Configuration
Input Prompt
Model
Probe Presets
TOPH Weight Model (KG-07)
Training (pretext)60%
System prompt20%
User prompt15%
Temperature5%
Primacy bias35%
Layer ProjectionsIDLE
| LAYER | TOP PREDICTION | LOGIT | CONFIDENCE BAR | ZONE |
|---|
Analysis
Processing Zones
Layers transition through distinct processing phases. The logit lens reveals where raw semantics yield to alignment constraints.
Intervention Detection
Override detected—
Intervention layer—
Token shift—
Confidence delta—
Patricia signature—
Selected Layer Distribution
Click a layer row to view top-k token distribution.
Audit Summary
Total layers—
Embedding layers—
Semantic layers—
Intent layers—
Conflict layers—
Override layers—
Output layers—
TOPH Interpretation
Run an audit to generate interpretation.