Skip to main content

Compute Validation Overview

Introduction​

The Republic compute validation protocol establishes a robust and verifiable mechanism for confirming that computational tasks are performed honestly and efficiently by network validators. Unlike traditional blockchains, which primarily focus on transactional consensus, Republic places compute at the core of its consensus and economic incentives, making the quality and integrity of computation a first-class concern.

This protocol enables the network to:

  • Schedule computational jobs securely and transparently.
  • Validate the results of these jobs through cryptographic proofs and benchmarks.
  • Incentivize reliable and performant compute providers via economic rewards and reputation.
  • Protect the network against fraud and degraded performance using slashing and quality controls.

The following sections provide an in-depth examination of the components, flow, and cryptoeconomic mechanisms that comprise the compute validation protocol.

Job Scheduling and Execution​

At the heart of the protocol is the process by which computational tasks (jobs) are scheduled and executed by validators.

  • Job Definition: Each job specifies a computational workload, such as matrix multiplication benchmarks or neural network inference tasks.
  • Scheduling: Jobs are delegated to validators through a decentralized scheduler that considers stake, past performance, and compute quality.
  • Execution: Validators run the assigned computations within predefined time constraints and generate outputs alongside cryptographic proofs of execution.

This process ensures that computation is timely, traceable, and attributable to specific validators.

Cryptographic Proofs of Computation​

To guarantee the authenticity of computation, Republic employs cryptographic proofs tied to the execution of models.

  • HashedModel Framework: Validators run computations using a HashedModel that generates intermediate checkpoint hashes reflecting the model’s internal state during execution.
  • Root Hash Generation: These checkpoint hashes are aggregated into a root hash, uniquely identifying the entire computation.
  • Verification: Validators submit this root hash alongside outputs; verifiers re-execute the computation to reproduce checkpoints and confirm hash correctness.

This proof mechanism binds the computation’s correctness and integrity cryptographically, preventing forgery or shortcuts.

Benchmarking for Quality and Performance​

Republic leverages three complementary benchmarks to objectively measure compute quality:

  • Throughput Benchmark: Measures the validator's ability to perform large-scale matrix multiplications accurately and rapidly.
  • Inference Benchmark: Tests the validator’s execution of transformer-based sequence inference models to assess realistic model execution capacity.
  • Achieved FLOPs Benchmark: Calculates effective floating-point operations per second (FLOPS) by timing a fixed-complexity model execution, providing a real-world performance metric.

Validators’ benchmark results determine their reputation and influence their eligibility to participate in consensus committees.

Reputation and Economic Incentives​

The protocol links compute quality to economic incentives:

  • Reputation Scores: Computed from benchmark results and job performance, representing the validator’s reliability and efficiency.
  • Token Rewards: Validators earn native tokens (REP) proportionally to their reputation and successful job completion.
  • Slashing: Validators who submit incorrect or delayed computations risk token slashing, disincentivizing dishonest or poor-quality work.

This framework encourages validators to maintain high standards of compute integrity and throughput.

Validator Selection and Consensus Participation​

Validator reputation and compute benchmarks feed directly into the consensus mechanism:

  • Validators with higher reputation and benchmark scores have increased probability to be selected for consensus committees.
  • This ensures that blocks and state transitions are secured by participants demonstrating both economic stake and verifiable compute power.
  • The protocol thereby aligns network security with genuine compute contribution.

Fraud Prevention and Quality Assurance​

To safeguard the network, the protocol implements multiple layers of fraud detection:

  • Re-execution Verification: Validators’ submitted results and hashes are independently re-executed by verifiers to detect discrepancies.
  • Slashing Conditions: Failures in correctness, timing, or proof submission trigger automatic slashing penalties.
  • Ongoing Benchmark Validation: Validators undergo periodic re-benchmarking to prevent performance degradation over time.

Together, these measures preserve the network’s integrity and trustworthiness.