01
Principles
- Primary-source first
- Evidence over narrative
- Versioned claims
- Explicit uncertainty
- Scientific fact separated from commercial inference
- Commercial inference separated from Qoriant judgment
- Corrections remain visible
- Sources remain traceable
02
Claim Types
- FACT CANDIDATE
- An AI-extracted statement tied to an exact source excerpt. It is not yet a FACT.
- FACT
- Directly supported by a cited source and approved by a human reviewer.
- INFERENCE
- A reasoned conclusion beyond what a source states.
- QORIANT VIEW
- Our explicit research or venture judgment.
- UNCERTAIN
- A material question with insufficient evidence.
03
Evidence Levels
- A — peer-reviewed primary evidence or an official result
- B — strong secondary evidence or cross-source synthesis
- C — company claim or preliminary result
- D — working hypothesis or unverified signal
04
AI and Human Review
AI may extract candidates, suggest relationships and prioritize signals. It cannot approve FACT, merge uncertain entities or publish a Qoriant object. Every decision preserves the proposed value, reviewed value, reviewer, time and decision.
05
Citation Integrity
Claims and Signals retain their Source ID, URL, retrieval date and supporting context. If an exact supporting excerpt cannot be verified, the candidate is marked unsupported and reduced to evidence level D.
06
Corrections
Material corrections update the object version, verification date and source registry. Qoriant does not silently convert uncertainty into fact.