Execution Phase (Off-Chain)

After a user submits an inference request, an executor node picks it up from the job queue and performs the AI computation off-chain. To ensure verifiability, the executor runs the model as a zk-circuit and produces a cryptographic proof of correctness.

This phase is computation-heavy and forms the core trust-minimized processing layer of the Hubic protocol.

🛠️ Execution Workflow:

  1. Executor receives the request tied to a model_hash.

  2. Loads the zk-circuit and model weights.

  3. Fetches the input payload using its input_hash (off-chain).

  4. Performs the AI inference using zk-compatible logic.

  5. Generates a zk-proof of the execution.

  6. Hashes the output and wraps the result into a verifiable package.

📦 Result Package Submitted to Verifier:

{
  "output": ["validator_07", "validator_21"],
  "zk_proof": "0xabc123...",
  "proof_hash": "0xdef456...",
  "output_hash": "0xghi789..."
}

  • zk_proof: The full cryptographic proof for the inference

  • output_hash: SHA-256 hash of the inference result

  • proof_hash: Hash of the zk-proof data structure

⚠️ Integrity Notes:

  • If the output doesn’t match the model’s registered circuit logic, the proof will fail on-chain.

  • Only registered models with valid verification keys can be executed.

  • No inference = no reward.

🌍 RWA Integration:

  • Yield Trigger Point: This is where tokenized AI models generate the raw material for RWA-based revenue.

  • Executor Accountability: Performance metrics (e.g., latency, proof size, error rate) impact future task allocation and reputation.

  • Pre-validation for Royalty Rights: Only output packages that pass structural pre-checks will proceed to revenue distribution on-chain.

The execution phase is where AI meets cryptographic enforcement — creating the only kind of work that can be trusted, tokenized, and monetized.

PreviousInference Request (User Job)NextVerification Phase

Last updated