// public sandbox · real alchemy ingestion · model v0.5.0-gov-expanded · status
v2.0 // public beta // MIT

HUNT SYBILS.
PROTECT REAL USERS.

Open-source Sybil detection for token distributions, DAO voting, and incentive programs. Score wallets, detect coordinated clusters, and give users a clear appeal path.

6 DETECTION METHODS ·5 CHAINS — ETH · ARBITRUM · OP · BASE · POLYGON ·DROP / REVIEW / KEEP VERDICTS ·EVIDENCE-BACKED · PER ADDRESS ·MODEL v0.5.0-GOV-EXPANDED ·APPEND-ONLY AUDIT LOG ·PUBLIC APPEALS · 48H SLA ·MIT LICENSED · OPEN SOURCE ·6 DETECTION METHODS ·5 CHAINS — ETH · ARBITRUM · OP · BASE · POLYGON ·DROP / REVIEW / KEEP VERDICTS ·EVIDENCE-BACKED · PER ADDRESS ·MODEL v0.5.0-GOV-EXPANDED ·APPEND-ONLY AUDIT LOG ·PUBLIC APPEALS · 48H SLA ·MIT LICENSED · OPEN SOURCE ·
// problem.txt

THE FARMER ECONOMY

Arbitrum
148,595

Sybils through Nansen → 22% supply stolen

zkSync
$753K

single farmer · 85 wallets · public extract

Linea
517K

filter flags · thousands of false-positive complaints

LayerZero
803K

filtered · zero public evidence per address

Aptos
40%

of distribution → exchanges within days

Industry
88%

of airdropped tokens dump < 3 months

// figures from public reporting + dune queries · not sybilshield-measured

// detection_stack

// 6 SIGNALS

01

FUNDING_SOURCE

Wallets funded by same source within tight time-window. Confidence scales with proximity.

02

BEHAVIORAL_HDBSCAN

Density clustering over 10-dim feature vectors. Scripts surface as tight blobs.

03

GRAPH_LEIDEN

Community detection on the tx graph. Isolated dense subgraphs = farms.

04

TEMPORAL_ENTROPY

Hour/day entropy + autocorrelation. Humans are messy; bots are not.

05

CROSS_CHAIN_LINK

Bridge events deterministically link wallet IDs across 8 chains.

06

ML_ENSEMBLE

LightGBM weighted by label-tier confidence. Honest holdout-only metrics.

// trust.md

// WHY TEAMS CAN TRUST THE SCORE

// evidence

Every score links to per-address evidence — which methods fired, which features pushed the number.

// open source

All six detection methods are MIT-licensed and documented. No vendor lock-in.

// sandbox vs prod

Every page distinguishes sandbox state from production state. No theatre.

// appeal flow

Public appeals endpoint with 48-hour SLA, built into the protocol, not a UX afterthought.

// audit log

Append-only event log. Any flag, appeal, or reversal is reproducible from the public methodology.

read the full methodology →

// terminal.out

// EVIDENCE, NOT JUST SCORES

sybilshield_inspect — 0xA12b…c4d7
● analyzed
target
0xA12b…c4d7
sybil_score
87
CLASSIFIED · sybil
CLUSTER: F-a8b2c1d4 (47 nodes)
[shared_funding]

Funded by 0xABC…123 → also funded 46 others within 1.0h

95%
[behavioral_clone]

Pattern matches cluster B-17 with 94% similarity

89%
[low_entropy]

Hour-of-day entropy 0.82 << human baseline (>2.5)

70%
// landscape.csv

// VS THE WORLD

TrustaNansenDIY DuneSybilShield
api_firsttruefalsefalsetrue
price/analysis$1-3K$50-150Kfree+4wfree
evidencePDFper-addr
audit_logimmutable
public_appeallive
open_sourceMIT
turnaroundhours4-8w2-4whours
// access.json

// FREE & OPEN

// a public good, not a product

Free for everyone. MIT-licensed. Grant-funded.

No prices, no plans, no checkout. Register with an email and score address cohorts with the full evidence pipeline. Fair-use limits keep the shared sandbox healthy, not a paywall.

  • - 100 calls/mo (reads free)
  • - Up to 1,000 addresses/analysis
  • - Public methodology + appeals
// audience.json

// BUILT FOR

Airdrop teams

Filter coordinated wallets before token distribution.

DAO operators

Review voter sets before controversial proposals.

DeFi incentive programs

Detect farming rings before rewards are paid.

Grant committees

Check whether applicants are connected entities.

READY

$> LAUNCH

Create a free sandbox account, run a demo analysis, and inspect evidence-backed scores.

./START