CrossSourceFactVerifier

A decentralized fact-verification primitive for the GenLayer blockchain. It cross-references multiple independent web sources to determine whether a given statement is TRUE, FALSE, or UNVERIFIABLE -- powered by AI consensus among decentralized validator nodes.

Why CrossSourceFactVerifier?

Traditional fact-checking is centralized: a single entity decides what is true. CrossSourceFactVerifier distributes this process across a network of independent AI validators, each fetching sources and reasoning independently. The result is a trustless, censorship-resistant verification record stored on-chain.

Key differentiator: Requires a minimum of 2 independent source URLs, enforcing genuine cross-verification rather than single-source reliance.

---

Public API

Write Methods

verify_fact(statement: str, source_urls: str) -> None

Verify a factual statement against multiple sources.

Parameter	Type	Description
statement	str	The claim to verify (must not be empty)
source_urls	str	Comma-separated URLs (min 2, max 10, each must start with http:// or https://)

Raises:

• UserError("statement must not be empty") -- empty or whitespace-only statement
• UserError("no valid source URLs provided") -- no URLs start with http/https
• UserError("at least 2 source URLs required for cross-verification") -- only 1 valid URL

View Methods

get_verification(record_id: str) -> str

Returns a single verification record as JSON.

{
  "id": "0",
  "statement": "Tesla delivered 500000 vehicles in Q3 2025",
  "sources": "https://reuters.com/article1,https://bbc.com/article2",
  "verdict": "TRUE",
  "agreement_ratio": 85,
  "conflicting_sources": "None",
  "verifier": "0xabc123..."
}

Raises: UserError("record not found") if record_id does not exist.

get_all_verifications() -> str

Returns all verification records as a JSON array.

get_stats() -> str

Returns aggregate statistics:

{
  "total_verified": 10,
  "true_count": 6,
  "false_count": 2,
  "unverifiable_count": 2
}

---

How Consensus Works

CrossSourceFactVerifier uses gl.vm.run_nondet_unsafe(leader_fn, validator_fn) -- the most powerful consensus pattern in GenLayer, where you write your own validator logic.

Leader Phase

1. The leader validator fetches content from each source URL using gl.nondet.web.render(url, mode="text").
2. Source content is truncated (3000 chars per source, 8000 total) and assembled into labeled sections.
3. The leader calls gl.nondet.exec_prompt() with a structured fact-checking prompt, requesting JSON output with verdict, agreement_ratio, and conflicting_sources.
4. Results are normalized (verdict mapped to TRUE/FALSE/UNVERIFIABLE, ratio clamped 0-100).

Validator Phase

Each validator independently:

1. Re-runs the entire leader logic -- fetching sources and querying their own LLM.
2. Compares semantics, not exact text:
   • Verdict must match (both say TRUE, or both say FALSE, etc.) after normalization
   • Agreement ratio must fall in the same band:
     • Band 0 (LOW): 0-35%
     • Band 1 (MID): 36-70%
     • Band 2 (HIGH): 71-100%
   • conflicting_sources text is not compared (different LLMs may list sources differently)

This band-based comparison allows different LLMs to reach the same semantic conclusion without requiring identical numeric outputs.

Why This Matters

• Not a format check: Validators verify the meaning of the result, not just JSON structure.
• Robust to LLM variance: Different models can word things differently but still agree on the core verdict and confidence level.
• Genuine cross-source analysis: Each validator independently fetches and analyzes the sources.

---

Use Cases

Use Case	Description
News Verification	Verify breaking news claims before sharing on decentralized social media
Research Claims	Check scientific claims against multiple journal/news sources
Due Diligence	Verify company information from independent sources before investing
Prediction Markets	Resolve event outcomes using multiple independent news sources
Content Platforms	Label user-submitted articles as "Verified" or "Disputed"
Insurance Claims	Cross-reference claim details against public records and news
Supply Chain	Verify product origin and certification claims from multiple registries

---

Usage Examples

Verify a News Claim

Method: verify_fact
Args:
  statement: "SpaceX successfully launched Starship on June 15 2025"
  source_urls: "https://reuters.com/space/spacex-starship,https://bbc.com/news/science-spacex,https://cnbc.com/spacex-launch"

Check the Result

Method: get_verification
Args:
  record_id: "0"

View Statistics

Method: get_stats
Args: (none)

---

State Design

VerificationRecord (on-chain struct):
  - statement: str
  - sources: str (comma-separated URLs)
  - verdict: str (TRUE | FALSE | UNVERIFIABLE)
  - agreement_ratio: bigint (0-100)
  - conflicting_sources: str (URLs or "None")
  - verifier: str (caller address as hex)

Contract state:
  - records: TreeMap[str, VerificationRecord]
  - next_id: bigint
  - total_verified: bigint
  - true_count: bigint
  - false_count: bigint
  - unverifiable_count: bigint

---

Edge Case Handling

Scenario	Behavior
Empty statement	UserError("statement must not be empty")
No valid URLs	UserError("no valid source URLs provided")
Only 1 valid URL	UserError("at least 2 source URLs required...")
All sources unreachable	Verdict = UNVERIFIABLE, ratio = 0
Some sources fail	Analyzes reachable sources only
Only 1 source reachable	Analyzes it, notes "Insufficient reachable sources"
LLM returns bad JSON	Fallback to UNVERIFIABLE, ratio = 0
Ratio out of range	Clamped to 0-100
More than 10 URLs	Only first 10 are used
Same statement verified twice	Allowed (information may change over time)

---

Live Deployment on Studionet

The contract is actively deployed on the GenLayer test network (studionet):

• Contract Address: 0x4afC08645cC7cd2074A8774e03C2fD162a5ee43f
• Network: studionet

Worked Example

Below is a realistic, worked example demonstrating how the primitive processes a claim.

1. Execute verification (Write Method)

Call verify_fact with a statement and at least two source URLs:

• Statement: "GenLayer is a blockchain Layer-1 that integrates AI at the consensus level."
• Source URLs: "https://genlayer.com,https://github.com/genlayer"

2. Query the result (View Method)

Call get_verification("0"). It returns the following verification record JSON:

{
  "id": "0",
  "statement": "GenLayer is a blockchain Layer-1 that integrates AI at the consensus level.",
  "sources": "https://genlayer.com,https://github.com/genlayer",
  "verdict": "TRUE",
  "agreement_ratio": 100,
  "conflicting_sources": "None",
  "verifier": "0x58f3a56e32a9637e8bcd44ccaeae794508b9b25e"
}

---

Deploy on GenLayer Studio

1. Open GenLayer Studio.
2. Click "New Contract" or use the file explorer.
3. Paste the contents of contracts/CrossSourceFactVerifier.py into the editor.
4. Select network Studionet.
5. Click "Deploy".
6. Wait for deployment confirmation -- you should see the contract address.
7. Test via the Studio UI:
   • Call verify_fact with a statement and comma-separated URLs.
   • Call get_verification with record_id = "0" to see the result.
   • Call get_stats to see aggregate counts.

---

Running Tests Locally

Prerequisites

• Python 3.11+
• genlayer-test package (gltest)

Install

pip install genlayer-test

Run

cd cross-source-fact-verifier
pytest tests/test_cross_source_fact_verifier.py -v

Test Coverage

• Happy paths: TRUE, FALSE, UNVERIFIABLE verdicts
• Edge cases: Empty statement, no URLs, single URL, nonexistent record
• Stats tracking: Counters increment correctly across multiple verifications
• Multiple verifications: Re-verifying updates stats properly

---

Submit to GenLayer Builder Portal

1. Go to the GenLayer Builder Portal.
2. Navigate to the Builder track, Intelligent Contracts category.
3. Fill in the submission form:
   • Title: CrossSourceFactVerifier
   • Description: A decentralized fact-verification primitive that cross-references multiple independent web sources using AI consensus to determine statement truthfulness.
   • Repository: Link to your repo or upload the project files.
   • Contract file: contracts/CrossSourceFactVerifier.py
   • Test file: tests/test_cross_source_fact_verifier.py
4. Submit and wait for review.

---

Project Structure

cross-source-fact-verifier/
  contracts/
    CrossSourceFactVerifier.py   # The intelligent contract
  tests/
    test_cross_source_fact_verifier.py  # Test suite with mocks
  README.md                      # This file

---

License

MIT