ArticlesSignalConceptsEvidenceQuestionsThe Inference AuditorAbout

Evidence

Delta Analysis

Blind evaluation by GPT-5 across 30 analytical dimensions. Configured output, using inference-time cognitive configuration, against default output from the same model on the same prompt.

Overall Score

9.2vs7.8

Average across all dimensions

Semantic Density

0.106vs0.055

Concepts per word

Meta-Reasoning

+8point advantage

On meta-cognitive execution

Dimension-by-Dimension Comparison

Selected dimensions. Full 30-dimension matrix in the delta analysis.

DimensionConfiguredDefaultDelta
Analytical Depth9.57.2+2.3
Structural Coherence9.38.0+1.3
Novel Insight Generation9.46.8+2.6
Meta-Cognitive Awareness9.65.5+4.1
Constraint Satisfaction9.17.9+1.2
Semantic Density9.07.0+2.0
Epistemic Precision9.37.5+1.8
Framework Originality9.56.2+3.3
Counter-Argument Handling9.27.8+1.4
Practical Actionability8.88.1+0.7
Reasoning Transparency9.47.3+2.1
Information Prioritization9.66.9+2.7

Semantic Density Comparison

Configured Output0.106
Default Output0.055

Unit: concepts per word. Higher = more information per token.

See the full evidence

The raw responses and the blind judge analysis

The full response from a standard, fresh Gemini 3 Deep Think instance. The full response from Gemini 3 Deep Think running the complete NovaThink Cognitive Seed stack of eight meta-cognitive priors that govern global reasoning. And the full delta analysis from a GPT-5 instance acting as a blind judge with no knowledge of which output came from which configuration.

  1. 01

    Default Response

    A fresh, unconfigured Gemini 3 Deep Think instance answers the synthesis prompt.

  2. 02

    Configured Response

    Gemini 3 Deep Think with the full NovaThink Cognitive Seed stack answers the same prompt.

  3. 03

    Delta Analysis

    A blind GPT-5 instance compares both outputs across 30 analytical dimensions.

Want to see how your own AI outputs compare?

Try The Inference Auditor