FactHarbor architecture

FactHarbor researches claims and articles, assesses supporting and opposing evidence, and produces reports whose conclusions can be examined alongside their sources. It remains an invite-gated Alpha; broader development is paused pending funding.

Analysis method

Full-size diagram · Mermaid source

The method separates what the user asserts from what the sources establish. AtomicClaims identify the assertions to assess. EvidenceScopes record the conditions under which individual evidence applies. ClaimAssessmentBoundaries group compatible evidence after research, so differences in methodology, time or geography remain visible instead of being flattened into one answer.

The debate stage tests an assessment against evidence-backed challenges. The resulting report distinguishes truth assessment from confidence, cites evidence, and describes limitations. Neither agreement among models nor a confident narrative establishes correctness by itself.

Application components

Component	Responsibility
Next.js web application	Reader interface, administration and analysis execution
ASP.NET Core API	Jobs, persisted reports and progress events
SQLite storage	Local persistence for jobs and configuration
Configured model and search services	Language-model tasks and evidence retrieval

Current implementation, schemas, prompts and default settings are part of the public repository. Language models perform semantic judgments; structural code handles contracts, identifiers and resource control. Providers and active configuration affect execution and must be recorded when comparing results.

Read and contribute

• Analysis responsibilities and stage contracts.
• Quality and trust: evidence, uncertainty and report interpretation.
• Security and operations: access boundaries and operational limits.
• Getting started, API source and terminology.

Research proposals, historical designs and current behavior are different kinds of evidence. Consult current status before treating a capability as available.