Release title: v1.2.0 — Claim Provenance & Temporal Consistency Engine

URL Intelligence Agent v1.2.0

Claim Provenance & Temporal Consistency Engine

v1.2.0 is a major verification and web-intelligence upgrade for URL Intelligence Agent.

The agent now goes beyond extracting a single value from a page. It preserves field-level evidence, tracks exactly where each value came from, compares multiple page representations, detects conflicts and drift, reasons about approximate values and numeric ranges, and explains why one claim was selected over another.

🔎 Field-Level Provenance

Evidence can now include:

• source representation
• extraction layer
• property and locator
• raw value
• normalized value
• timestamps
• confidence signals
• evidence hashes

Supported evidence layers include:

• Source HTML
• Rendered DOM
• Open Graph
• Meta tags
• Twitter Cards
• JSON-LD
• Microdata
• RDFa
• Visible content
• HTTP headers
• Runtime API evidence where permitted

🧠 Claim Resolution

New claim states include:

• consensus
• compatible_variation
• drift
• conflict
• insufficient_evidence

The engine understands the difference between exact values, approximations, ranges and lower/upper bounds.

Example:

Open Graph: 80,000+
Rendered page: 100,502

Instead of incorrectly reporting a contradiction, v1.2 can determine:

Logical conflict No
Precision difference Yes
Representation drift Yes
Freshness divergence Yes
Stale metadata suspected Yes
Preferred value 100502

⏱ Temporal Consistency

Added support for:

• representation drift detection
• freshness divergence
• stale metadata detection
• temporal claim history
• provenance-aware snapshots
• ETag and Last-Modified evidence
• conditional HTTP requests

🔐 Evidence Integrity

Evidence can include SHA-256 integrity hashes so downstream systems can verify exactly which representation produced a result.

A W3C PROV-shaped export is also available.

🛠 New Actions

New capabilities include:

• inspect_provenance
• verify_claim

Existing investigation, SEO, structured-data, comparison, monitoring and knowledge-export actions now also benefit from provenance intelligence.

🔌 MCP Upgrade

v1.2.0 supports:

• MCP 2026-07-28
• MCP 2025-11-25
• MCP 2025-06-18
• MCP 2025-03-26

Also included:

• modern stateless MCP operation
• legacy session compatibility
• strict action-specific schemas
• structured output schemas
• MCP Tasks support
• stronger hosted anti-abuse protections

🌐 Evidence Inspector

The live Hugging Face application now includes an Evidence Inspector for exploring:

• observations
• resolved claims
• conflicts
• drift
• suspected stale metadata
• representations
• evidence layers
• confidence
• resolution explanations

📊 Benchmark

The v1.2.0 production release passed the current core benchmark:

• 55 / 55 benchmark cases
• 120 / 120 assertions
• 107 / 107 deterministic assertions
• 18 / 18 security checks
• 6 / 6 full-agent completion checks
• 100% current core benchmark score

A dedicated provenance regression track has also been added.

🛡 Security

This release also improves:

• SSRF protection boundaries
• browser request restrictions
• private-address checks
• download restrictions
• Remote MCP rate-limit handling
• charset-aware decoding
• evidence isolation

Unrestricted browser rendering remains intentionally disabled on the public Hugging Face deployment because browser networking represents a separate SSRF and egress trust boundary.

🚀 Live

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

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

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

Created by Vincenzo Picciuolo / HRN Innovation Technologies Ltd

Developed inside the HORNO Network ecosystem.

Licensed under the MIT License.