Core concepts

A Dataset is a validated numeric time axis plus named numeric columns. Time values must be finite and strictly increasing. A World contains validated identifiers, dimensional units, variables, scalar parameters, and one continuous or discrete transition law per state. An expression is a finite scalar tree of constants, symbols, and the supported arithmetic and unary operations.

Discovery profiles input, estimates derivatives, builds deterministic feature columns, fits sparse coefficients, scores candidates, and creates a World. Simulation evaluates that World with RK4 for continuous laws or discrete stepping for discrete laws. A .lsworld bundle is the canonical serialized artifact with integrity checks.

The current bundle format intentionally excludes stochastic laws, delays, regimes, custom operators, signatures, and causal metadata. Those inputs are rejected rather than silently simplified.