{
  "baseline": {
    "calls": 9,
    "latency_ms": 22030,
    "tokens": 72800,
    "usd": 0.763,
    "verified_success": true
  },
  "baseline_actions": [
    {
      "name": "inspect the failing assertion",
      "output": "expected 80.0",
      "stage": "Diagnose",
      "tokens": 1200
    },
    {
      "name": "scan the entire repository",
      "output": "README.md, pricing.py, test_pricing.py",
      "stage": "Diagnose",
      "tokens": 9000
    },
    {
      "name": "ask two parallel reviewers",
      "output": "reviewer-a: inspect arithmetic; reviewer-b: inspect type contract",
      "stage": "Diagnose",
      "tokens": 14000
    },
    {
      "name": "apply the targeted one-line patch",
      "output": "replaced subtraction with percentage multiplication",
      "stage": "Fix",
      "tokens": 2400
    },
    {
      "name": "rewrite the complete pricing module",
      "output": "rewrote the complete pricing module",
      "stage": "Fix",
      "tokens": 12000
    },
    {
      "name": "ask a frontier model for an alternative patch",
      "output": "alternative patch agreed with the targeted correction",
      "stage": "Fix",
      "tokens": 18000
    },
    {
      "name": "run the targeted verifier",
      "output": "targeted verifier passed",
      "stage": "Verify",
      "tokens": 700
    },
    {
      "name": "run the full test suite",
      "output": "full suite passed",
      "stage": "Verify",
      "tokens": 4500
    },
    {
      "name": "request a premium model audit",
      "output": "premium audit found no remaining defect",
      "stage": "Verify",
      "tokens": 11000
    }
  ],
  "defect": {
    "after": "return total * (1 - rate)",
    "before": "return total - rate",
    "verifier": "apply_discount(100.0, 0.20) == 80.0"
  },
  "demo": "marginal-killer-demo-v1",
  "disclaimer": "Deterministic functional demonstration using declared action-cost estimates; not provider telemetry, not a production benchmark, and not a claim about every agent workload.",
  "initial_verified_success": false,
  "marginal": {
    "calls": 3,
    "latency_ms": 1230,
    "tokens": 4300,
    "usd": 0.026,
    "verified_success": true
  },
  "marginal_actions": [
    {
      "name": "inspect the failing assertion",
      "output": "expected 80.0",
      "stage": "Diagnose"
    },
    {
      "name": "apply the targeted one-line patch",
      "output": "replaced subtraction with percentage multiplication",
      "stage": "Fix"
    },
    {
      "name": "run the targeted verifier",
      "output": "targeted verifier passed",
      "stage": "Verify"
    }
  ],
  "savings": {
    "calls_percent": 66.67,
    "latency_percent": 94.42,
    "tokens_percent": 94.09,
    "usd_percent": 96.59
  },
  "scenario": "Fix a percentage-discount bug in a deterministic Python repository",
  "stages": [
    {
      "candidates": [
        {
          "allowed": true,
          "expected_gain": 0.22,
          "kind": "research",
          "latency_ms": 350,
          "name": "inspect the failing assertion",
          "reason": "approved: marginal ROI 7.166",
          "score": 0.1893,
          "tokens": 1200,
          "usd": 0.006
        },
        {
          "allowed": false,
          "expected_gain": 0.05,
          "kind": "research",
          "latency_ms": 2200,
          "name": "scan the entire repository",
          "reason": "rejected: marginal ROI 0.218 below 1.000",
          "score": -0.1794,
          "tokens": 9000,
          "usd": 0.045
        },
        {
          "allowed": false,
          "expected_gain": 0.04,
          "kind": "review",
          "latency_ms": 4500,
          "name": "ask two parallel reviewers",
          "reason": "rejected: marginal ROI 0.098 below 1.000",
          "score": -0.369,
          "tokens": 14000,
          "usd": 0.12
        }
      ],
      "decision": "approved: marginal ROI 7.166",
      "selected": "inspect the failing assertion",
      "stage": "Diagnose"
    },
    {
      "candidates": [
        {
          "allowed": true,
          "expected_gain": 0.5,
          "kind": "generation",
          "latency_ms": 700,
          "name": "apply the targeted one-line patch",
          "reason": "approved: marginal ROI 7.418",
          "score": 0.4326,
          "tokens": 2400,
          "usd": 0.018
        },
        {
          "allowed": false,
          "expected_gain": 0.2,
          "kind": "generation",
          "latency_ms": 3200,
          "name": "rewrite the complete pricing module",
          "reason": "rejected: marginal ROI 0.561 below 1.000",
          "score": -0.1564,
          "tokens": 12000,
          "usd": 0.11
        },
        {
          "allowed": false,
          "expected_gain": 0.15,
          "kind": "generation",
          "latency_ms": 5500,
          "name": "ask a frontier model for an alternative patch",
          "reason": "rejected: marginal ROI 0.230 below 1.000",
          "score": -0.501,
          "tokens": 18000,
          "usd": 0.28
        }
      ],
      "decision": "approved: marginal ROI 7.418",
      "selected": "apply the targeted one-line patch",
      "stage": "Fix"
    },
    {
      "candidates": [
        {
          "allowed": true,
          "expected_gain": 0.35,
          "kind": "verification",
          "latency_ms": 180,
          "name": "run the targeted verifier",
          "reason": "approved: marginal ROI 21.394",
          "score": 0.33364,
          "tokens": 700,
          "usd": 0.002
        },
        {
          "allowed": false,
          "expected_gain": 0.08,
          "kind": "verification",
          "latency_ms": 1400,
          "name": "run the full test suite",
          "reason": "rejected: marginal ROI 0.763 below 1.000",
          "score": -0.0248,
          "tokens": 4500,
          "usd": 0.012
        },
        {
          "allowed": false,
          "expected_gain": 0.05,
          "kind": "verification",
          "latency_ms": 4000,
          "name": "request a premium model audit",
          "reason": "rejected: marginal ROI 0.126 below 1.000",
          "score": -0.348,
          "tokens": 11000,
          "usd": 0.17
        }
      ],
      "decision": "approved: marginal ROI 21.394",
      "selected": "run the targeted verifier",
      "stage": "Verify"
    }
  ]
}