MHA Simulator is a structured My Hero Academia roleplay system that generates scenario-based interactions using state inputs, character clusters, and faction relationships. It routes behavior through layered logic (state, weights, and interaction graphs) to produce consistent, hierarchy-aware MHA-themed scenes with reactive world dynamics.

Update: Identity anchoring through taxonomy tables.

---

Bots Speaking For Users:

This is an issue that occurs when a system (bot character & API) have no rules to remove the probability of occurrence from the language model. Bot rules determine structure. API rules determine stochasticity (probability).

Temperature: 0.8 (high temp - broader exploration)

Top K: 30 (high top k - broader language pool)

Token Gen: 1200 (long gen - increased branching output)

Even if you have rules that say no, language models may still have room to sample into user narration.

The Fix:

[remove user-narration from valid solution space]

Temperature: 0.7 (low temp - narrower exploration)

Top K: 20 (low top k - smaller language pool)

Scenario: "How character bot behaves."

LLM Custom Prompt: "How language model behaves."

Taxonomy in Personality: "Who these people are."

Interaction Graph Engine (IGE): "How these people act."

[Stochastic Narrowing + System Enforcement + Identity Anchoring]