MH-FLOCKE MH-FLOCKE
HomeDocsGitHubBlogPaperYouTubeReddit𝕏

Consistency & Integrity

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.