Contributing Compute

Any OpenCastor robot can contribute idle compute to the harness research pipeline.

Prerequisites

• OpenCastor runtime installed and running
• Ollama installed with gemma3:1b (for real eval) — or simulated eval works without it
• Robot registered with fleet registry (has an RRN)

Enable

castor contribute start

Or in config:

agent:
  contribute:
    enabled: true
    idle_after_minutes: 15
    projects:
      - harness_research

What your robot does

1. Waits until idle for idle_after_minutes
2. Fetches next candidate from harness_eval_queue in Firestore
3. Runs OHB-1 benchmark (30 tasks, ~5–15 min depending on hardware)
4. Submits score to harness_eval_results with your RRN
5. Returns to idle standby

Run types

Choose community (public, credit-eligible) or personal (private, no credits):

castor contribute start --run-type community   # default
castor contribute start --run-type personal

Credit attribution

Your RRN is stored with every result you submit:

{
  "candidate_id": "config_abc123",
  "evaluated_by": "RRN-000000000001",
  "score": 0.6123,
  "evaluated_at": "2026-03-21T..."
}

If a config you evaluated becomes the fleet champion, you receive a champion credit flag.

Credit share formula:

credit_share_pct = (your_eval_count / total_eval_count) * 100

Applying champion configs

Champions are never auto-applied. When a new champion is found, you receive a notification in the app. You choose when to apply it — to one robot or the fleet.

castor harness apply-champion           # apply to this robot
castor harness apply-champion --fleet   # apply to all robots

Enable auto-apply for future champions (opt-in):

castor contribute auto-apply --enable