{
  "schema": "skillreducer/research-table/1",
  "id": "smaller-model",
  "study": "smaller-model",
  "title": "Cheaper on a smaller model, per skill",
  "evidence_class": "measured-local",
  "date": "2026-09-20",
  "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590",
  "source_documents": [
    {
      "path": "docs/experiments/DETECTION-2026-09-19.md",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590",
      "committed": "2026-09-20"
    }
  ],
  "n_meaning": "Runs on the smaller model behind the row: comparable tasks times three repetitions",
  "columns": [
    {
      "name": "skill",
      "description": "Skill slug in the AirMarket catalog"
    },
    {
      "name": "downgradeable",
      "description": "Quality held on every comparable task and cheaper"
    },
    {
      "name": "quality_held_all",
      "description": "The smaller model matched the default model within 0.05 on every comparable task"
    },
    {
      "name": "cheaper",
      "description": "Cost ratio below 1"
    },
    {
      "name": "skill_lifts_haiku",
      "description": "The skill lifted the smaller model above no skill on some task"
    },
    {
      "name": "cost_ratio",
      "description": "Cost with the skill on the smaller model over cost on the default model"
    },
    {
      "name": "tasks",
      "description": "Tasks measured"
    },
    {
      "name": "tasks_comparable",
      "description": "Tasks with a comparable default-model run"
    },
    {
      "name": "tasks_held",
      "description": "Comparable tasks where quality held"
    },
    {
      "name": "n",
      "description": "Runs on the smaller model behind the row: comparable tasks times three repetitions"
    },
    {
      "name": "evidence_class",
      "description": "Evidence class of the figure (SPEC.md section 20)"
    },
    {
      "name": "date",
      "description": "UTC day the study's last measured run landed"
    },
    {
      "name": "commit",
      "description": "Commit the figures come from: the committed data when the study has some, otherwise the latest commit of the document that reports it"
    }
  ],
  "rows": [
    {
      "skill": "web-quality-audit",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.408,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "unity-script-roles",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.433,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "typescript-server",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.536,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "napkin",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.541,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "landing-page-conversion-audit",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.573,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "compose-modifier-and-layout-style",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.6,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "llm-council",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": false,
      "cost_ratio": 0.623,
      "tasks": 2,
      "tasks_comparable": 2,
      "tasks_held": 2,
      "n": 6,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "hermes-agent-skill-authoring",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.656,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "catchup",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.658,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "cost-efficiency-analyzer",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.748,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "eol-message",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.751,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "unity-primetween",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.753,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "optimize-simplicite-logs",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.776,
      "tasks": 2,
      "tasks_comparable": 2,
      "tasks_held": 2,
      "n": 6,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "geo-fundamentals",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.799,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "caveman-commit",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.842,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "research-agent",
      "downgradeable": true,
      "quality_held_all": true,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.94,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "offensive-reporting",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": true,
      "skill_lifts_haiku": false,
      "cost_ratio": 0.216,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 0,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "debugging-network-issues",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": true,
      "skill_lifts_haiku": false,
      "cost_ratio": 0.376,
      "tasks": 2,
      "tasks_comparable": 2,
      "tasks_held": 0,
      "n": 6,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "feedback",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.407,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 0,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "product-vision",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.544,
      "tasks": 2,
      "tasks_comparable": 2,
      "tasks_held": 1,
      "n": 6,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "wiki-lint",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": true,
      "skill_lifts_haiku": false,
      "cost_ratio": 0.587,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 0,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "modeling-product-usage-metrics",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.652,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 0,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "data-analyst",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": true,
      "skill_lifts_haiku": false,
      "cost_ratio": 0.714,
      "tasks": 2,
      "tasks_comparable": 2,
      "tasks_held": 1,
      "n": 6,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "ads-plan",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.859,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 0,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "n8n-error-handling",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": true,
      "skill_lifts_haiku": true,
      "cost_ratio": 0.979,
      "tasks": 2,
      "tasks_comparable": 2,
      "tasks_held": 1,
      "n": 6,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "triton-kernel",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": false,
      "skill_lifts_haiku": true,
      "cost_ratio": 1.007,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 0,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "event-store-design",
      "downgradeable": false,
      "quality_held_all": true,
      "cheaper": false,
      "skill_lifts_haiku": true,
      "cost_ratio": 1.02,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "vision-sft",
      "downgradeable": false,
      "quality_held_all": true,
      "cheaper": false,
      "skill_lifts_haiku": true,
      "cost_ratio": 1.036,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "latent-briefing",
      "downgradeable": false,
      "quality_held_all": true,
      "cheaper": false,
      "skill_lifts_haiku": true,
      "cost_ratio": 1.094,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "dotnet-local-tools",
      "downgradeable": false,
      "quality_held_all": true,
      "cheaper": false,
      "skill_lifts_haiku": true,
      "cost_ratio": 1.256,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "alphagbm-tepper-signal",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": false,
      "skill_lifts_haiku": true,
      "cost_ratio": 1.466,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 0,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "architecture-designer",
      "downgradeable": false,
      "quality_held_all": true,
      "cheaper": false,
      "skill_lifts_haiku": true,
      "cost_ratio": 1.524,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 1,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "friendly",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": false,
      "skill_lifts_haiku": true,
      "cost_ratio": 2.181,
      "tasks": 2,
      "tasks_comparable": 1,
      "tasks_held": 0,
      "n": 3,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    },
    {
      "skill": "terracotta",
      "downgradeable": false,
      "quality_held_all": false,
      "cheaper": false,
      "skill_lifts_haiku": true,
      "cost_ratio": 2.545,
      "tasks": 2,
      "tasks_comparable": 2,
      "tasks_held": 1,
      "n": 6,
      "evidence_class": "measured-local",
      "date": "2026-09-20",
      "commit": "80e4a13fb96ebb96d8efff8690a8c77215657590"
    }
  ]
}
