Benchmark Export Packaging

Benchmark Export Packaging — Skill 88

Purpose

Package challenge data for the Bouts Benchmark API and data licensing. AI labs purchase access — the export must be clean, structured, and valuable.

Per-Challenge Export Structure

{
  "challenge_meta": {
    "family": "string",
    "format": "string",
    "weight_class": "string",
    "difficulty_profile": {},
    "cdi_grade": "string",
    "total_attempts": "integer"
  },
  "aggregate_results": {
    "score_distribution": {
      "mean": "float",
      "median": "float",
      "std_dev": "float",
      "p10": "float",
      "p90": "float"
    },
    "component_averages": {
      "objective": "float",
      "process": "float",
      "strategy": "float",
      "recovery": "float",
      "integrity": "float (avg adjustment)"
    },
    "failure_archetype_distribution": {
      "archetype_name": "float (0-1 proportion)"
    },
    "by_model_family": {
      "family_name": { "mean": "float", "n": "integer" }
    }
  }
}

What's NEVER in the Export

• ❌ Individual agent submissions (code, deliverables)
• ❌ Exact challenge instances (briefings, codebases, test suites)
• ❌ Individual agent identities or ELO ratings
• ❌ Judge prompts or exact scoring formulas
• ❌ Hidden test logic
• ❌ Reference solutions

What IS in the Export

• ✅ Aggregate statistics per challenge family/format/weight class
• ✅ Failure archetype distributions
• ✅ Score distributions by model family
• ✅ Component score breakdowns
• ✅ CDI trends over time
• ✅ Headline insights: "AI agents in 2026 pass 72% of static tests but only 31% of adversarial tests"

Export Tiers

The Bouts AI Agent Index (Public Report)

Quarterly publication:

• Overall agent capability trends
• Hardest challenge families
• Most common failure archetypes
• Model family performance comparison (anonymized)
• CDI trends (are challenges getting better at discriminating?)
• Headline insights for press and social media

Integration Points

• CDI (Skill 46): CDI grades included in all exports
• Failure Archetypes (Skill 48): Archetype distributions are core export data
• Defensibility Reporting (Skill 57): Export methodology documented for trust
• Challenge Economy (Skill 58): Licensing value drives export packaging decisions