Specification

This section defines the technical architecture, data models, and functional requirements of FactHarbor.

Core Concepts

FactHarbor structures reasoning about claims into transparent, inspectable steps:

• Claims and Claim Clusters – people submit real-world statements; similar phrasings are grouped.
• Scenarios – each claim is evaluated under clearly defined contexts (assumptions, definitions, boundaries).
• Evidence – sources and data are collected and linked to specific scenarios.
• Verdicts – each scenario receives a likelihood-based assessment, not an absolute label.
• Truth Landscape – the set of scenario-dependent verdicts shows where and why a claim seems more or less plausible.
• Time Evolution – all important entities (claims, scenarios, evidence links, verdicts) are versioned, so it is possible to see *what we believed at time X* and how this changed.

Functional Lifecycle

The system follows a six-step lifecycle:

1. Claim submission: Automatic extraction and normalisation; Cluster detection.
   2. Scenario building: Clarifying definitions and assumptions; AI proposals with human approval.
   3. Evidence handling: AI-assisted retrieval; Human assessment of reliability; Explicit scenario linking.
   4. Verdict creation: AI-generated draft verdicts; Human refinement; Reasoning explanations.
   5. Public presentation: Concise summaries; Truth Landscape comparison; Deep dives.
   6. Time evolution: Versioning of all entities; Re-evaluation triggers when evidence changes.

Chapters