{
  "schema": "anvil-serving.recipe-feasibility-result/v1",
  "campaign": "128K minimum BF16 target-only scout",
  "equations": {
    "required_tokens": "sum(requirements.tokens)",
    "policy_available_vram": "physical_vram - sum(all vram_reserves)",
    "demand": "resident + concurrency*per_sequence + required_tokens*kv_token_multiplier*per_token",
    "relative_quality_loss": "1 - quality/reference_quality",
    "warm_e2e_gain": "1 - warm_e2e/no_spec_warm_e2e",
    "tasks_per_hour_ratio": "tasks_per_hour/reference_tasks_per_hour"
  },
  "required_tokens": {
    "min": 131072,
    "max": 131072
  },
  "physical_available_bytes": {
    "min": 34190917632,
    "max": 34190917632
  },
  "policy_available_bytes": {
    "min": 33117175808,
    "max": 33117175808
  },
  "candidates": [
    {
      "id": "128k-nospec-bf16-095",
      "classification": "unresolved",
      "reasons": [
        "context or VRAM bounds overlap or remain unbounded"
      ],
      "axes": {
        "context_limit": "pass",
        "measured_context": "not-measured",
        "measured_context_evidence_status": "not-supplied",
        "memory": "unknown",
        "correctness": {
          "status": "unknown",
          "missing_variables": [
            "deterministic_pass_rate"
          ]
        },
        "quality": {
          "status": "unknown",
          "missing_variables": [
            "quality_score",
            "reference_quality_score"
          ]
        },
        "warm_e2e_speed": {
          "status": "unknown",
          "missing_variables": [
            "warm_e2e_seconds",
            "no_spec_warm_e2e_seconds"
          ]
        },
        "tasks_per_hour": {
          "status": "unknown",
          "missing_variables": [
            "successful_tasks_per_hour",
            "reference_tasks_per_hour"
          ]
        }
      },
      "required_tokens": {
        "min": 131072,
        "max": 131072
      },
      "demand_bytes": {
        "min": 32357209866,
        "max": 35374424391
      },
      "physical_available_bytes": {
        "min": 34190917632,
        "max": 34190917632
      },
      "policy_available_bytes": {
        "min": 33117175808,
        "max": 33117175808
      },
      "physical_margin_bytes": {
        "min": -1183506759,
        "max": 1833707766
      },
      "policy_margin_bytes": {
        "min": -2257248583,
        "max": 759965942
      },
      "estimated_max_tokens_physical": {
        "min": 116624,
        "max": 159052
      },
      "estimated_max_tokens_policy": {
        "min": 103517,
        "max": 142668
      },
      "load_bearing_unknowns": [],
      "planning_bound_inputs": [
        "requirements.tokens.prompt",
        "requirements.tokens.output_reserve",
        "requirements.tokens.protocol_margin",
        "requirements.concurrency",
        "requirements.vram_reserves.free_headroom.variable",
        "candidates[0].runtime_context_limit_tokens",
        "candidates[0].resident_components.runtime_workspace_and_recurrent",
        "candidates[0].per_token_components.draft_and_padding",
        "candidates[0].kv_token_multiplier"
      ],
      "missing_evidence": [
        "deterministic_pass_rate",
        "measured_max_stable_context_tokens",
        "no_spec_warm_e2e_seconds",
        "quality_score",
        "reference_quality_score",
        "reference_tasks_per_hour",
        "successful_tasks_per_hour",
        "warm_e2e_seconds"
      ]
    }
  ],
  "variables": [
    {
      "path": "requirements.tokens.prompt",
      "status": "assumed",
      "unit": "tokens",
      "min": 125952,
      "max": 125952,
      "source": "64K request plan",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    },
    {
      "path": "requirements.tokens.output_reserve",
      "status": "assumed",
      "unit": "tokens",
      "min": 4096,
      "max": 4096,
      "source": "reasoning and visible output reserve",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    },
    {
      "path": "requirements.tokens.protocol_margin",
      "status": "assumed",
      "unit": "tokens",
      "min": 1024,
      "max": 1024,
      "source": "token calibration margin",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    },
    {
      "path": "requirements.concurrency",
      "status": "assumed",
      "unit": "sequences",
      "min": 1,
      "max": 1,
      "source": "single candidate C1",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    },
    {
      "path": "requirements.physical_vram_bytes",
      "status": "measured",
      "unit": "bytes",
      "min": 34190917632,
      "max": 34190917632,
      "source": "prior local inventory and current GPU identity",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    },
    {
      "path": "requirements.vram_reserves.free_headroom.variable",
      "status": "assumed",
      "unit": "bytes",
      "min": 1073741824,
      "max": 1073741824,
      "source": "128K retainedprofile minimum1GiB, lowerGPUfraction plusweightoffloadifneeded",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    },
    {
      "path": "candidates[0].runtime_context_limit_tokens",
      "status": "assumed",
      "unit": "tokens",
      "min": 131072,
      "max": 131072,
      "source": "candidateAPIcap",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    },
    {
      "path": "candidates[0].resident_components.runtime_workspace_and_recurrent",
      "status": "estimated",
      "unit": "bytes",
      "min": 858993459,
      "max": 1717986918,
      "source": "prior.87VRAMtarget minus22.14weights minus4.73KV ~.83GiB, widened for contextlayout",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    },
    {
      "path": "candidates[0].resident_components.target_vision",
      "status": "measured",
      "unit": "bytes",
      "min": 22908281815,
      "max": 22919019233,
      "source": "Prior no-spec weights21.34GiB, completevision; MTPdisabled",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    },
    {
      "path": "candidates[0].per_token_components.target_full_attention",
      "status": "confirmed",
      "unit": "bytes/token",
      "min": 65536,
      "max": 65536,
      "source": "16fullattention*K/V*4heads*256dim*2bytes",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    },
    {
      "path": "candidates[0].per_token_components.draft_and_padding",
      "status": "estimated",
      "unit": "bytes/token",
      "min": 0,
      "max": 16384,
      "source": "NoMTP; context-dependenthybridrecurrent/page overhead, boundedscout estimate anchored by earliernoMTP5.61GiB/79872tokens",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    },
    {
      "path": "candidates[0].kv_token_multiplier",
      "status": "assumed",
      "unit": "ratio",
      "min": 1,
      "max": 1,
      "source": "C1",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    },
    {
      "path": "requirements.thresholds.min_deterministic_pass_rate",
      "status": "assumed",
      "unit": "ratio",
      "min": 1,
      "max": 1,
      "source": "allhardgates",
      "observed_at": "2026-09-24T04:12:05.848923+00:00",
      "notes": null
    }
  ],
  "unresolved_variables": [],
  "promotion_authority": false
}
