The Neural Reality Gate (services/neural_reality_gate.py, ~820 lines) is Frank's grounding validator — every thought is checked against reality before it can influence behavior.
Architecture: 5 Hippocampal-Prefrontal Modules
| Module | Analog | Function |
|---|---|---|
| HC | Hippocampus | Memory consistency — does this match stored facts? |
| mPFC | Medial PFC | Self-model check — is this consistent with who Frank is? |
| Perirhinal | Perirhinal Cortex | Familiarity detection — has Frank seen this before? |
| Temporal | Temporal Cortex | Semantic plausibility — does this make sense? |
| aPFC | Anterior PFC | Meta-judgment — how confident is the overall assessment? |
21K Parameters
Multi-task supervised training from hallucination_log and stored reflections. Trains during consolidation with SFT anchoring (same system as Neural Conscience).
Integration
Called from _validate_thought_output() in the consciousness daemon. Extends regex-based violation detection. Can escalate to full suppression if confidence is high. Catches ~7.9% of autonomous thoughts as hallucinations (92.1% integrity rate).