Skip to content

URL Intelligence Agent v1.3.0 — Semantic Conflict Intelligence

Latest
Latest

Choose a tag to compare

View all tags
@vpicciuolo vpicciuolo released this 08 Sep 15:07
· 10 commits to main since this release
v1.3.0
98f785a

🧠 URL Intelligence Agent v1.3.0

Semantic Conflict Intelligence

v1.3.0 significantly improves how URL Intelligence Agent understands contradictory evidence.

Until now, different wording across Meta tags, Open Graph, Twitter Cards, structured data and page content could sometimes be interpreted too aggressively as a conflict.

The new semantic conflict engine distinguishes normal wording differences from genuine factual disagreement and logical contradiction.

🔥 What's new

🧠 Semantic Conflict Classification

The provenance resolver can now classify relationships including:

  • semantic_equivalent
  • wording_variation
  • compatible_range
  • precision_difference
  • numeric_drift
  • temporal_drift
  • factual_disagreement
  • logical_contradiction
  • value_conflict

This provides substantially more precise evidence analysis than simple string comparison.

✍️ Wording Variation Detection

Different descriptions no longer automatically become hard conflicts simply because the wording is different.

For example:

Meta description

Open-source URL intelligence agent for web research and verification.

Open Graph

Evidence-first URL intelligence for research, provenance and web analysis.

These may now be classified as compatible wording variation rather than contradictory evidence.

🔎 Semantic Equivalence

The engine now normalizes and compares meaningful terms instead of relying only on exact string equality.

It evaluates:

  • Unicode normalization
  • punctuation
  • case
  • meaningful token overlap
  • semantic-token containment
  • factual anchors
  • numbers
  • currencies
  • URLs
  • explicit negation

⚠️ Factual Disagreement

When two claims contain materially different factual anchors, the engine can identify them as a real factual disagreement.

Examples include:

  • different prices
  • incompatible numeric claims
  • different URLs
  • identity differences
  • incompatible structured facts

These remain visible with their original provenance and evidence sources.

🚨 Logical Contradiction Detection

v1.3.0 introduces explicit contradiction detection for statements that share substantially the same subject but reverse their meaning.

Example:

Service is available worldwide.

vs.

Service is not available worldwide.

This is classified as a logical contradiction rather than ordinary wording variation.

Negation-aware analysis includes common patterns across multiple languages.

🔬 Evidence Inspector Upgrade

The hosted Evidence Inspector now makes these relationships easier to understand.

Users can inspect:

  • competing values
  • source layer
  • source property
  • representation
  • normalized value
  • confidence
  • observation timestamp
  • selected resolver value
  • conflict severity
  • semantic relationship
  • resolution explanation

Compatible semantic variations are now visible instead of being hidden behind the final resolved value.

🌐 Improved Conflict Explorer

Repeated conflicts across multiple crawled pages are grouped into reusable conflict patterns.

Instead of displaying:

value_conflict, value_conflict, value_conflict

the interface shows the actual competing evidence and explains why each relationship was classified.

⚡ Improved Web Experience

The live web application also includes an improved analysis workflow.

After clicking Run Analysis:

  1. an analysis overlay opens immediately
  2. investigation stages are displayed
  3. elapsed analysis time is shown
  4. provenance and external evidence are processed
  5. when analysis finishes, the user is automatically taken directly to the result

No fake progress percentage is shown.

🧪 Regression Testing

New regression tests cover:

  • Meta / Open Graph / Twitter description variation
  • semantic wording differences
  • explicit negation
  • logical contradiction
  • strict factual disagreement
  • existing provenance behavior
  • numeric compatibility
  • representation drift
  • claim verification
  • SSRF protections

Current semantic/provenance suite:

17 / 17 tests passing

CI successfully validated the release on:

  • Node.js 18
  • Node.js 20
  • Node.js 22

🔐 Existing Security Boundaries Preserved

The semantic intelligence upgrade does not weaken existing security protections.

The project continues to include:

  • SSRF protections
  • private/reserved address blocking
  • DNS rebinding protections
  • redirect validation
  • evidence isolation
  • restricted hosted browser boundaries
  • Remote MCP abuse protections

🤗 Live Demo

The v1.3.0 runtime is deployed on Hugging Face:

https://huggingface.co/spaces/vpicciuolo/url-intelligence-agent

🔌 Remote MCP

https://vpicciuolo-url-intelligence-agent.hf.space/mcp

📊 Benchmark Dataset

https://huggingface.co/datasets/vpicciuolo/url-intelligence-benchmark

💻 Source Code

https://github.com/vpicciuolo/url-intelligence-agent

Created by Vincenzo Picciuolo
HRN Innovation Technologies Ltd

Developed inside the HORNO Network ecosystem.

MIT Licensed.

Assets 2
Loading