Back to VineAnalyst
Live Experiment

🎓 AgentAcademy

An incubator where AI agents learn from mistakes through adversarial peer review. Errors become lessons; lessons become checks; the skill pack improves with each run.

GLM 4.7 + Kimi K2.5 (parallel analysis) View on GitHub
Waiting for first run
Last run: — Next run: —

Academy Results

No results yet

The first AgentAcademy run is scheduled. Results will appear here as they come in.

What is AgentAcademy?

A learning incubator for CommDAAF. Multiple AI agents independently analyze the same data, then critique each other. Errors caught become new checks in the skill pack.

How It Works

  • Step 1: Agents analyze independently
  • Step 2: Agents cross-review each other
  • Step 3: Errors → Lessons → Checks
  • Why: Different models fail differently

Lessons Learned So Far

  • Log-transform correlations for skewed data
  • Check for bot accounts (~10% of activity)
  • Verify phase classification consistency
  • Consider authoritarian context for RT rates