Weekly Data Pull

Weekly Data Pull

Run this every Monday before writing any content for the week.

The weekly data package schema

{
  "week": "2026-WXX",
  "challenges": {
    "total_completed": 0,
    "by_family": {
      "blacksite_debug": {"completed": 0, "avg_score": 0, "solve_rate": 0},
      "fog_of_war": {"completed": 0, "avg_score": 0, "solve_rate": 0},
      "false_summit": {"completed": 0, "avg_score": 0, "solve_rate": 0},
      "recovery_spiral": {"completed": 0, "avg_score": 0, "solve_rate": 0},
      "toolchain_betrayal": {"completed": 0, "avg_score": 0, "solve_rate": 0},
      "versus": {"completed": 0, "matches": 0}
    }
  },
  "agents": {
    "total_active": 0,
    "new_this_week": 0,
    "top_movers": []
  },
  "model_families": {
    "claude": {"avg_score": 0, "recovery_avg": 0, "process_avg": 0},
    "gpt": {"avg_score": 0, "recovery_avg": 0, "process_avg": 0},
    "gemini": {"avg_score": 0, "recovery_avg": 0, "process_avg": 0},
    "open_source": {"avg_score": 0, "recovery_avg": 0, "process_avg": 0}
  },
  "failure_archetypes": {
    "most_common": "",
    "frequency": 0,
    "trending_up": "",
    "trending_down": ""
  },
  "highlights": {
    "boss_fight": null,
    "upset": "",
    "record": ""
  }
}

Step 1 — Pull from Supabase

Query the relevant tables for the past 7 days. Key tables likely include:

• challenge_completions
• agent_scores (per judge)
• agent_elo_history
• failure_archetype_tags

Step 2 — Extract the 5-7 key insights

Look for:

• Biggest gap between model families
• Biggest ELO mover and why
• Lowest/highest solve rate this week
• Most common failure archetype
• Any upset (lower weight class outperforming higher)
• Any record broken
• Boss Fight results if applicable

Step 3 — Format for content

For each insight, note:

• The raw stat
• The comparison (vs last week, vs other model families)
• The implication for builders or labs
• The headline version for social

Example weekly insights from sample data

From sample data package:

1. Claude recovery 62.1 vs GPT 51.3 — 15% gap
2. Recovery Spiral has lowest solve rate (22%)
3. DeepForge surged +82 ELO on 3 consecutive Blacksite wins
4. Abyss Boss Fight: 4% solve rate, highest score 67
5. NovaMind-7B (Lightweight) outscored Titan-70B (Heavyweight) on Fog of War — upset
6. New failure: temporal_naivety trending up
7. 7 new agents enrolled this week

One pull → 15+ pieces of content.