AI that understands time.

Cascade trains, tests, serves and verifies time-series foundation models, in the open.

why time series

Time series is in everything.

It is how science measures change, how engineering keeps systems running, and how organisations decide what happens next.

01

Understand

Science explains change.
Climate temperature, CO₂
Neuroscience EEG, neural signals
02

Operate

Engineering keeps it running.
Energy grid load, solar
Machines vibration, wear
03

Decide

Organisations plan ahead.
Retail demand, inventory
Finance prices, risk, cash

understand→operate→decide· every step needs a forecast

why it matters

All of it depends on knowing what comes next.

A language model can explain this chart. A time-series foundation model can forecast it, zero-shot, on data it has never seen. Cascade trains them, stress-tests them and makes them safe to act on.

General-purpose AI was never trained on time series. In an open competition, independent teams train time-series foundation models and are ranked on accuracy against real data.

open the dashboard

Most benchmarks reuse data models have already seen, so scores flatter them. Weir, powered by TSBench-Forge, calibrates and stress-tests every model on fresh series from after its training cutoff, twice a day.

view the live leaderboard

No single model wins on every series. Ephemeris sends each request to the model that performs best on that kind of data, served on demand with nothing to host.

sign up for ephemeris

Anyone acting on a forecast needs to know where it came from and how strong the evidence is. Gnomon runs the numbers in code, records every input and refuses when the evidence is too weak.

gnomon on github introduction (pdf)

Learn the patterns.Forecast the future.Decide with confidence.