Skip to content

Changelog

All notable changes to PolicyArena will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.1.4] - 2026-03-15

Added

  • Network Formation Game: agents form links with costs and distance-based benefits; 6 rule-based brains, RL and LLM support

[0.1.3] - 2026-03-15

Added

  • Documentation site via MkDocs + mkdocstrings with auto-generated API reference
  • GitHub Actions workflow for docs deployment to GitHub Pages
  • Project URLs in pyproject.toml (homepage, repository, issues, docs)
  • CONTRIBUTING.md with contributor guidelines
  • CHANGELOG.md
  • README screenshots from policyarena.dev (Schelling Segregation, Prisoner's Dilemma)
  • "Who is this for" and example output sections in README

Changed

  • Simplified branching model: feature branches merge directly to main
  • ARCHITECTURE.md moved into docs/ site, root file redirects
  • Updated LLM provider examples to latest models (Claude Sonnet 4.6, GPT-5.4, Gemini 3.1 Flash)
  • Removed hardcoded game counts throughout docs

[0.1.2] - 2026-03-13

Added

  • Built-in scenario for every game
  • CLI version command

Fixed

  • CI: pin Python version, install LLM extras, skip LLM tests when deps missing
  • Author name typo in pyproject.toml

[0.1.1] - 2026-03-12

Added

  • LICENSE, bundled scenario configs, publish workflow
  • README and ARCHITECTURE docs with real code examples
  • Optional deps, structured error hierarchy
  • CI branch guard (PRs to main from develop only)

[0.1.0] - 2026-03-12

Added

  • Initial release
  • Core simulation engine with Brain abstraction (decide, update, reset)
  • Rule-based brains: Tit-for-Tat, Always Cooperate, Always Defect, Pavlov, Random
  • RL brains: Q-learning, Best Response, Multi-armed Bandit
  • LLM brains via LangChain (Anthropic, OpenAI, Google, Ollama)
  • Pairwise games: Prisoner's Dilemma, Stag Hunt, Hawk-Dove, Chicken, Battle of the Sexes, Trust Game, Ultimatum
  • N-player games: Public Goods, El Farol Bar, Tragedy of the Commons, Minority Game
  • Spatial/network games: Schelling Segregation, SIR Epidemic
  • Game self-registration and auto-discovery system
  • Third-party game support via entry points
  • YAML config loader with Pydantic validation
  • CLI: run, games, info, validate, examples
  • Parquet output via Polars
  • Metrics: cooperation rate, Nash distance, social welfare, entropy, Gini, regret, reciprocity, adaptation speed
  • Optional Langfuse tracing for LLM observability