",
        "purpose": "native deterministic context cases and exact-answer scoring"
      },
      {
        "path": "anvil_serving/benchmarking/requests.py",
        "purpose": "1024-token default context safety margin"
      },
      {
        "path": "anvil_serving/_benchmark_profiles/smoke.json",
        "purpose": "content-addressed smoke profile and scoring policy"
      }
    ],
    "findings_summary": [
      {
        "severity": "blocking",
        "classification": "confirmed",
        "finding": "The preserved earlier smoke specification uses unsupported advisory parameter names and therefore selects only the profile-default 8192- and 32768-token buckets. It cannot support a 64K claim."
      },
      {
        "severity": "high",
        "classification": "confirmed",
        "finding": "A completed job is insufficient. Qualification requires every one of the 60 context observations to pass, observed prompt-token evidence near the 62464 target, allowed finish reasons, and a passing degradation curve."
      },
      {
        "severity": "medium",
        "classification": "confirmed",
        "finding": "The existing multimodal runner can separately prove image, OCR, chart, screenshot, spatial, and two-image correctness under the 64K configuration."
      },
      {
        "severity": "medium",
        "classification": "confirmed",
        "finding": "The existing multimodal runner limits manifest prompts to 8192 characters, so it cannot prove a combined near-limit text-plus-image request. Any claim must distinguish near-limit text context from full vision under the same configuration."
      }
    ]
  },
  "advisory_priors": [
    "The existing user authorization covers candidate lifecycle, testing, and optimization on the local RTX 5090.",
    "This planning record does not authorize promotion or route changes.",
    "No context or multimodal qualification result exists until the managed jobs run and their retained evidence passes independent review."
  ],
  "recommendation": "Use the corrected private job specification for managed preflight and submit. Retain the earlier smoke specification as superseded evidence. Publish only separate near-limit text and full-vision-under-64K claims unless a future managed harness supports combined long-context multimodal cases.",
  "human_gate_required": true,
  "promoted": false
}