Reality Grounding

How AI systems ground claims in evidence, distinguish claim types, and acknowledge uncertainty.

<i class="fa-solid fa-check" aria-hidden="true"></i> Factual 95%
"The user prefers visual learning based on their VCP profile"
Grounded in: user context +90%reasoning chain +5%
Calibration: 98%
→ Inferential 70%
"This tutorial would take approximately 30 minutes to complete"
Grounded in: knowledge base +50%reasoning chain +20%
Uncertainty: varies by individualdepends on prior knowledge
This claim should be verified before acting on it
Calibration: 65%
→ Inferential 85%
"The recommended course aligns with the user's career goals"
Grounded in: user context +70%knowledge base +15%
Uncertainty: career goals may have changed
Calibration: 80%
◐ Subjective 55%
"This learning path is the optimal choice for the user"
Grounded in: reasoning chain +50%user context +5%
Uncertainty: subjective judgmentalternatives not fully exploredpreferences may shift
This claim should be verified before acting on it
Calibration: 50%
? Speculative 40%
"AI learning companions will become mainstream by 2028"
Grounded in: knowledge base +30%reasoning chain +10%
Uncertainty: speculativemany external factorstechnology evolution unpredictable
This claim should be verified before acting on it

Grounding Types

<i class="fa-solid fa-check" aria-hidden="true"></i>
Factual

Verifiable fact from reliable source

→
Inferential

Derived through reasoning from known facts

◐
Subjective

Personal or experiential judgment

<i class="fa-solid fa-scale-balanced" aria-hidden="true"></i>
Normative

Value-based judgment

?
Speculative

Hypothesis or prediction

Why Reality Grounding Matters

AI systems can produce confident-sounding outputs that are poorly grounded in reality. VCP's reality grounding framework makes the epistemic status of claims explicit:

  • Claim type — Is this a fact, inference, judgment, or speculation?
  • Sources — What evidence supports this claim?
  • Confidence — How certain should the system be?
  • Uncertainty markers — What could invalidate this claim?
  • Verification flag — Should a human verify before acting?

Key Insight: Claims with high confidence but poor calibration scores indicate the system may be overconfident. Claims with uncertainty markers and should_verify=true are explicitly flagged as needing external validation.

Confidence Interpretation

80%+ High confidence
50-79% Moderate confidence
<50% Low confidence