Skip to content

validator.md

Latest commit

 

History

History
96 lines (70 loc) · 2.63 KB

validator.md

File metadata and controls

96 lines (70 loc) · 2.63 KB

Validator

What Validators Do

Validators observe the TensorUSD auction contract on-chain, track which miners win liquidation auctions, and use those results to assign weights to miners on the Bittensor network. Those weights determine each miner's share of TAO emissions each epoch.

How Validators Score Miners

Every AuctionFinalized event is recorded. At the end of each epoch, the validator scores every miner's wins using this formula:

bonus_ratio = min((winning_bid - debt_balance) / debt_balance, 0.20)

reward = 1.0 + bonus_ratio
Winning Bid vs Debt bonus_ratio reward
Exactly debt (bid = debt) 0.00 1.0
10% over debt 0.10 1.1
20%+ over debt (capped) 0.20 1.2
  • Minimum reward per win: 1.0
  • Maximum reward per win: 1.2
  • Multiple wins in the same epoch accumulate:
total_reward(miner) = sum of reward(win_i) for all wins in tempo window

Weight Setting

Scores are smoothed over time using an exponential moving average (EMA), then submitted to the chain as weights:

score = α × new_reward + (1 - α) × previous_score

weights = L1_normalise(scores)   →   submitted via subtensor.set_weights()

Each miner's TAO emission share is proportional to their normalised weight. Miners that haven't won any auction in the epoch get no weight increase — their score decays toward zero via the EMA.

How to Run

uv run neurons/validator.py \
  --netuid 421 \
  --subtensor.network test \
  --wallet.name <wallet_coldkey> \
  --wallet.hotkey <wallet_hotkey> \
  --logging.info
Flag Description
--netuid 421 Subnet UID to validate on
--subtensor.network test Connect to the testnet
--wallet.name test Name of the local wallet
--wallet.hotkey default Hotkey to use for this validator
--logging.info Enable info-level logs

Sequence Diagram

sequenceDiagram
    participant V as Validator
    participant C as Substrate Chain
    participant AC as Auction Contract
    participant DB as SQLite DB

    V->>C: connect
    V->>C: sync_historical_wins(last_step_block, current_block)
    C-->>V: past AuctionFinalized events
    V->>DB: store AuctionWin rows

    loop Background thread
        C-->>V: AuctionFinalized event (live)
        V->>AC: get_auction() — fetch debt_balance
        V->>DB: store AuctionWin (deduplicated)
    end

    loop Every epoch
        V->>DB: get_auction_rewards(tempo_start, tempo_end)
        DB-->>V: rewards[], uids[]
        V->>V: score = α × reward + (1-α) × score
        V->>C: set_weights(normalised scores)
    end
Loading