LAPLACE — the archive of consequence

THE BUTTERFLY MACHINE

Two worlds, identical except for one nudge smaller than a dust mote. Watch how long knowing almost everything is worth.

WORLD A — what you measured

WORLD B — what is actually there

System Your measurement error: 1 part in 10⁹

0.0s

elapsed

LOCKSTEP

the two worlds

—

prediction horizon

The cruel arithmetic: divergence grows exponentially, so measuring a thousand times better buys only a couple of seconds more foresight. That — not missing data — is why the demon starves. Perfect prediction would need infinite precision: the universe does not hand out infinite decimal places.

THE HORIZON TABLE

"Foretold, with enough data points" — here is how far enough actually gets you, domain by domain, and why the wall stands where it does.

System	Honest horizon	Why the wall is there
Solar eclipses	millennia	Two-body geometry is integrable — no sensitive dependence. The one place Laplace actually wins.
Ocean tides	centuries	Driven and dissipative: the moon re-disciplines the ocean every cycle, erasing small errors instead of amplifying them.
Planetary orbits	~50–100 million years	Weakly chaotic (Laskar, 1989): Lyapunov time ~5 Myr. Laplace's own solar system eventually betrays him.
Weather	~10 days, forever	Lorenz (1963): error doubles every ~2 days no matter how dense the sensors. Not an engineering gap — a property of the equations.
A double pendulum	~10 seconds	Two hinges are already enough for full chaos. You just watched it.
Dice, mid-tumble	under a second	Each bounce multiplies angular error; a millionth of a degree at release is a different face by the third bounce.
Markets	effectively zero	Reflexive: a credible forecast changes the behavior it forecast. The prediction eats itself.
History's forks	unknowable	A driver's wrong turn in Sarajevo, 1914. The cascade is only visible backwards — which is why the archive next door catalogs time travel instead.