How it works
Critical information about scams, frauds, and dangerous products gets buried, glossed over, or actively suppressed by the people with financial interests in hiding it.
AVOID.NET is a verified-claims index for the post-truth era. AI drafts the case file. The community verifies it. Solana anchors the audit log. Every editorial decision is public and permanent.
The system
┌──────────────────────────────────────────────────────────┐
│ SOURCES │
│ @zachxbt (Telegram) · community submissions · GoPlus│
└────────────┬─────────────────────────┬───────────────────┘
│ │
▼ ▼
┌──────────────────────┐ ┌──────────────────────┐
│ Ingestion agents │ │ Web app │
│ Python · src/ │ │ Next.js · web/ │
│ │ │ │
│ connect → archive │ │ pages · forms │
│ → draft → anchor │ │ · auth · admin │
└──────────┬───────────┘ └──────────┬───────────┘
│ │
└─────────────┬─────────────┘
▼
┌──────────────────────────────┐
│ Supabase (Postgres + Auth) │
│ investigations · evidence │
│ · decisions · versions │
└──────────────┬───────────────┘
▼
┌──────────────────────────────┐
│ Solana (SPL Memo v2) │
│ publish + decision anchors │
│ · daily heartbeat │
└──────────────────────────────┘Three layers, single responsibilities: ingestion pulls from external sources and runs the Claude pipeline; the web app handles public pages and moderation; Supabase holds the queryable state and Solana holds the tamper-evident audit log.
The pipeline
Agents monitor sources like ZachXBT's Telegram channel, pulling scam warnings and threat intelligence in real time. URLs are archived via the Wayback Machine before they can disappear.
Claude analyzes the raw intelligence — extracting entities, addresses, timelines, and severity — then drafts a structured investigation page with sourced claims and a trust score.
Anyone can submit evidence: URLs, on-chain transactions, addresses, or firsthand claims. Each submission is hashed and deduplicated. Corroboration from distinct submitters strengthens the case.
Every editorial decision — accept, reject, correct, revoke — is hashed and written to Solana as an SPL Memo. The audit trail is append-only and publicly verifiable by anyone, without trusting avoid.net.
A cryptographic heartbeat is anchored on Solana every day, even when there are no editorial changes. A gap in heartbeats is a publicly visible signal that something stopped — we cannot silently turn it off.
Anatomy of an investigation page
Every entity gets the same dossier-style layout. The signals you see on a page are designed to be checkable, not taken on faith.
Trust model
The AI is the journalist. The community is the editor. Solana is the notary. Every page carries a status pill showing where it sits in this lifecycle.
Each entity page carries a trust score derived from sourced evidence, weighted by source credibility and corroboration. The confidence percentage reflects how complete the evidence base is — a high score with low confidence means more research is needed.
Every anchor signature on avoid.net can be independently verified on Solana Explorer. Our signer transparency page discloses the public keys we use, so you can check our on-chain history directly.
History
AVOID.NET launches as an open wiki for sharing warnings about products, companies, and practices worth avoiding. Without strong verification, it is eventually overwhelmed by spam and taken offline.
Relaunched with AI-drafted investigations, community verification, and on-chain audit anchoring on Solana. The spam problem is addressed by economic friction at the source layer and by making every editorial decision publicly verifiable, not by trusting moderators.
Participate
Do diligence better together. If you have evidence about an entity — a URL, a transaction hash, an address, a firsthand account — you can submit it directly on any investigation page.
Verified advice for a post-truth world.