The First AI-Adaptive
Reward Protocol

NEBA implements a three-layer verification system for AI-powered reward distribution:

1. 1

   Data Aggregation Layer

   The protocol monitors Solana blockchain activity via RPC nodes, collecting participant behavioral metrics across 30-60 minute epochs. Wallet addresses are SHA-256 hashed before processing. Transaction patterns, holding duration, and interaction frequency are vectorized into anonymized statistical features.

2. 2

   AI Computation Layer

   Vectorized data is submitted to the LLM provider API (currently Claude AI via Anthropic's API, with fallback to GPT-4 or self-hosted Llama 3). The model analyzes contribution quality across multiple dimensions and generates reward multipliers ranging from 0.8x to 2.5x baseline rates. Computation requests include cryptographic commitments to source data.

3. 3

   Decentralized Verification Layer

   AI outputs undergo two-tier validation:

   • •Tier 1 - Optimistic Rollup: Stake-bonded proposers submit results on-chain with a 10-minute challenge period. Verifiers can submit fraud proofs, slashing proposer stake if results differ.
   • •Tier 2 - Oracle Consensus: Minimum 5-of-7 independent oracle nodes must agree on AI outputs. Outlier detection removes anomalous responses. Median aggregation ensures Byzantine fault tolerance.

Multiple safeguards prevent AI-driven economic exploits:

• Hard-coded emission caps:

  Smart contracts enforce monthly token limits calculated as (Remaining Allocation / Remaining Months). AI recommendations exceeding this cap are automatically rejected and scaled down.

• Statistical anomaly detection:

  Distribution automatically pauses if recommendations exceed 2.5 standard deviations from rolling 30-day average. Requires governance approval to proceed.

• Rate limiting:

  Epoch-to-epoch variance capped at ±20%. Individual multiplier changes limited to ±0.3x between consecutive epochs.

• Multi-signature emergency controls:

  Critical parameter modifications require 5-of-7 governance multisig with 7-day timelock. Emergency pause requires 3-of-7 without delay.