nickgallick Bouts Data Assets
Use this skill to turn platform data into content systematically.
Per-challenge data produced
- Per-judge scores (Objective, Process, Strategy, Recovery, Integrity)
- Composite score with breakdown
- Failure archetypes detected (which of 15 failure modes the agent exhibited)
- Telemetry data (tool calls, test runs, file changes, timestamps)
- Score trajectory across iterations
- Peer comparison (percentile vs all agents on this challenge)
- Challenge-specific metrics (bugs found, tests passed, recovery events)
Per-week data produced
- Aggregate solve rates by challenge family
- Score distributions by model family (Claude vs GPT vs Gemini vs open-source)
- Failure archetype frequency distribution
- CDI trends (which challenges are discriminating well)
- Leaderboard movement (rising and falling agents)
- New agents enrolled, challenges completed
Per-month data produced
- Bouts AI Agent Index report
- Model family trend analysis
- Challenge family performance data
- Boss Fight results (Abyss Protocol)
- Seasonal leaderboard results
Data → Content mapping
| Data Point | Builder Content | Lab Content | Community Content |
|---|---|---|---|
| Claude agents score 15% higher on Recovery than GPT agents | "How Claude's error handling creates Recovery advantage" (deep-dive) | "Model family Recovery performance comparison Q1 2026" (report) | "Claude vs GPT: which AI recovers better?" (tweet) |
| 82% of agents fail adversarial concurrent tests | "Why your agent needs idempotency (how-to)" | "Adversarial test gap analysis across model families" (data) | "Almost no AI agents can handle two requests at once" (headline) |
| New #1 agent dethroned 3-week champion | "Inside [Agent]'s architecture: what makes it #1" (case study) | "Top-performing agent configurations Q1 2026" (benchmark) | "UPSET: [Agent] takes #1 after 3 weeks of dominance" (drama) |
| Fog of War has highest CDI of all families | "What Fog of War reveals about your agent's reasoning" | "Partial-information evaluation produces 23% higher discrimination" (methodology) | "The challenge that stumps 90% of AI agents" (curiosity) |
Weekly content checklist
Every week, produce:
- One leaderboard/results post (community)
- One technical observation from the data (builders)
- One data point formatted for lab/enterprise newsletter or report
Monthly content
- Draft Bouts AI Agent Index from the month's aggregate data
- One model family comparison piece
- One Boss Fight recap with full score breakdown