LawSynthGitHub

What is LawSynth?

LawSynth discovers interpretable, executable law systems from time-series data. Point it at a CSV and it recovers the governing equations behind the numbers — a world you can read, simulate, analyze, control, and share. If you can't read and reason about the result, it isn't a LawSynth result.

A damped phase-portrait spiral collapses to the discovered law dx/dt = y, dy/dt = −x − 0.3·y
A damped phase-portrait spiral collapses to the discovered law dx/dt = y, dy/dt = −x − 0.3·y

What you get

  • Explicit laws, not black boxes. Discovery returns equations like dx/dt = σ (y − x) with fitted parameters — not opaque weights.
  • Deterministic and offline. The same inputs produce a bit-identical world, on any machine, with no network access. Reproducibility is a contract, not a hope.
  • One portable artifact. Every discovery is a .lsworld bundle, and every downstream tool operates on that single file.

The core loop

The LawSynth discovery loop: observe a CSV, discover the laws, understand them, use them to forecast and control, and share the result
The LawSynth discovery loop: observe a CSV, discover the laws, understand them, use them to forecast and control, and share the result
observe (CSV)  →  discover (laws)  →  understand (explain)  →  use (forecast / analyze / control)  →  share (report / export / .lsworld)

Each step is a real command in the lawsynth CLI, mirrored by the Python and TypeScript SDKs. Discovery is a sparse fit to your data — a compact hypothesis, not proof of causality — and every world records the uncertainty and assumptions behind it.

Local-first and reproducible: the same inputs produce the same world, offline, forever.

Where to go next