ZeruAI · Client Work & Research

Client Deployments,
Field Research & Published Research

From live airdrop optimisations to peer-reviewed neural network models — everything ZeruAI has built, measured, and published.

Client Work

Live deployments where protocols ran zScore in production and measured the outcomes on-chain.

Live ClientV1 vs V2 DeltaOn-Chain Verified

GAIB

V1 → V2 Airdrop Optimisation

GAIB ran V1 without behavioral filtering. Sybils captured 21% of tokens and dumped within hours. For V2, they applied zScore before distribution. These are the measured on-chain results.

−56%
Sybil allocation cut
+49%
Quality wallet gain
−50%
Sell pressure
Read Full Report

Research

Airdrop Research

Retrospective simulations showing what zScore would have prevented in past airdrop distributions — backed by real on-chain wallet data.

Airdrop AnalysisSybil Detection

EigenLayer

Season 2 Airdrop Analysis

Deep analysis of 11,176 wallets across the EigenLayer S2 airdrop. 63% dump rate, with 68.4% of dumpers selling within 24 hours. 81.1% of recipients scored below 200. A zScore filter would have prevented 76% of token leakage.

11,176
Wallets analysed
63%
Dump rate
76%
Leakage preventable
Read Full Report
Airdrop AnalysisToken Retention

ether.fi

Season 3 Airdrop Analysis

ether.fi Season 3 distributed ETHFI to 10,015 wallets. 25.5% sold within days — 72.9% of those were serial repeat offenders who had dumped previous airdrops too. A zScore filter could have prevented 47% of token leakage, saving 686K ETHFI.

10,015
Wallets analysed
25.5%
Dump rate
686K
ETHFI saveable
Read Full Report
Airdrop AnalysisRepeat Offenders

Optimism

Seasons 4 & 5 Airdrop Analysis

11,024 wallets, 6.63M OP tokens. 22% immediate exits within 24 hours. 74.3% of wallets scored below 100. A zScore ≥ 200 filter simulation shows 73% reduction in same-day leakage.

11,024
Wallets analysed
22%
Same-day exits
73%
Leakage preventable
Read Full Report
Airdrop AnalysisExit Velocity

Elixir

ELX Airdrop Analysis

15,622 wallets, 40.9M ELX tokens. 35% bad actors, 85.1% of dumpers exited same-day. Extreme exit velocity — tokens to CEX within hours of claiming across every score band.

15,622
Wallets analysed
35%
Bad actor rate
5.0M
ELX saveable
Read Full Report
Protocol Analysis

Deep-dive analysis of protocol-level data — LP behaviour, incentive efficiency, and TVL concentration across DeFi ecosystems.

LP AnalysisTVL ConcentrationIncentive Design

Aave V3

Liquidity Provider Analysis

132,064 LP wallets on Ethereum mainnet. 4.9% of wallets hold 93% of all TVL. Capital efficiency gap of 241,129× between top and bottom LP cohorts — the case for score-weighted incentives.

132,064
LP wallets
93%
TVL in top 4.9%
241K×
Efficiency gap
Read Full Report
Model Research Papers

Peer-reviewed papers on the AI and machine learning models that power zScore — from the foundational reputation framework to deep behavioral scoring for DeFi.

Peer-Reviewed PaperFoundational ResearchNeural NetworkszkTLSReputation

zScore: A Universal Decentralised Reputation System

arXiv · 2503.05718

The foundational paper introducing zScore — a core primitive for on-chain reputation derived from wallet behaviour using state-of-the-art AI neural network models and real-world credentials ported on-chain through zkTLS. Validated on retroactive data from lending protocols, showing strong correlation between zScore and healthy borrowing and repayment behaviour.

AI + zkTLS
Methodology
Lending
Validated on
Mar 2025
Published
Read Paper
Peer-Reviewed PaperDeep LearningU-Net ArchitectureLP ScoringUniswap v3

Deep Reputation Scoring in DeFi

arXiv · 2507.20494 · Uniswap v3 Scoring

A behavioral scoring framework for Uniswap assigning two complementary scores: a Liquidity Provision Score and a Swap Behavior Score. Built on a deep residual neural network with densely connected skip blocks inspired by U-Net, incorporating pool-level context (TVL, fee tiers, pool size) to differentiate user behaviors across varying pool characteristics.

Deep ResNet
Model
Uniswap v3
Protocol
Jul 2025
Published
Read Paper

Want Your Protocol Analysed?

All reports are built from public on-chain data using the zScore API. If you're running an airdrop, liquidity programme, or incentive round — we can produce a tailored pre-distribution analysis or a live integration in under a day.

BOOK A DEMOEXPLORE API