Nobody gets the server’s answer#
The simulation knows what happened. The people inside it do not.
A character can believe something because they saw it, read it or heard it from somebody else. Those are not interchangeable sources. If you watched a man leave through the north gate, a stranger insisting that he went south should not overwrite your own eyes. If three people repeat the same accusation, you may take it more seriously—but repetition still does not turn hearsay into a fact.
This distinction is the base of Satisfy’s social mechanics. A fact belongs to the world. A belief belongs to a person. A rumour is a belief that arrived through somebody else’s telling, with a trail behind it: who said it, where it was heard and what claim was actually passed on.
That means two honest people can disagree. Neither needs access to a secret global answer, and neither has to be pretending for the sake of a scripted scene. They simply reached different conclusions from what they perceived.
The road changes the news#
I do not want rumours to teleport between NPCs. Someone has to speak, and someone else has to be close enough to hear. Taverns, markets, workshops and wells matter because people meet there. A traveller can carry news between places; an empty road cannot.
Trust changes what survives the journey. A report from a friend should carry more weight than the same words from a stranger. A hostile witness may still plant doubt without persuading anyone. Several independent tellers can strengthen a claim, while the same person repeating it five times remains one source wearing out everybody’s patience.
Official documents are a special kind of hearsay. A gazette or decree can preserve its wording across a long distance because each copy comes from the issuer rather than from a chain of retellings. It is still not automatically true. A character may trust the office, distrust it or believe their own eyes instead. The seal proves where the statement came from; seals have never shown much interest in proving reality.
Retelling should also blur a claim instead of replacing it with random nonsense. A woman sees a courier leave by the north gate. By evening, the market may have reduced this to “someone important fled town.” It has lost detail, but it has not somehow become “the mayor was turned into a goose” unless somebody deliberately introduced a new lie. The journey should be visible in the information itself.
Reputation lives in people#
Most games store reputation as a number attached to the player. Help the bakers, rob a baker’s brother, and the city quietly averages the matter into three and a half stars.
Satisfy is meant to keep the untidy version. Relationships are directed: I may trust you while you barely know me. They can include affection, fear, familiarity, debts and social roles such as kinship or employment. What people believe about your actions changes those relationships. Reputation is then a view across many minds, not a score owned by your character, so it can differ between a family, a guild and a district. It can also be wrong.
This matters for play. A person should offer work according to what they need and what they believe you have done. A witness may open a door for you. A lie may close another one later if its path leads back to you. The same rules apply to players and NPCs: both act through the world, both can become sources, and neither receives a private exemption from consequences.
The semantic layer keeps the claim separate from its phrasing. One NPC can tell it in Russian, another in English, and a third in the clipped language of somebody who has decided that details are a luxury. An LLM may help give each speaker a voice, but it does not decide what happened or what anyone believes. Those decisions stay in the simulation, where they can be checked and replayed.
The useful edge of uncertainty#
This is the social model I am building toward, not a claim that the whole system is already present in the playable prototype. The first useful version only needs to prove a small chain: somebody witnesses an event, tells another person, and the report reaches someone who acts differently because of it.
The difficult part is not making information spread. It is stopping the result from becoming either a perfect distributed database or a fog of random gossip. Memory has to forget without erasing every consequence. Distortion has to create uncertainty without destroying meaning. Trust must matter without becoming a number players grind by repeating the safest action.
Most importantly, the game must be able to answer “why?” Why does this person fear me? Who told the guard? Where did the accusation lose a name? If the path cannot be inspected, the world is not behaving—it is merely producing surprises.
When a rumour finally reaches the player, I want it to feel as though it has been somewhere.