Orchestrator registration always rejected with orchestrator_not_routable - requirements undocumented

[saurabhp1801]

Setup: hotkey 5GbhZBYMunT3Rv7Bd1twYjHPNX6v7af4Zi4LMqJJLJgBLzcW (UID 254, netuid 105), repo v0.2.1, Python 3.10, mainnet endpoints from docs.

What works: /auth/challenge + /auth/verify succeed (API key issued), NATS control connects to tls://orch-gateway.b1m.ai:4222, worker registration works (worker connects to our gateway).

What fails: every register message returns register_error: orchestrator_not_routable.

Evidence it is not a reachability problem:

• Advertised url/gateway_url tested as both http://212.43.154.172:8000 and https://212-43-154-172.sslip.io (valid TLS cert)
• GET /health returns 200 from the public internet on both; WebSocket upgrade on /ws/... responds correctly
• Tested regions EU/europe/north-america, uid set and unset - identical rejection
• Across 5,000+ registration attempts, access logs (direct and via reverse proxy) show zero inbound HTTP requests from BeamCore to the advertised URL - the check never contacts the orchestrator

Questions: What does the routability check validate? Are new orchestrator registrations open, or is onboarding gated/allowlisted? If gated, what is the approval process?

Client-side bugs found while debugging (separate but related):

1. SubnetCoreClient.start_polling() is awaited during FastAPI lifespan startup, before uvicorn binds the port - any live probe at first registration would always fail; a fresh install can never register if the check probes synchronously.
2. Field(env=...) names in neurons/orchestrator/core/config.py are silently ignored by pydantic-settings v2 - e.g. documented ORCHESTRATOR_WORKER_GATEWAY_URL does nothing; only WORKER_GATEWAY_URL (the uppercased field nameworks. Same forORCHESTRATOR_UIDvsUID`.