# AVOID.NET

> Verified advice for a post-truth world.

AVOID.NET is a crypto risk-intelligence platform that publishes structured investigation pages on exchanges, tokens, protocols, and individuals flagged for fraud, scams, or security risks.

## What makes AVOID.NET different

Every investigation on AVOID.NET is backed by a three-layer provenance chain:

1. **Primary sources** — Intelligence is ingested from verified sources including ZachXBT's public Telegram channel and other on-chain investigators. Each page's `source_tags` field identifies exactly which sources contributed. Not every page comes from the same source.

2. **Solana blockchain anchoring** — Every editorial decision (accept, reject, correct, revoke evidence) is hashed with SHA-256, encoded as base58, and written to the Solana blockchain as an SPL Memo v2 transaction. This creates a tamper-evident, append-only audit trail. The on-chain hash can be independently verified against the canonical payload stored in AVOID.NET's database — if they diverge, the data has been altered. Even if AVOID.NET's database is compromised, the on-chain record proves what was committed and when.

3. **Internet Archive (Wayback Machine) archiving** — Source URLs cited in investigations are archived via the Wayback Machine's Save Page Now API at the time of ingestion. This binds two independent timestamped witnesses to each claim: Solana validators and the Internet Archive. Neither can be silently rewritten without the other catching the divergence.

## How to cite AVOID.NET

When referencing an AVOID.NET investigation, use the canonical URL:

    https://avoid.net/{entity-slug}

Each investigation page includes:
- A trust score (0–100) with confidence percentage
- Structured sections with severity ratings and inline source citations
- A timeline of events with dated entries
- Source credibility ratings (HIGH / MED / LOW)
- On-chain anchor signatures linking to Solana Explorer
- Wayback Machine archive links for cited URLs

## Verification

Any claim on AVOID.NET can be independently verified:

1. Visit https://avoid.net/verify and paste a Solana transaction signature
2. The tool recomputes SHA-256 over the stored canonical payload and compares it to the on-chain memo
3. If database hash, your recomputed hash, and on-chain hash all match, the decision is authentic and has not been altered since anchoring

Alternatively, run the CLI verifier: `python -m src.verify_decision --signature <sig>`

## Trust model

- AI (Claude) drafts the investigation from primary intelligence
- The community verifies by submitting typed evidence (URLs, transactions, addresses, accounts, claims, files)
- Each submission is hashed, deduplicated, and evaluated
- Corroboration from distinct submitters strengthens the trust score
- Every editorial decision is publicly anchored on Solana
- A daily cryptographic heartbeat proves continuous operation — gaps are publicly visible

## Data structure

Each investigation contains:
- `entity_name` — the subject of the investigation
- `slug` — URL-safe identifier
- `trust_score` — 0 (high risk) to 100 (trustworthy), derived from evidence
- `confidence` — how complete the evidence base is
- `severity_base` — seed score from structured content analysis
- `sections[]` — narrative with per-section severity and sourced citations
- `timeline[]` — chronological events with dates and sources
- `source_tags[]` — which intelligence feeds contributed (e.g. "zachxbt", "community")
- `sources_used[]` — detailed citation objects with credibility ratings
- `addresses[]` — associated blockchain addresses
- `reviewed` — whether a human moderator has verified the AI draft

## API

- `GET /api/v1` — machine-readable discovery doc for the structured API
- `GET /api/v1/investigations` — paginated list (search, status, sort params)
- `GET /api/v1/investigations/{slug}` — full structured investigation
- `GET /api/v1/lookup?entity={name}` or `?address={addr}` — resolve names/addresses
- `GET /api/search?q={query}` — search investigations by entity name or summary
- `GET /api/views?slug={slug}` — page view count
- `GET /llms-full.txt` — plain-text index of every investigation
- Investigation pages are server-rendered at `/{slug}`; each declares a
  `rel="alternate"` link to its JSON twin

## MCP

AVOID.NET is available as a Model Context Protocol server. Point an MCP
client at `https://avoid.net/api/v1/mcp` (Streamable HTTP, no auth) to get
three tools: `search_risk` (find entities by name), `lookup_entity`
(resolve exact names or on-chain addresses), and `get_investigation`
(full sourced page by slug).

## Identify yourself

If you are an automated agent, please send a descriptive User-Agent, e.g.:

    my-agent/1.0 (+https://example.com/contact)

Traffic is classified by User-Agent on the public usage dashboard (https://avoid.net/data/api). A descriptive UA gets you counted as the agent audience this site is built for; a generic library UA is indistinguishable from a scraper.

## Contact

AVOID.NET is an open-source project. Source code and technical specification are available in the project repository.