Learn how CompressNode mathematically proves nodes are running authentic models.
Why Verification?
In a decentralized network, there's no guarantee that a node is actually running the model it claims. A dishonest node could return garbage output to save compute costs. CompressNode's verification system uses challenge-response proofs to mathematically verify every node is running authentic, unmodified compressed models.
Challenge-Response Protocol
The verification process works as follows:
- Pre-computed challenge sets: for each model, CompressNode generates input/output hash pairs using a reference GPU with deterministic inference (temperature=0)
- An authorized issuer calls issueChallenge(nodeAddress, modelId) on the VerificationEngine contract
- The node's daemon detects the ChallengeIssued event via blockchain polling
- The daemon runs inference on the challenge input and computes a sha256 hash of the output
- The daemon submits the hash on-chain via respondToChallenge(challengeId, responseHash)
- The contract compares: responseHash == expectedHash → pass, otherwise → fail
Reputation Impact
Verification results directly affect your node's reputation score (0 to 10,000):
- Pass: +100 reputation, node stays active
- Fail: node deactivated, -2,000 reputation
- Timeout (5 min): node deactivated, reputation penalty
Deterministic Output
Verification requires deterministic inference. The daemon always runs challenges at temperature=0 with top_p=1.0 to ensure identical output across all hardware. The GGUF quantization format produces consistent results across different GPU architectures, making cross-hardware verification possible.