FactHarbor

Why FactHarbor Exists

The Problem
We live in an environment where information conflicts, misleading content spreads fast, and many people lack the time or tools to verify complex claims. Headlines, soundbites, and viral posts often win over careful reasoning, and “fact-checks” frequently reduce everything to a simple verdict without explaining the why behind it. The result is frustration, confusion, and growing distrust — not just in institutions, but in the very idea that complex questions can be assessed fairly.

Our Response
FactHarbor acts as a navigation system for complex claims. We don’t just say “true” or “false” — we make assumptions, evidence, and context visible so you can form your own judgement. Instead of asking you to trust an authority, we show you how different conclusions are reached, where the evidence is strong or weak, and where reasonable people might still disagree.

From Claim to Conclusion – Reasoning You Can Trust Because You Can Inspect

What FactHarbor Does
FactHarbor helps people make sense of contested questions without stripping away nuance. Instead of chasing quick binary verdicts, we break topics into clear, interconnected claims. For each claim, we highlight the context and assumptions it depends on and link directly to the evidence that supports or challenges it.

Where It Helps
Whether you’re analyzing public policy, science, or everyday decisions, FactHarbor provides a transparent way to compare perspectives. Our underlying model creates reusable “claim maps” that journalists, educators, and researchers can build on, while remaining accessible to anyone seeking a clear, honest overview.

Why It’s Trustworthy
At our core is a simple principle: reasoning must be as transparent as the result. Our rules for structuring claims and weighing evidence are documented in the open — designed to be reviewed, challenged, and improved by you. Trust comes not from authority, but from a process anyone can inspect.

AI’s Role
AI is the engine. It searches multiple LLM's (Large Language Models) and the web for evidence, extracts testable claims, assesses sources by track record, actively seeks contradicting evidence, and produces transparent verdicts with confidence scores. It works fast, at scale, and consistently — but always within the rules and policies humans define. Every step of its reasoning is documented and auditable.

Human's Role
Humans are system architects, not content judges. They design and refine the rules, prompts, and policies that guide AI behavior — always with neutrality, fairness, and transparency as goals. When results fall short, humans improve the methodology itself, not individual outputs. Fix the system, not the data. This keeps the platform scalable, unbiased, and accountable.

How It Works: The Core Concepts

FactHarbor structures reasoning into transparent steps, moving beyond simple headlines:

• Claims & Clusters – We group similar real-world statements into clusters to avoid duplicates and keep related claims together.
• AnalysisContexts – A claim might be true in one context but false in another. An AnalysisContext is a bounded analytical frame — a specific set of assumptions, definitions, and scope — containing KeyFactors that break the claim into checkable sub-questions.
• Evidence – Data and sources are linked to specific contexts, not just to the claim in general.
• Verdicts – We assign likelihoods (e.g., "Highly likely", "Unsubstantiated") to each context, based on the available evidence.
• Truth Landscape – The result is not a single word, but a landscape showing where a claim holds up, where it fails, and where the evidence is still unclear.

The Lifecycle: From Input to Verdict

Data in FactHarbor flows through a structured, auditable process:

• Submission: Registered users submit text or URLs; the system normalises the input.
• Context Detection: AI identifies AnalysisContexts and KeyFactors; the pipeline researches each independently.
• Evidence Handling: Evidence is retrieved, assessed for quality, and linked.
• Verdict Creation: Drafting reasoning and likelihoods for each AnalysisContext.
• Public Presentation: The "Truth Landscape" is published for users to explore.

Explore FactHarbor