reality-engine

Reality Engine's central idea is that simulation should model when and why things happen, not where they sit. Its Temporal Graph Network reifies time as a traversable dimension: events are nodes, consequences are edges, and the graph automatically prunes low-probability branches and merges convergent timelines so compute stays tractable. That pruning-and-merging behavior is the part most homegrown alternate-history scripts lack, and it's what lets the same engine run a village over a decade on a laptop or a planetary civilization over ten thousand years on distributed clusters (the README describes that scaling, though distributed runs are described as theoretical rather than battle-tested). Strengths cluster around four capabilities you can actually invoke: alternate history modeling forked at a named historical event, probabilistic forecasting from macro-economic or climate inputs, generative narrative design that outputs a chronicle of alliances, betrayals, discoveries, and wars from agent interactions, and educational sandboxing that shows small initial changes diverging exponentially. The agent layer is more than decoration — agents carry unique drives, knowledge, and memory, form coalitions and governments through hierarchical cognition, and spread cultural memes via contagion models weighted by trust, authority, and recency bias. Resource and energy economics track depletion, efficiency gains, and collapse cycles. Multilingual output in 40+ languages with a Cultural Context Engine that adapts names, titles, units, and calendars (including right-to-left and CJK) means you don't get Anglo-centric chronicles by default. Weaknesses are equally concrete. There is no GUI — it is CLI-only — and the documentation is essentially the README, so onboarding is read-the-source work. The GitHub repository shows 128 stars and 782 commits with no releases, no formal support channels, and no commercial backing; if your team needs SLAs or professional services, this is the wrong shelf. There is no 3D spatial world-building or rendered environment output, and the 40+ language output is text and data, not polished presentation for non-technical stakeholders. Distribution to clusters, while described, is not yet proven at scale. Where it fits: researchers studying historical causality or chaos theory, writers and game masters who want a simulated chronicle rather than an outline, game developers prototyping emergent narrative, and educators who need a live demonstration that a treaty clause or a mistimed natural disaster reshapes centuries. Where it doesn't: anyone who wants drag-and-drop worldbuilding, a rendered map, or a vendor to call when a simulation breaks at 2 a.m.