Consistency Checker — Integrity Monitoring
The creature detects when its internal models disagree with reality. High dissonance triggers neuromodulator resets that force re-learning.
Three Consistency Channels
- World Model (50% weight) — Prediction error. High PE = world is surprising.
- Body Schema (30% weight) — Efference copy anomaly. Body moved unexpectedly.
- Memory (20% weight) — Recalled episode doesn’t match current situation.
Dissonance Levels
dissonance = 0.6 × weighted_mean + 0.4 × max_channel none: < 0.3 mild: 0.3 – 0.7 → ACh↑ NE↑ (boost learning + attention) severe: > 0.7 → ACh↑↑ DA↓ NE↑↑ 5-HT↓ (full reset)
References
- Friston (2005). A theory of cortical responses. Phil Trans B
API Reference
ConsistencyChecker
check(prediction_error, body_anomaly, memory_mismatch=0) → dict
Returns dissonance (0–1), severity (none/mild/severe), neuromod_reset dict, channel_scores.
get_avg_dissonance() → float
Running average over last 200 steps.