EXTERNAL TEST LAB

Test network changes
before they ship

A test environment for protocol upgrades, DevShards, inference flows, integrations and operational changes – before governance decisions or production rollout

Overview dashboard ↗Network dashboard ↗Inference dashboard ↗Chain dashboard ↗Components dashboard ↗

OPERATE A HOST

How to Join node

git clone https://github.com/paranjko/external-test-lab.git
alias gdc="$PWD/external-test-lab/net-deployment-runbook/gdc.sh"
gdc host join --model=<MODEL_ID> --public-host <IP_or_DOMAIN> <ssh-alias>

Minimum Host requirements For more details, read Community DevNet runbook/JOIN: add a Host

Component	Minimum
OS	Ubuntu 22.04, 24.04, or 26.04 (amd64)
GPU	NVIDIA GPU with 16 GB VRAM
Driver	NVIDIA R580 or newer
CUDA	GPU works in a CUDA 12.8 Docker container
CPU	8 threads or more (practical baseline)
Memory	32 GB RAM or more (practical baseline)
Storage	50 GiB free required, with 100–150 GB NVMe recommended
Network	Stable internet with a stable public IPv4 address or a DNS name that resolves to it
Ingress	TCP 80, 443, 5000, and the configured SSH port, plus UDP 443
Uptime	Always on with sleep and address changes disabled

Choose MODEL_ID from the live model registry, JOIN rejects an ID absent from the registry before Host preparation

Chain ID–

Allocated models

Approved DevShards

Latest height–

Largest voting share

CHAIN-RECORDED EPOCH TRANSITION

Previous epoch –

Waiting for a finalized epoch

PoC evidence–Waiting for a chain observation

cPoC–Waiting for a chain observation

Consensus transition–Waiting for a validator-set observation

Inspect chain evidence

PoC evidence

Participant	Committed artifacts	Validator attestations

Consensus set at activation

Participant	Previous power	Effective power	Observed change

PUBLIC DEVSHARD OBSERVATIONS

DevShard execution

Waiting for a network observation

Public endpoints–Health and runtime inventory

Runtime inventory–Compared with the chain-approved catalog

Host-local sessions–Not a fleet or GPU-utilisation measure

Inspect public endpoint observations

Public endpoint	Running runtimes	Host-local sessions

LIVE DEVNET

Participant status

Waiting for status refresh

OBSERVED NETWORK HOSTS

Network node map

Validating
Active
Wait

LIVE GATEWAY LOAD

Inference traffic

Loading gateway metrics…

Active requests–currently in flight

Input tokens–currently in flight

Successful requests–completed since gateway restart

Rate-limited requests–rejected since gateway restart

Network capacity–available model weight

Gateway ACHECKINGwaiting for read-only route status

Gateway BCHECKINGwaiting for read-only route status

Gateway trafficCHECKING

– active now– successful since restart– rate-limited since restart