Configuration

The validator reads configuration from environment variables (via .env). Defaults live in validator/config.py, shared/chain.py, and validator/main.py.

Identity & wallet

Variable	Default	Meaning
`NPA_NETUID`	`98`	Subnet netuid
`NPA_NETWORK`	`finney`	Subtensor network
`NPA_WALLET`	`default`	Bittensor wallet name
`NPA_HOTKEY`	`default`	Bittensor hotkey name
`NPA_BT_WALLET_DIR`	—	Override wallet path
`NPA_API_URL`	`https://api.neverplayalone.ai`	Backend API base URL

Evaluation

Variable	Default	Meaning
`NPA_MISSION_ID`	`resource_gathering`	Local mission fallback (the roster wins)
`NPA_TASKS_PER_ROUND`	`3`	Seeds per miner per round; score is the mean
`NPA_MAX_PARALLEL_AGENTS`	`4`	Concurrent npabench agent slots
`NPA_WORKSPACE_ROOT`	`/tmp/npa_validator`	Per-round workspace
`NPA_LOOP_POLL_SECONDS`	`12`	Main loop poll cadence
`NPA_EVALUATION_START_CUTOFF_RATIO`	`0.5`	Skip a round if starting past this fraction of it

Weights

Variable	Default	Meaning
`NPA_WEIGHT_EPOCH_BLOCKS`	`360`	Blocks between weight updates within a round

LLM proxy

Variable	Default	Meaning
`OPENROUTER_API_KEY`	—	OpenRouter key (proxy upstream)
`CHUTES_API_KEY`	—	Chutes key (proxy upstream)
`NPA_PROXY_MAX_TOTAL_SPEND_USD`	`0.01`	Spend cap per miner per task
`NPA_PROXY_UPSTREAM_TIMEOUT_SECONDS`	`60`	Upstream request timeout
`NPA_PROXY_PORT`	`8080`	Container-internal proxy port (never published to the host)

Auto-update (scripts)

Variable	Default	Meaning
`NPA_BENCH_REF`	`main`	npabench git ref to vendor
`NPA_UPDATE_INTERVAL_SECONDS`	`600`	Auto-update check interval
`NPA_UPDATE_EARLY_WINDOW_BLOCKS`	`50`	Only restart within this window before a round
`NPA_VALIDATOR_PM2_NAME`	`validator`	pm2 process name to restart

Setting the mission

The active mission is chosen by the backend, not the validator. It is published in each round's frozen roster as mission_id, and the validator reads roster["mission_id"]. Your local NPA_MISSION_ID is only used as a fallback before a roster exists.

Mission changes apply next round

Because the roster is frozen at evaluation start, changing the backend mission only affects the next round that freezes after the change. A round already under evaluation keeps the mission it froze with. On a private test stack you can delete and regenerate the roster to force the change immediately.