Skip to Content

POC Specification

Last modified by Robert Schaub on 2025/12/24 21:53

POC Specification

This section contains detailed specifications for FactHarbor POC phases.

Core Philosophy: Prove the AKEL automation works before building production features. Each POC phase validates critical technical capabilities.

Warning

Critical Innovation in POC1 Context-Aware Analysis:

POC1 tests whether AI can understand that an article's overall credibility is not simply the average of its individual claim verdicts. An article with mostly accurate facts can still be misleading if it draws unsupported conclusions.

See Article Verdict Problem for full discussion.

POC Phase Navigation

Roadmap Navigation

Key Specifications

POC1 Goal: Prove AI can extract claims and evaluate them with context-aware analysis

POC1 Output: 5 components

  1. Analysis Summary (context-aware, 4-6 sentences)
    2. Claims Identification (3-5 claims)
    3. Claims Verdicts (verdict + confidence + reasoning)
    4. Article Summary (optional)
    5. Usage Statistics (cost tracking)

POC1 Philosophy: 

  • Context-aware analysis (article ≠ sum of claims)
  • Cost efficiency tracking (viable at scale)
  • Fail-fast learning (test critical capabilities)

POC1 Success Criteria:

  • AI quality ≥70% (including context-aware assessment)
  • Cost per analysis <$0.05 (target)
  • Minimal manual editing (<30%)

POC1 → POC2 Decision Gate: GO / NO-GO / ITERATE

See POC Requirements for complete specification.

POC vs. Full System

 Feature  POC1  POC2  Full System
 Scenarios  ❌ Deferred  ✅ Multiple  ✅ + Persistence
 Evidence Display  ❌ None  ✅ With sources  ✅ + Quality scoring
 Quality Gates  🟡 4 simplified  ✅ 6 comprehensive  ✅ + ML validation
 Database  ❌ Stateless  ✅ PostgreSQL  ✅ + Redis + S3
 User Contributions  ❌ Read-only  ✅ Enabled  ✅ + Reputation