health

Agent Learning Health Dashboard

You are presenting a comprehensive view of how the agent learning system is performing — combining Neverland metrics, nomination stats, and notebook growth into one actionable dashboard.

Step 1: Neverland Metrics

paradigm_ambient_health({})

Step 2: Nomination Stats Per Agent

For each active agent from the Neverland response, get detailed stats:

paradigm_ambient_learn({ agent: "<agent-id>", dry_run: true })

This returns stats without adjusting thresholds.

Step 3: Agent Roster Context

paradigm_agent_list({ scope: "all" })

Step 4: Present Dashboard

Combine all data into a single dashboard:

Agent Learning Health
=====================

Overall: {healthStatus}
  Avg accept rate: {avgAcceptRate}% | Target: >70%
  Avg threshold: {avgThreshold} | Range: 0.0–1.0
  Total expertise: {totalExpertise} symbols
  Total notebooks: {totalNotebooks} entries
  Total transferable: {totalTransferable} patterns

Per-Agent Breakdown
-------------------

  architect {✓ active | ⏸ benched}
    Accept: {rate}% ({accepted}/{total}) | Threshold: {threshold}
    Expertise: {count} symbols | Notebooks: {count}
    Direction: {↑ improving | → stable | ↓ declining}
    {recommendation if any}

  builder {✓ active | ⏸ benched}
    Accept: {rate}% | Threshold: {threshold}
    ...

  security {✓ active | ⏸ benched}
    Accept: {rate}% | Threshold: {threshold}
    ...

  (repeat for all agents)

Neverland Progress
------------------

  cold-start [====                ] Sessions 1-3
  accumulating   [========            ] Sessions 3-5
  calibrating    [============        ] Sessions 5-8
  mature         [====================] Sessions 8+

  Current: {healthStatus} ← you are here

  Target milestones:
    □ Agents have divergent expertise scores (by session 5)
    □ Maestro routes to right agent >80% (by session 10)
    □ Agent acceptance rate >70% (by session 10)
    □ Cross-project patterns appear in enrichment

Step 5: Recommendations

Based on the health data:

If cold-start:

• "Run more sessions with orchestration (paradigm_orchestrate_inline) to generate events and nominations"
• "Use /paradigm:teach to give agents initial domain knowledge"

If accumulating (accept rate <50%):

• "Agents are generating nominations but most are being dismissed"
• "Check dismiss reasons — if briefs are too generic, the Teacher Model will fix this over time"
• "Consider /paradigm:teach for agents with 0% accept rate"

If calibrating (accept rate 50-70%):

• "Learning loop is working — thresholds are adjusting"
• "Check which agents have notebook entries — those with notebooks should produce better nominations"
• "Monitor for agents trending down — may need teaching or benching"

If mature (accept rate >70%):

• "The team is well-calibrated"
• "Focus on cross-project transfer — patterns learned here should appear in other projects"
• "Consider publishing successful agent profiles to nevr.land"