SN83 - centralization & miner incentive

bit_contraction · November 21, 2025, 2:16pm

Could the burn police have a look at SN83? Rules broken :

Rule 4: Random graphs are generated from a single API by the devs. No explanation why this is needed since validators could generate their own locally. Apart from a clear single point of failure, it also allows potential cheating by injecting the optimal solution.

CliqueAI/graph/client.py

main


      
              number_of_nodes_max: int | None = None,
              number_of_edges_min: int | None = None,
              number_of_edges_max: int | None = None,
          ) -> LambdaGraph:
              """
              This function is called by the validator to get a graph.
          
              Args:
                  wallet (bt.Wallet): Validator wallet.
              """
              url = f"{LAMBDA_URL}/graph/lambda"
              payload = GraphPayload(
                  timestamp=time.time(),
                  hotkey=wallet.hotkey.ss58_address,
                  uuid=str(uuid.uuid4()),
                  netuid=netuid,
                  label=label,
                  number_of_nodes_min=number_of_nodes_min,
                  number_of_nodes_max=number_of_nodes_max,
                  number_of_edges_min=number_of_edges_min,
                  number_of_edges_max=number_of_edges_max,

Rule 6: Type of problems received (difficulty and amount) is proportional to alpha staked to owner’s validator.

github.com/toptensor/CliqueAI

CliqueAI/selection/miner_selector.py

main


      
              if S == 0:
                  x_m = np.array([1] * len(self.miner_uids))
              else:
                  x_m = np.sqrt(1 + s_m / S)
          
              delta = x_m - difficulty - 0.5
              P = 1 - np.exp(-np.maximum(0, delta))
              self.miner_weights_cache[difficulty] = P
              return P
          
          def sample_miner_uids(self, difficulty: float) -> list[int]:
              """
              Sample miners based on their stake weights and a given difficulty.
          
              Args:
                  difficulty (float): The difficulty threshold for sampling miners.
          
              Returns:
                  list[int]: A list of selected miner UIDs.
              """
              P = self.miner_weights(difficulty)

github.com/toptensor/CliqueAI

CliqueAI/selection/miner_selector.py

main


      
          
                  if (
                      self.current_block - self.snapshot.metagraph.last_update[uid]
                      <= self.snapshot.epoch_length
                  ):
                      continue
          
                  uids.append(uid)
              return uids
          
          def miner_weights(self, difficulty: float) -> np.ndarray:
              """
              Calculate the weights of miners based on their stake and the difficulty.
          
              Args:
                  difficulty (float): The difficulty threshold for sampling miners.
          
              Returns:
                  np.ndarray: An array of weights for each miner.
              """
              if difficulty in self.miner_weights_cache:

Apart from this, there are other issues:

No comms by team for > 1 month
The incentive mechanism heavily promotes one big miner entity. This is more subtle and debatable but clearly described by several people in the SN discord.

Topic		Replies	Views
SN-15 is using a centralized scoring mechanism and probably gaming burn police investigation	2	82	November 18, 2025
SN27 - setting all miner emissions to a "treasury" key burn police investigation	1	34	May 8, 2026
SN-88 requires miners to stake alpha in order to mine burn police investigation	0	24	November 19, 2025
SN-87 was asking for external tao payment like 102 before they sold it burn police outdated	1	28	November 16, 2025
Chain Decentralization NPoS (Incentives) Roadmap on-mainnet	0	21	December 29, 2025

SN83 - centralization & miner incentive

Related topics