Posterior Network — A Minimal Bayesian Reputation Simulator

This project implements the core mechanism behind reputation systems based on Bayesian updating.
It models a world with two types of sellers, two types of signals, and a buyer who updates their beliefs after each interaction.

Overview

The system follows a simple structure:

• Seller Types

  • Loyal: Usually provides high-quality outcomes.
  • Strategic: Acts based on long-term payoff but may provide lower quality in the short run.

• Signals

  • Good: Positive outcome delivered by the seller.
  • Bad: Negative outcome delivered by the seller.

• World Model The world model stores likelihood values:

P(signal | seller_type)

These likelihoods represent how each seller type tends to behave.

• Belief Updating The buyer maintains a prior belief:

P(Loyal), P(Strategic)

After observing a signal, the buyer updates these beliefs using Bayes’ rule:

P(type | signal) ∝ P(signal | type) * P(type)

Files

• computation.go — Contains the Bayesian model, likelihoods, and belief update logic.
• main.go — Demonstrates how to create a model and update beliefs after a signal.

Running the Example

1. Build the project:

go build

2. Run the executable: ./posterior-network You should see output similar to: map[0:0.6923076923 1:0.3076923077]