Can the computer find a planet you hide? Hide one in a star's light and find out, then see where searches break.
A passing planet dims its star by a tiny amount, once every orbit. Search software hunts the star's brightness record for those repeating dips. Real stars also flicker on their own, which is exactly where searches go wrong.
START HERE
Three one-click experiments, about 5 seconds each
Or , measured by NASA's Kepler telescope.
OR BUILD YOUR OWN
Choose the star, the planet and how to clean the data.
1 · PICK A STAR
2 · HIDE A PLANET
How easy is it to spot?
How long is its year?
3 · CLEAN UP THE STAR'S FLICKER
Everything runs in your browser and takes a few seconds.
4 · DID IT FIND IT?
THE GRAPHS
The star's brightness over the whole observation. Orange: where your planet dims it, usually far too small to see by eye.
After cleaning: the star's own flicker removed, leaving (ideally) just noise and the planet's dips.
The search tries thousands of possible orbits. Tall peaks are candidate planets; above the red line counts as a detection. Orange dashes: your planet's real orbit.
All the data stacked on the orbit the search picked, so every transit lands on top of the others. A real planet shows a dip in the middle.
5 · WHAT THIS TELLS US
Advanced settings: exact numbers
The big experiments behind the Transit Lab: thousands of hidden planets per run, searched with the full SpaceSift pipeline. Each one records what it asked, how it was run and what it found, so anyone can rerun it from its code version, settings and random seed.
Choose an experiment.
What is this?
Astronomers find planets around other stars by watching for tiny, repeating dips in a star's brightness as a planet passes in front. Search programs hunt for those dips. The Transit Lab asks a different question from "can it find planets?": when, and why, does the search miss them?
To answer that, SpaceSift plants fake planets of known size and orbit in real and simulated light curves, runs the search, and counts how often it succeeds, like hiding eggs and grading the searcher.
What it has found so far
SS-0001. On 200 quiet Kepler stars, the standard search (box least squares) recovers half of planted planets at a signal-to-noise of 10.9 and 97% at high signal. The pipeline behaves as expected.
SS-0002A (simulation). Stellar variability can hide planets entirely. On a star pulsating every 4 hours at 100 ppm, the standard cleaning step found 0 of 106 planted planets; removing the pulsation's rhythm first (prewhitening) found 105. For fast-spinning spotted stars it is the reverse: prewhitening fails and only a short cleaning window helps. No single method works on every star.
SS-0002B (real stars, in progress). Whether the simulated map predicts what happens on 316 real variable Kepler stars. Predictions were committed before the run.
How the demo relates to the experiments
The experiments run the Python pipeline on many thousands of planted planets. This page runs a JavaScript version of the same steps in your browser, checked against the Python one; its detection threshold is calibrated separately (1% false alarms on noise-only stars), so its numbers can differ slightly from the experiments'. One run is one example, not a result.