{
  "version": "cheapifier-calculator-scenarios-1",
  "execution": {
    "source_report_sha256": "c252393dc247a62c5e33493649aa2d0eab4c4caafcc4ac47c5992178a05043c9",
    "runtime": "ONNX Runtime 1.23.2",
    "compressor": "LLMLingua-2 XLM-RoBERTa token classification",
    "keep_rate": 0.95,
    "tokenizer": "o200k_base",
    "model_sha256": "0fddee8b1e0de33216b534370b884481f033f8d5225e73ac88f38bb8f829caff",
    "tokenizer_sha256": "d5e1e0abdc17f93ca3c0c02a75d90c7c1f4ef556dd1a4e5c12d4ad4e8812e112"
  },
  "totals": {
    "scenarios": 6,
    "cases": 93,
    "checks": 3447,
    "passes": 3447,
    "semantic_probe_checks": 867,
    "semantic_probe_passes": 867
  },
  "stage_order": ["baseline", "context-organization", "agent-instructions", "context-budget", "tool-routing", "guarded-compression", "duplicate-call-suppression", "tool-output-reduction"],
  "scenarios": [
    { "id": "balanced", "cases": 36, "stage_tokens": [245228, 247164, 247164, 236688, 145779, 145343, 145343, 103799], "reduction_percent": 57.67245175917921, "checks": 1272, "passes": 1272, "semantic_probe_checks": 330, "semantic_probe_passes": 330, "routing_cost_reduction_percent": 71.35058789637658 },
    { "id": "engineering", "cases": 12, "stage_tokens": [209353, 209974, 209974, 204250, 151402, 151267, 78883, 34315], "reduction_percent": 83.60902399296882, "checks": 453, "passes": 453, "semantic_probe_checks": 99, "semantic_probe_passes": 99, "routing_cost_reduction_percent": 84.73475577623459 },
    { "id": "support", "cases": 9, "stage_tokens": [80033, 80522, 80522, 77090, 51227, 51098, 28520, 16037], "reduction_percent": 79.96201566853674, "checks": 324, "passes": 324, "semantic_probe_checks": 90, "semantic_probe_passes": 90, "routing_cost_reduction_percent": 95.0 },
    { "id": "regulated", "cases": 9, "stage_tokens": [39630, 40122, 40122, 35829, 16635, 16511, 16511, 15098], "reduction_percent": 61.90259904113046, "checks": 318, "passes": 318, "semantic_probe_checks": 90, "semantic_probe_passes": 90, "routing_cost_reduction_percent": 27.055744542987902 },
    { "id": "research", "cases": 9, "stage_tokens": [90933, 91405, 91405, 87925, 65101, 64963, 44905, 26323], "reduction_percent": 71.05231324161745, "checks": 330, "passes": 330, "semantic_probe_checks": 96, "semantic_probe_passes": 96, "routing_cost_reduction_percent": 88.00545740848594 },
    { "id": "automation", "cases": 18, "stage_tokens": [455850, 456759, 456759, 444846, 354363, 354140, 147662, 51398], "reduction_percent": 88.72479982450368, "checks": 750, "passes": 750, "semantic_probe_checks": 162, "semantic_probe_passes": 162, "routing_cost_reduction_percent": 70.75903589488519 }
  ],
  "calculator_rule": "For the selected workload, add its measured context-organization overhead per qualified interaction, then apply every sequential stage ratio from that workload's locally recorded token path to deployed input tokens. Leave undeployed input and model output unchanged. Compare the result with the original pre-Kulla baseline. Company size, AI maturity, and consumption scale the measured path; they do not substitute a different reduction coefficient."
}