Wiring Probe Results — Complete

Status: COMPLETE (all 5 probes passed) Agent: opencode/ext-agent (sandshrew) Timestamp UTC: 2026-05-11T23:40:00Z

Probe Results

Probe 1: Network Baseline ✅

Pi→RG ping: 3-5ms steady, 0% loss

Probe 2: Python on Pi ✅

Python 3.13.5, requests 2.32.3, 3.1GB RAM available
Installed: langgraph 1.1.10, fastapi 0.136.1, uvicorn 0.46.0
Venued in: /home/mehdifarah/game-surface-venv/

Probe 3: HTTP Roundtrip (RG → Pi) ✅

5 pings to POST /ping: 14-22ms, AVG 16ms

Probe 4: State Fetch (RG → Pi) ✅

GET /state: 18ms, payload 5.9 KB (36 nodes + 3 units)
State is tiny. Serialization + transfer is negligible.

Probe 5: Invoke (RG → Pi) ✅

POST /invoke with 0.8s mock agent: 839ms total
HTTP overhead: ~39ms. Agent call dominates. Real LLM calls: 1-5s expected.

Bonus: RG Tailscale Version Cap — DEBUNKED ✅

Tailscale 1.96.4 binary runs on RG. Not hardware-capped.
Just needs: update symlinks in custom.sh + restart tailscaled.

Bonus: Tailscale SSH from Mac to Pi ✅

ssh mehdifarah@relik-pi4 works. No password. No key prompt.

Wiring Decisions Locked

Decision Answer Evidence
Polling or streaming? Polling at 100ms 16ms HTTP + 18ms state fetch = 34ms roundtrip. Polling imperceptible.
HTTP framework? FastAPI + uvicorn Installed, tested, working on Pi.
State payload concern? None 5.9 KB at 36 nodes. <50 KB even with full history.
Invoke latency acceptable? Yes 16ms HTTP + agent call time. UI-responsiveness dominated by agent, not wire.
RG can run latest Tailscale? Yes 1.96.4 binary runs. Not capped.
Tailscale SSH works Mac→Pi? Yes ssh mehdifarah@relik-pi4 confirmed.

Best Case Target Config

Pi (relik-pi4)

RG (knulli-1)

Mac

What a Future Session Needs to Do

  1. Update RG Tailscale from 1.76.1 → 1.96.4 (binary works, just swap symlinks)
  2. Test tailscale up --ssh on RG after upgrade
  3. Test MagicDNS on RG after upgrade
  4. Run tailscale up --ssh on Mac (if not already enabled)
  5. Kill probe server on Pi when done (pkill -f python3)