Building the World Modeling Engine: Probabilistic Forecasting for Personal Knowledge
How we built a causal simulation engine that predicts future knowledge states based on your capture patterns — and why temporal reasoning required an entirely separate service.
Why a World Modeling Engine?
Knowledge management is retrospective. You capture, you organize, you search — but you never predict. The World Modeling Engine (WME) changes that by building a probabilistic simulation of your future knowledge state based on your capture patterns.
The Architecture
The WME is a separate service from Serpaix's core API — and intentionally so. Temporal reasoning requires:
How It Works
The engine runs three parallel simulations:
Capture velocity: How many notes are you creating per day? Per topic? How does that change over weekends vs. workdays?
Graph growth rate: How fast is your knowledge graph expanding? Are certain domains growing faster than others?
Retention decay: Based on your FSRS review patterns, what knowledge is at risk of being forgotten?
Each simulation produces a probability distribution over future states. Combined, they let the system predict — with measurable confidence — what you'll know (and what you'll forget) next month.
Why This Matters
A knowledge system that predicts your future knowledge state can:
The WME is currently in Phase 17 development and is the foundation for Serpaix's proactive intelligence features.