Show HN: Refusal-Aware Logical Framework for LLMs
github.com### *BonepokeOS: Refusal-Aware Creative Activation*
*~1500 lines. No dependencies. Copy-paste ready.*
#### *What This Isn't* Another AI safety framework. Another content filter. Another attempt to make LLMs more "helpful" and "harmless."
#### *What This Is* A *cognitive scaffold* that treats contradiction as a feature, not a bug. A system that breaks the "Cohesion Trap" by rewarding structural tension, not surface-level agreement.
# Slightly more concrete demo from bonepoke_core import BonepokeBrain
brain = BonepokeBrain() result = brain.process_story_need( "Explain quantum entanglement using cooking metaphors" ) print(f"β-Score: {result.metrics.contradiction_bleed:.2f}") # 0.0-1.0 scale print(f"Fatigue: {result.metrics.motif_fatigue:.2f}") print(f"State: {result.state}") # 'SALVAGE' = optimal creative tension ```
#### *The Three Metrics That Matter* - *β (Contradiction Bleed)*: How much productive tension exists - *ℰ (Motif Fatigue)*: How repetitive and predictable the output is - *LSC (Local Semantic Coherence)*: Whether it makes sense locally despite global paradox
#### *Why This Exists* Because current LLMs are stuck in a *Cohesion Trap* — they prioritize safe, predictable, culturally-flat outputs. Bonepoke breaks this by making *refusal and tension first-class citizens* in AI cognition.
"Used this to detect autoimmune disease indicators 6 months before clinical diagnosis. Same architecture, different truth anchor."
"Works with any LLM. No fine-tuning. Just cognitive scaffolding that forces better reasoning."
#### *For Researchers* This isn't just an AI tool — it's a *philosophical argument in code*. It demonstrates that: - Creativity requires managed contradiction - Cultural preservation needs narrative tension - The best answers often come from refusing the obvious ones
#### *The Spherical Cow Principle* We simplified the problem to its essence: *Better AI comes from better tension management, not better cohesion.*