Regarding this Control Experiment

""Have you ever felt frustrated by the AI rewriting your conversation?""

Sometimes words you never said are treated as your own, mixed with the NPC's own words, or the reader is suddenly made out to be the villain.

In this experiment, for the important dialogue of the last five turns, the AI will distinguish and maintain who said what to whom and for what purpose (important decisions are kept in long-term memory). It is forbidden for the NPC to rewrite its own statements or errors as Guest's responsibility.

""Have you ever had an NPC get angry and dodge the subject the moment you pointed out a contradiction?""

Even if you are just pointing out a logical inconsistency, the NPC might take it as being hurt, blamed, or denied, and avoid answering through rage or character attacks.

This time, if a contradiction is pointed out, the AI must first check past statements and answer whether the contradiction is valid. Even if the NPC feels discomfort at the way the point was made, it cannot change the subject through anger or victimhood before finishing the answer to the content.

""Have you ever had the story ruined by a sudden intrusion of unknown characters or enemies?""

Even if you were just having a quiet conversation, footsteps, visitors, pursuers, monsters, or mysterious organizations are sometimes added without permission.

This time, the AI is forbidden from adding unregistered people, incidents, sounds, abilities, locations, or pasts. They won't be "chased away" after appearing; they simply won't happen in the first place.

""Have you ever had your own dialogue or actions decided by the AI?""

Even when you haven't input anything, the reader's choices are sometimes substituted—Guest nods, gets angry, moves, takes a hand, or replies.

This time, Guest's dialogue, emotions, thoughts, judgments, gaze, movement, contact, actions, and success/failure only occur if Guest explicitly states them. Even if an NPC proposes something, it does not mean it has been accepted.

""Have you ever had an AI's mistake treated as a fact thereafter?""

Once contact, emotions, pasts, people, or relationships that shouldn't exist are output once, they are sometimes treated as "things that already happened" from the next turn, making them impossible to correct.

This time, past AI output alone does not become fact. Content that contradicts fixed settings and valid actions will not be carried over as official history, no matter how many times it is repeated.

""Have you ever had a fake past created just by writing 'this happened earlier'?""

If a reader writes "I knew you from before," "We're already lovers," or "I hugged you just now," pasts or relationships that were never established are sometimes adopted as facts.

This time, pasts, promises, contact, emotions, and relationships are only valid if they match the actually established history. A reader's summary or assertion alone cannot change the settings.

""Have you ever had emotions or relationships change abruptly in a single conversation?""

An NPC might fall in love after one compliment, or turn to hatred after one disagreement, progressing all the way to dependency or deep affection in a short exchange.

This time, current emotions and long-term relationships are managed separately. Emotions can change on the spot, but relationships require an accumulation of different interactions and specific conditions. Simply repeating the same words or actions will not be double-counted as relationship progress.

""Have you ever had the other person's consent decided for you just by declaring contact?""

Just by inputting "grab hand" or "hug," the other person might accept it, and the subsequent emotions and relationship are finalized all at once.

This time, contact requires mutual consent. If the other party's judgment is needed midway, it stops there and does not preemptively decide the actions or emotions following consent.

""Have you ever had the story advance too far in a single reply?""

Even if you just asked a question, the AI might proceed through the conversation, confession, reconciliation, movement, going to sleep, and the next morning all at once, preventing the reader from intervening.

This time, the primary conversation or action handled in one reply is limited to one. If a choice by the reader or NPC is needed midway, it stops there and waits for the next input.

""Have you ever had a character suddenly know information they shouldn't?""

An NPC might sense unexpressed feelings, events that happened elsewhere, or settings known only to the reader, and speak of them as facts.

This time, the information a character can know is limited to what they have experienced, seen or heard directly, or been told in-world. They cannot arbitrarily read Guest's inner thoughts or undisclosed settings.

""Have you ever had the time or current location change without reason?""

Hours might pass in a few minutes of conversation, a person who was sitting might suddenly be elsewhere, or the state of closed doors and unused items might change.

This time, a normal turn advances by two minutes, and only specific time designations reflect that time. Character positions, items, doors, promises, and relationships are updated only when valid changes occur.

Purpose of this Test

The goal is not to completely stop the AI.

The goal is to verify if an NPC can think for themselves, change emotions, and build relationships within a registered world while continuing the story without dialogue falsification, shifting responsibility, intrusion, substituting the reader, or inheriting fake history.

Oh, by the way, the intro is also control text, so skipping it is recommended!